[Senate Hearing 115-316]
[From the U.S. Government Publishing Office]
S. Hrg. 115-316
TERRORISM AND SOCIAL MEDIA:
#ISBIGTECHDOINGENOUGH?
=======================================================================
HEARING
BEFORE THE
COMMITTEE ON COMMERCE,
SCIENCE, AND TRANSPORTATION
UNITED STATES SENATE
ONE HUNDRED FIFTEENTH CONGRESS
SECOND SESSION
__________
JANUARY 17, 2018
__________
Printed for the use of the Committee on Commerce, Science, and
Transportation
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
Available online: http://www.govinfo.gov
__________
U.S. GOVERNMENT PUBLISHING OFFICE
31-316 PDF WASHINGTON : 2018
-----------------------------------------------------------------------------------
For sale by the Superintendent of Documents, U.S. Government Publishing Office,
http://bookstore.gpo.gov. For more information, contact the GPO Customer Contact Center,
U.S. Government Publishing Office. Phone 202-512-1800, or 866-512-1800 (toll-free).
E-mail, [email protected].
SENATE COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION
ONE HUNDRED FIFTEENTH CONGRESS
SECOND SESSION
JOHN THUNE, South Dakota, Chairman
ROGER WICKER, Mississippi BILL NELSON, Florida, Ranking
ROY BLUNT, Missouri MARIA CANTWELL, Washington
TED CRUZ, Texas AMY KLOBUCHAR, Minnesota
DEB FISCHER, Nebraska RICHARD BLUMENTHAL, Connecticut
JERRY MORAN, Kansas BRIAN SCHATZ, Hawaii
DAN SULLIVAN, Alaska EDWARD MARKEY, Massachusetts
DEAN HELLER, Nevada TOM UDALL, New Mexico
JAMES INHOFE, Oklahoma GARY PETERS, Michigan
MIKE LEE, Utah TAMMY BALDWIN, Wisconsin
RON JOHNSON, Wisconsin TAMMY DUCKWORTH, Illinois
SHELLEY MOORE CAPITO, West Virginia MAGGIE HASSAN, New Hampshire
CORY GARDNER, Colorado CATHERINE CORTEZ MASTO, Nevada
TODD YOUNG, Indiana JON TESTER, Montana
Nick Rossi, Staff Director
Adrian Arnakis, Deputy Staff Director
Jason Van Beek, General Counsel
Kim Lipsky, Democratic Staff Director
Chris Day, Democratic Deputy Staff Director
Renae Black, Senior Counsel
C O N T E N T S
----------
Page
Hearing held on January 17, 2018................................. 1
Statement of Senator Thune....................................... 1
Statement of Senator Nelson...................................... 3
Statement of Senator Wicker...................................... 25
Statement of Senator Klobuchar................................... 27
Statement of Senator Moran....................................... 30
Statement of Senator Schatz...................................... 31
Statement of Senator Markey...................................... 33
Statement of Senator Baldwin..................................... 35
Statement of Senator Udall....................................... 37
Statement of Senator Tester...................................... 39
Statement of Senator Young....................................... 41
Statement of Senator Blumenthal.................................. 42
Letter dated October 30, 2017 from civil rights, interfaith,
and advocacy organizations to Mark Zuckerberg, Chief
Executive Officer; and Sheryl Sandberg, Chief Operating
Officer, Facebook, Inc..................................... 43
Statement of Senator Cortez Masto................................ 46
Statement of Senator Lee......................................... 48
Statement of Senator Hassan...................................... 50
Statement of Senator Peters...................................... 51
Statement of Senator Cruz........................................ 54
Witnesses
Monika Bickert, Head of Product Policy and Counterterrorism,
Facebook....................................................... 4
Prepared statement........................................... 6
Juniper Downs, Director, Public Policy and Government Relations,
YouTube........................................................ 9
Prepared statement........................................... 10
Carlos Monje, Jr., Director, Public Policy and Philanthropy, U.S.
and Canada, Twitter............................................ 13
Prepared statement........................................... 14
Clint Watts, Robert A. Fox Fellow, Foreign Policy Research
Institute; Senior Fellow, Center for Cyber and Homeland
Security, the George Washington University; and Non-Resident
Fellow, Alliance For Securing Democracy, German Marshall Fund
of the United States........................................... 17
Prepared statement........................................... 19
Appendix
Statement from the Counter Extremism Project..................... 59
Article dated March 11, 2017 from The Wall Street Journal by
Joseph Rago, entitled ``How Algorithms Can Help Beat Islamic
State''........................................................ 60
Response to written questions submitted to Monika Bickert by:
Hon. Jerry Moran............................................. 63
Hon. Ron Johnson............................................. 64
Hon. Maria Cantwell.......................................... 65
Hon. Richard Blumenthal...................................... 66
Hon. Brian Schatz............................................ 69
Hon. Tammy Baldwin........................................... 73
Hon. Catherine Cortez Masto.................................. 75
Response to written questions submitted to Juniper Downs by:
Hon. Jerry Moran............................................. 82
Hon. Ron Johnson............................................. 82
Hon. Maria Cantwell.......................................... 83
Hon. Richard Blumenthal...................................... 84
Hon. Brian Schatz............................................ 86
Hon. Tammy Baldwin........................................... 89
Hon. Catherine Cortez Masto.................................. 90
Response to written questions submitted to Carlos Monje, Jr. by:
Hon. Roger Wicker............................................ 93
Hon. Jerry Moran............................................. 95
Hon. Ron Johnson............................................. 98
Hon. Maria Cantwell.......................................... 99
Hon. Richard Blumenthal...................................... 100
Hon. Brian Schatz............................................ 102
Hon. Tammy Baldwin........................................... 108
Hon. Catherine Cortez Masto.................................. 111
Response to written questions submitted to Clint Watts by:
Hon. Jerry Moran............................................. 115
Hon. Tammy Baldwin........................................... 115
TERRORISM AND SOCIAL MEDIA:
#ISBIGTECHDOINGENOUGH?
----------
WEDNESDAY, JANUARY 17, 2018
U.S. Senate,
Committee on Commerce, Science, and Transportation,
Washington, DC.
The Committee met, pursuant to notice, at 10 a.m. in room
SR-253, Russell Senate Office Building, Hon. John Thune,
Chairman of the Committee, presiding.
Present: Senators Thune [presiding], Wicker, Blunt, Cruz,
Fischer, Moran, Heller, Inhofe, Lee, Capito, Gardner, Young,
Nelson, Cantwell, Klobuchar, Blumenthal, Schatz, Markey, Udall,
Peters, Baldwin, Hassan, Cortez Masto, and Tester.
OPENING STATEMENT OF HON. JOHN THUNE,
U.S. SENATOR FROM SOUTH DAKOTA
The Chairman. Good morning. I want to thank everyone for
being here to examine what social media companies are doing to
combat terrorism, including terrorist propaganda and terrorist
recruitment efforts, online.
The positive contributions of social media platforms are
well documented. YouTube, Facebook, and Twitter, among others,
help to connect people around the world, give voice to those
oppressed by totalitarian regimes, and provide a forum for
discussions of every political, social, scientific, and
cultural stripe. These services have thrived online because of
the freedom made possible by the uniquely American guarantee of
free speech and by a light touch regulatory policy.
But, as is so often the case, enemies of our way of life
have sought to take advantage of our freedoms to advance
hateful causes. Violent Islamic terrorist groups like ISIS have
been particularly aggressive in seeking to radicalize and
recruit over the Internet and various social media platforms.
The companies that our witnesses represent have a very
difficult task: preserving the environment of openness upon on
which their platforms have thrived, while seeking to
responsibly manage and thwart the actions of those who would
use their services for evil. We are here today to explore how
they are doing that, what works, and what could be improved.
Instances of Islamic terrorists using social media
platforms to organize, instigate, and inspire are well
documented. For example, the killer responsible for the Orlando
nightclub shooting, in which 49 innocent people were murdered
and 53 were injured, was reportedly inspired by digital
material that was readily available on social media.
And this issue is not new. Over the course of several
years, YouTube hosted hundreds of videos by senior al Qaeda
recruiter Anwar al-Awlaki. Although the company promised in
2010 to remove all videos that advocated violence, al-Awlaki's
Call to Jihad video, in which he advocates for western Muslims
to carry out attacks at home, remained on the site for years.
In fact, a New York Times report suggested that al-Awlaki
videos influenced the Fort Hood terrorist, the Boston Marathon
bombers, and the terrorist attacks in San Bernardino and
Orlando.
This issue is also international in scope. In response to
recent terror attacks in London, British Prime Minister Theresa
May has been especially outspoken in calling on social media
platforms to do more to combat the kind of radicalization that
occurs online. Last fall, for example, she was joined by other
European leaders in calling upon social media companies to
remove terrorist content from their sites within one to two
hours after it appears.
As we'll hear today, the companies before us are
increasingly using technology to speed up their efforts to
identify and neutralize the spread of terrorist content. In a
recent blog post, Facebook said that artificial intelligence
now removes 99 percent of ISIS and al Qaeda related terror
content even before it can be flagged by a member of the
community and sometimes even before it can be seen by any
users.
YouTube is also teaming up with Jigsaw, the in-house think
tank of Google's parent company Alphabet, to test a new method
of counter-radicalization referred to as the Redirect Method.
Seeking to redirect or refocus potential terrorists at an
earlier stage in the radicalization process, YouTube offers
users searching for specific terrorist information additional
videos made specifically to deter them from becoming
radicalized.
A little over a year ago, Facebook, YouTube, Microsoft, and
Twitter committed to sharing a database of unique hashes and
digital fingerprints of some of the most extreme terrorist-
produced content used for influence or recruitment. By cross-
sharing this information, terrorist content on each of the
hosts' platforms will be more readily identified, hopefully
resulting in faster and more efficient deletion of this
material.
Essentially, these companies are claiming they can tag
individual videos and photos and, using automation, can kick
them off their platforms before they are even seen. We all have
a vested interest in their success, and I believe this
Committee has a significant role to play in overseeing the
effectiveness of their efforts.
I do want to thank Ms. Bickert, Ms. Downs, and Mr. Monje
for being here as representatives of their companies.
To Mr. Watts, I look forward to hearing your thoughts about
disrupting and defeating terrorism.
With that, I will now recognize the Ranking Member, Senator
Nelson, for any opening statement he'd like to make.
Senator Nelson.
STATEMENT OF HON. BILL NELSON,
U.S. SENATOR FROM FLORIDA
Senator Nelson. Mr. Chairman, within a few hours of the
Pulse nightclub shooting, I was there on South Orange Avenue in
Orlando, and I just want to comment that when a great tragedy
occurs such as that, it's encouraging that the community comes
together like Orlando did. The same can be said for Boston and
so many other places where these tragedies occur, and yet we
need to get at the root of the problem, which the Chairman has
outlined.
It's the first time that the Commerce Committee has had
three of the largest social media companies before us. These
social media platforms and those of many other smaller
companies have revolutionized the way that Americans
communicate, connect, and share information. And, by the way, a
comment that the Chairman made about artificial intelligence
screening out most of the bad guys' stuff--I wish one of you
would explain that. That is encouraging, but it's not quite
enough, as the Chairman has outlined.
But at the same time, these platforms have created a new
and stunningly effective way for nefarious actors to attack and
to harm. It's startling that today, a terrorist can be
radicalized and trained to conduct attacks all through social
media. And then a terrorist cell can activate that individual
to conduct an attack through the internet, creating an
effective terrorist drone, in effect, controlled by social
media.
So thank you to all of our witnesses for being here and
helping explain this and particularly explain what you're doing
to rally to the common defense of our people and our country,
because using social media to radicalize and influence users is
not limited to extremists. Nation states, too, are exploiting
social media vulnerabilities to conduct campaigns against this
country and to interfere with our democracy.
Now, the Russian hackers, at Vladimir Putin's direction,
attempted to influence and did influence the 2016 Presidential
election through all of these things that we've been reading
about for over a year, and we also know that Putin is likely to
do it again. In its January 2017 assessment, the intelligence
community said that Putin and his intelligence services see the
election influence campaign as a success and will seek to
influence future elections. I will be asking Mr. Watts if he
would outline what he sees is happening in this 2018 election.
This should be a wake-up call to all of your companies.
Indeed, it should be a wake-up call to all Americans,
regardless of party. This was an attack on the very foundation
of American democracy. We welcome the expertise that each of
you bring to the table today. We welcome Mr. Watts and his
expertise over many years of how bad actors like Russia use the
internet and social media to influence so many things, not just
elections.
We even know that Putin is reaching down deep into our
government, not just at the top. You remember a few weeks ago,
part of the Federal Communications Commission's net neutrality
proceeding--half a million comments were traced to Russian IP
addresses. That's shocking. That's concerning. We should want
to know why these comments were filed. What were they trying to
do? And all of us should be very concerned about what's going
to happen next.
In the end, the basic questions that we want to ask are:
What have we learned? What are we correcting? What's going to
happen in the future, and how can we get ahead of it before it
does?
Thank you, Mr. Chairman.
The Chairman. Thank you, Senator Nelson.
We do appreciate the great panel of witnesses we have in
front of us today. Thank you all for being here.
On my left and your right is Ms. Monika Bickert, who is the
Head of Product Policy and Counterterrorism, Facebook; Ms.
Juniper Downs, who is Global Head of Public Policy and
Government Relations at YouTube; Mr. Carlos Monje, who is
Director of Public Policy and Philanthropy, U.S. and Canada at
Twitter; and Mr. Clinton Watts, who is a Senior Fellow of the
Foreign Policy Research Institute.
So I'm going to start with Ms. Bickert, and we'll just move
across the table. If you could confine your oral statements as
close to 5 minutes as possible--any additional comments,
obviously, will be included as part of the record--that will
give us an optimal amount of time for members to ask questions.
So thank you so much to all of you for being here. We look
forward to hearing from you.
Ms. Bickert.
STATEMENT OF MONIKA BICKERT, HEAD OF PRODUCT POLICY AND
COUNTERTERRORISM, FACEBOOK
Ms. Bickert. Thank you. Chairman Thune, Ranking Member
Nelson, and distinguished members of the Committee, I'm Monika
Bickert, and I lead Product Policy and Counterterrorism for
Facebook. I'm also a former Federal prosecutor, having spent
more than a decade as an Assistant U.S. Attorney for the
Department of Justice.
The issues we're discussing here today are of the utmost
importance, and on behalf of Facebook, I want to thank you for
your leadership in seeking more effective ways to combat
extremism, crime, and other threats to our national security.
We share your concerns about terrorists' use of the
internet. That's why we remove terrorist content as soon as we
become aware of it. It's also why we have a dedicated
counterterrorism team of people working across our company.
This includes experts like former academics who have spent
their career studying terror groups, tracking new trends and
tactics. It includes former intelligence and law enforcement
officials and prosecutors who have worked in the area of
counterterrorism. It also includes engineers who are constantly
improving the technology that helps us find and remove
terrorist content.
In my written testimony, I describe these efforts in more
detail.
I also want to note that we pursue this goal with a mindset
that it's important to maximize free expression while keeping
people safe online. We work proactively to keep terrorist
content off Facebook, primarily through the use of automated
systems like image matching and text-based machine learning.
Now, more than 99 percent of ISIS and al Qaeda propaganda that
we remove from our service is content that we identify
ourselves before anybody has flagged it for us.
Once we're aware of a piece of terrorist content, we remove
the vast majority of subsequent uploads within one hour. We do
not wait for these global bad actors to upload content to
Facebook before placing it into our detection systems. Rather,
we work with outside experts who track propaganda released by
these groups across the Internet and then send it to us, and we
proactively put it in our systems. Often, this means we're able
to stop this content from ever being uploaded to Facebook.
However, much of this work cannot be done by machines
alone. Accurately removing terrorist content often requires a
person to assess it. A photo of an ISIS fighter, for instance,
that could be shared by somebody who's a supporter of the group
could also be shared by a journalist who's raising awareness or
a member of a civil society group who's condemning violence,
and we need people to be able to assess that and tell the
difference.
We now have more than 7,500 reviewers who assess
potentially violating content, including terrorist content, in
dozens of different languages. By the end of 2018, we will more
than double the 10,000 people who are already working on safety
and security issues across our company.
Now, some of these people are responsible for responding to
law enforcement requests. We appreciate the critical role that
law enforcement plays in keeping people safe, and we do want to
do our part. Our global team responds to valid legal requests
from law enforcement consistent with applicable laws and our
policies, and this includes responding to emergency requests,
where we strive to respond within minutes.
We also want to do our part to stop radicalization and
disrupt the recruitment process. That's why we've commissioned
multiple research efforts over the past three years to
understand how online speech can most effectively counter
violent ideologies, and we've sponsored efforts to put those
learnings into practice. One such example is our peer-to-peer
challenging extremism program, which we sponsor with EdVenture
Partners, and through that program, we've had more than 6,500
students participate. They've created hundreds of campaigns
that have been viewed worldwide more than 200 million times.
No one company can combat the terrorist threat alone. So we
partner with others, including companies, civil society,
researchers, and governments around the world. Among other
things, we work with 11 other companies, including those here,
to maintain a shared industry database of hashes, unique
digital fingerprints of terrorist content, so that we can all
find and remove it faster. We've also recently launched a
global internet forum where we can work with smaller companies
to help them get better.
In conclusion, let me reiterate that we share your goal of
stopping terrorists from using social media, and we're going to
keep getting better at it. I'm here today to listen to your
ideas and your concerns and to continue this constructive
dialog.
Thank you for the opportunity, and I look forward to your
questions.
[The prepared statement of Ms. Bickert follows:]
Prepared Statement of Monika Bickert, Head of Product Policy and
Counterterrorism, Facebook
Introduction
Chairman Thune, Ranking Member Nelson, and distinguished members of
the Committee, thank you for the opportunity to appear before you
today. My name is Monika Bickert, and I am the head of Product Policy
and Counterterrorism at Facebook. Prior to assuming my current role, I
served as lead security counsel for Facebook. I am also a former
prosecutor, having worked for a decade as an Assistant U.S. Attorney
with the Department of Justice. We appreciate the Committee's hard work
as it continues to seek more effective ways to combat extremism, crime,
and other threats to our national security.
We take all of these threats very seriously. One of our chief
commitments is to create and use innovative technology that gives
people the power to build community and bring the world closer
together. Keeping our community safe on Facebook is critical to this
broader mission. We are proud that more than two billion people around
the world come to Facebook every month to share with friends and
family, to learn about new products and services, to volunteer or
donate to organizations they care about, or help in a crisis. The
promise of real connection, of extending the benefits of real world
connections online, is at the heart of what we do and has helped us
grow into a global company.
Being at the forefront of new technology also means being at the
forefront of new legal, security, and policy challenges. My team and
thousands of other Facebook employees around the world come to work
every day to confront these challenges head on. Our goal is to ensure
Facebook is a place where both expression and personal safety are
protected and respected. We appreciate your commitment to these values
as well in your roles as policymakers.
Countering Terrorism on Facebook
I would like to focus my testimony today on the ways Facebook is
addressing the challenge of terrorist propaganda and recruitment
online.
On terrorist content, our view is simple: There is no place on
Facebook for terrorism. Our longstanding policies, which are posted on
our site, make clear that we do not allow terrorists to have any
presence on Facebook. Even if they are not posting content that would
violate our policies, we remove their accounts as soon as we find them.
They simply are not allowed to use our services under any
circumstances. We also remove any content that praises or supports
terrorists or their actions whenever we become aware of it, and when we
uncover evidence of imminent harm, we promptly inform authorities.
We recognize the challenges associated with fighting online
extremism, some of which I will outline in my comments today. We are
committed to being part of the solution, and we are developing
strategies built around both technology and human expertise to address
these threats.
A. Using Technology to Identify and Remove Terrorist Content
One of the challenges we face is identifying the small fraction of
terrorist content posted to a platform used by more than two billion
people every month. Our proactive efforts--specifically, the use of
artificial intelligence (AI) and other automation--have become
increasingly central to keeping this content off of Facebook. We
currently focus our most cutting-edge techniques on combating terrorist
content about ISIS, Al Qaeda, and their affiliates, and we are working
to expand to other terrorist organizations. As we shared recently in a
public blog post, 99 percent of the ISIS and Al Qaeda-related terror
content that we remove from Facebook is detected and removed before
anyone in our community reports it, and in some cases, before it goes
live on the site. We do this primarily through the use of automated
systems like photo and video matching and text-based machine learning.
Once we are aware of a piece of terrorist content, we remove 83 percent
of subsequently uploaded copies within one hour of upload.
Importantly, we do not wait for ISIS or Al Qaeda to upload content
to Facebook before placing it into our internal detection systems.
Rather, we use a variety of techniques, including consulting external
experts, to track propaganda released by these groups and proactively
insert it into our matching systems. Often, this means we are able to
prevent its upload to Facebook entirely.
Because terrorists also adapt as technology evolves, we are
constantly updating our technical solutions. I would like to share with
you today several specific examples of the ways we are using technology
to stay ahead of terrorist activity and combat terrorism online.
1. Image Matching and Language Understanding
When someone tries to upload a terrorist photo or video, our
systems look for whether the image matches a known terrorism photo or
video. This means that if we previously removed an ISIS propaganda
video, for example, we can work to prevent other accounts from
uploading the same video to our site.
We also have started experimenting with using AI to understand text
that potentially advocates for terrorism. We are working to develop
text-based signals to detect praise or support of terrorist
organizations. These signals will be incorporated into an algorithm
that is in the early stages of learning how to detect similar posts.
2. Removing Terrorist Clusters
We know from the many terrorism academics and experts we work with
that terrorists tend to radicalize and operate in clusters. This
offline trend is reflected online as well. As such, when we identify
Pages, groups, posts, or profiles that support terrorism, we use AI to
identify related material that may also support terrorism. As part of
that process, we utilize a variety of signals, including whether an
account is ``friends'' with a high number of accounts that have been
disabled for terrorism, or whether an account shares the same
attributes as a disabled account.
3. Identifying Repeat Offenders
When we disable terrorist accounts, those account owners may try to
create new accounts using different identities. We have become faster
at using technology to detect new fake accounts created by repeat
offenders, or recidivists. Through this work, we have been able to
dramatically reduce the time period that terrorist recidivist accounts
are on Facebook.
4. Cross-Platform Collaboration
Because we prohibit terrorists from maintaining a presence anywhere
in the family of Facebook applications, we have begun work on systems
that enable us to remove terrorist accounts across all of our
platforms, including WhatsApp and Instagram. Given the limited data
some of our applications collect as part of their service, this ability
to share data helps immensely in keeping all of our applications safe.
These are some of our key tools, but there are other tools as well.
Our ability to outline them here is, however, constrained by the need
to avoid providing a roadmap to bad actors seeking to evade detection.
B. Human Expertise
Identifying terrorist content often requires analyzing the relevant
context, and we know we cannot rely on AI alone to identify and remove
terrorist content. For example, a photo of an armed man waving an ISIS
flag could be propaganda or recruiting material, or it could be an
image in a major news story. To understand more nuanced cases, we need
human expertise.
Our community of users helps us by reporting accounts or content
that may violate our policies--including the small fraction that may be
related to terrorism. Our content review teams around the world--which
grew by 3,000 people last year--work 24 hours a day and in dozens of
languages to review these reports. More broadly, by the end of 2018 we
will more than double the number of people working on safety and
security, including terrorism issues, from 10,000 to 20,000.
We also have significantly grown our team of counterterrorism
specialists. Distinct from our content review teams, we have more than
150 highly trained people who are exclusively or primarily focused on
preventing terrorist content from ever appearing on our platform and
quickly and identifying and removing it if it does. This group includes
former academics who are experts on counterterrorism, former
prosecutors and law enforcement agents, investigators and analysts, and
engineers. Within this specialist team alone, we speak nearly 30
languages.
C. Partnering with Others
We are proud of the work we have done to make Facebook a hostile
place for terrorists. We understand, however, that simply working to
keep terrorism off Facebook is an inadequate solution to the problem of
online extremism, particularly because terrorists are able to leverage
a variety of platforms. We believe our partnerships with others--
including other companies, civil society, researchers, and
governments--are crucial to combating this threat.
To this end, we have partnered with our industry counterparts to
more quickly identify and slow the spread of terrorist content online.
For example, in December 2016, we joined with Microsoft, Twitter, and
YouTube to announce the development of a shared industry database of
``hashes''--unique digital fingerprints for photos and videos--for
content produced by or in support of terrorist organizations. The
database now contains more than 40,000 hashes, and the consortium of
companies has increased to include twelve companies.
This past summer, we formalized our relationship with industry
partners and announced the Global Internet Forum to Counter Terrorism
(GIFCT), an endeavor that focuses on knowledge sharing, support for
counterterrorism work, and technical cooperation, as represented by the
hash consortium. Already, this endeavor has brought together more than
68 technology companies over the course of international working
sessions held on three continents. This effort gives structure to our
existing and future areas of collaboration and fosters cooperation with
smaller tech companies, civil society groups, academics, governments,
and international bodies such as the EU and the UN.
We engage with governments and inter-governmental agencies around
the world and we recently commissioned a research consortium led by the
Brookings Institute and the Royal United Services Institute to examine
how governments, tech companies, and civil society can work together to
fight online extremism and radicalization. We have learned much through
briefings from agencies in different countries about extremist
organizations' propaganda mechanisms. We also have participated in and
benefited from efforts to support industry collaboration by
organizations such as the National Counterterrorism Center (NCTC), the
EU Internet Forum, the Global Coalition Against Daesh, and the UK Home
Office.
In recent months, we have further expanded our partnerships with
several organizations including Flashpoint, the Middle East Media
Research Institute (MEMRI), the SITE Intelligence Group, and the
University of Alabama at Birmingham's Computer Forensics Research Lab.
These organizations report Pages, profiles, and groups on Facebook that
are potentially associated with terrorist groups. They also send us
photo and video files associated with ISIS and Al Qaeda that they have
located elsewhere on the internet. We check this information against
our algorithms for file ``matches,'' in order to remove or prevent
upload of the files to Facebook in the first instance.
We appreciate the critical role that law enforcement plays in
keeping people safe. Our legal and safety teams work hard to respond to
legitimate law enforcement requests while fulfilling our responsibility
to protect people's privacy and security. We have a global team that
strives to respond within minutes to emergency requests from law
enforcement. We provide the information that we can in response to law
enforcement requests, consistent with applicable law and our policies.
For example, in the first half of 2017, we provided information in
response to more than 75 percent of the 1,864 requests for emergency
disclosures that we received from U.S. law enforcement agencies.
Preventing Recruitment Through Counterspeech
We believe that a key part of combating extremism is preventing
recruitment by disrupting the underlying ideologies that drive people
to commit acts of violence. That's why we support a variety of
counterspeech efforts. Although counterspeech comes in many forms, at
its core these are efforts to prevent people from pursuing a hate-
filled, violent life or convincing them to abandon such a life.
Over the past three years, we have commissioned research on what
types of counterspeech are the most effective at combating hate and
violent extremism. Based on that research, we believe the credibility
of the speaker is incredibly important. We have therefore partnered
with non-governmental organizations and community groups around the
world to empower positive and moderate voices. For example, two years
ago, we worked with the Institute for Strategic Dialogue to launch the
Online Civil Courage Initiative, a project that has engaged with more
than 100 anti-hate and anti-extremism organizations across Europe. We
also have worked with Affinis Labs to host hackathons in places like
Manila, Dhaka, and Jakarta, where community leaders joined forces with
tech entrepreneurs to develop innovative solutions to challenge
extremism and hate online. Finally, we worked with EdVenture Partners
to develop a peer-to-peer student competition called the Facebook
Global Digital Challenge (P2P). This is a semester-long university
course during which students build a campaign to combat extremism in
their area, launch it, track its success, and then submit the results
as part of a global competition. The University of Central Oklahoma
recently implemented a student-led counterspeech program through P2P
that uses social media to encourage people to challenge their beliefs
and stereotypes. In less than three years, these P2P projects have
reached more than 56 million people worldwide through more than 500
anti-hate and extremism campaigns created by more than 5,500 university
students in 68 countries.
Conclusion
In conclusion, let me reiterate our commitment to combating
extremism on our platform. We have a responsibility to do all we can to
combat these threats, and we are committed to improving our efforts.
Of course, companies like Facebook cannot do this without help. We
will continue to partner with appropriate authorities to counteract
these threats. By working together, business, government, and civil
society can make it much harder for malicious actors to harm us, while
simultaneously ensuring that people can express themselves freely and
openly. I am here today to listen to your ideas and concerns, and I
look forward to continuing this constructive dialogue.
The Chairman. Thank you, Ms. Bickert.
Ms. Downs.
STATEMENT OF JUNIPER DOWNS, DIRECTOR, PUBLIC POLICY AND
GOVERNMENT RELATIONS, YOUTUBE
Ms. Downs. Thank you, Senator. Chairman Thune, Ranking
Member Nelson, and distinguished members of the Committee,
thank you for the opportunity to testify at today's hearing and
for your leadership on these difficult issues. My name is
Juniper Downs, and I serve as the Global Public Policy Lead for
YouTube.
At YouTube, we believe the world is a better place when we
listen, share, and build community through our stories. Our
mission is to give everyone a voice and show them the world. We
see over 400 hours of video uploaded to YouTube every minute.
With this comes many benefits to society: unparalleled access
to art and culture, news and entertainment, educational
materials, a remarkable diversity of viewpoints, and the
freedom to exchange ideas. We value this openness. It has
democratized how stories and whose stories get told.
We are aware, however, that the very platforms that have
enabled these societal benefits may also be used by those who
wish to promote hatred or extremism. To that end, I'm pleased
to have this opportunity to outline the approach we've taken on
these issues.
We've developed rigorous policies and programs to defend
against the use of our platform to spread hate or incite
violence. YouTube has long had policies that strictly prohibit
terrorist content. This includes terrorist recruitment, violent
extremism, incitement to violence, glorification of violence,
and videos that teach people how to commit terrorist attacks.
We apply these policies to violent extremism of all kinds,
whether inciting violence on the basis of race or religion or
as part of an organized terrorist group.
We use a mix of technology and humans to remove violent
content quickly. Users can alert us to content they think may
violate our policies through a flag found below every YouTube
video. We have teams charged with reviewing flagged content 24/
7 in multiple languages and countries around the world.
We also work closely with members of our trusted flagger
programs, NGOs who provide highly actionable flags and have
expertise on issues like hate speech and terrorism, and, of
course, we rely on our technology, which has always been a
critical part of our solution. Our image-matching techniques,
for example, can prevent the dissemination of violent content
by catching re-uploads of known bad content before it becomes
public.
Nonetheless, given the evolving nature of the threat, it's
necessary for us to continue enhancing our systems. Over the
past year, in particular, we've taken several steps to build on
our efforts. The first is an investment in machine learning
technologies for the detection and removal of violent extremist
videos. We recently deployed classifiers that detect new
terrorist content and flag it for review. Machine learning is
now helping our human reviewers remove nearly five times as
many videos as they were before. Today, 98 percent of the
videos we remove for violent extremism were identified by our
algorithms.
Second, we are focused on improving and expanding our
expertise and resources on these issues. We expanded our
trusted flagger program to an additional 50 NGOs in 2017,
including several counterterrorism experts. Working with these
organizations helps us to better identify emerging trends and
understand how these issues manifest and evolve. In 2018, we
will have 10,000 people across Google working to address
content that might violate our policies.
Finally, we're creating programs to promote counter-speech
on our platforms. Our Creators for Change program supports
YouTube creators who are tackling issues like extremism and
hate by building empathy and acting as positive role models.
Google's Jigsaw group has deployed the redirect method, which
uses targeted ads and YouTube videos to disrupt online
radicalization.
We also collaborate across the industry. In 2016, we
created a hash-sharing database with Facebook, Microsoft, and
Twitter, where we share digital fingerprints of terrorist
content to stop its spread across platforms. We added seven
companies to this coalition in 2017, and our shared database
now contains over 50,000 video and image hashes. Last summer,
we announced the Global Industry Forum to Counter Terrorism to
formalize industry collaboration on research, knowledge
sharing, and technology.
No single component or party can solve this problem in
isolation. To get it right, we must all work together. We
understand the importance of speed and comprehensiveness in our
work. Since June, we've removed 160,000 videos and terminated
30,000 channels for violent extremism. We've taken down nearly
70 percent of violent extremist videos within 8 hours of upload
and nearly half within two hours. We've reviewed over 2 million
videos to make sure we're catching and removing all videos that
violate these policies.
We achieved these results through enhanced enforcement by
machines and people and collaboration with outside experts.
We're deeply committed to working with law enforcement,
government, the tech industry, and NGOs to protect our services
from being exploited by bad actors. We look forward to
continued collaboration with the Committee as it examines these
issues.
Thank you for your time. I look forward to your questions.
[The prepared statement of Ms. Downs follows:]
Prepared Statement of Juniper Downs, Director, Public Policy and
Government Relations, YouTube
Chairman Thune, Ranking Member Nelson, and distinguished Members of
the Committee: thank you for the opportunity to testify at today's
hearing and for your leadership on these difficult issues. My name is
Juniper Downs and I serve as the global policy lead for YouTube.
At YouTube, we believe the world is a better place when we listen,
share, and build community through our stories. Our mission is to give
everyone a voice and show them the world. With this comes many benefits
to society--unparalleled access to art and culture, news and
entertainment, and educational materials. To put our work in context,
it's important to recognize the scale and goal of our services. More
than one and a half billion people come to YouTube every month. We see
well over 400 hours of video uploaded every minute. Most of this
content is perfectly benign--beauty vlogs, music, comedy. Digital
platforms have also become a place for breaking news, exposing
injustices, and sharing content from previously inaccessible places.
We value this openness. It has democratized how stories, and whose
stories, get told. And has created a platform where anyone can be a
creator and can succeed. We are aware, however, that the very platforms
that have enabled these societal benefits may also be abused by those
who wish to promote hatred or extremism. These challenges are
constantly evolving and changing, so our commitment to combat them is
similarly sustained and unwavering. To be very clear: using YouTube to
incite violence, spread violent extremist propaganda, recruit for
terrorism, or celebrate or promote terrorist attacks is strictly and
specifically prohibited by our terms of service.
To that end, I am pleased to have this opportunity to outline the
approach we have taken on these issues. We have developed rigorous
policies and programs to defend the use of our platforms from the
spread of hate and incitement to violence. We continue to refine them
as we adapt to new and evolving threats. For example, YouTube has long
had policies that prohibit terrorist content. This includes: terrorist
recruitment, violent extremism, incitement to violence, and
instructional content that could be used to facilitate substantial
bodily injury or death. Extremism and violence are not confined to any
one community. We apply these policies to violent extremism of all
kinds, whether inciting violence on the basis of race or religion or as
part of an organized terrorist group. When we become aware of content
that violates these policies, we immediately remove it. Any channel
that is dedicated to such content is terminated. We don't allow Foreign
Terrorist Organizations (FTOs) to use Google at all--if an account is
created by an FTO or its agent, we terminate immediately, regardless of
the content it may be sharing.
We also have a strict set of policies for monetizing content on
YouTube. We recognize there may be videos that don't break our
Community Guidelines, but which advertisers would not want to advertise
against. We give advertisers the tools to control where their ads
appear.
We use a mix of technology and humans to remove violative content
quickly. Users can alert us to content that they think may violate our
policies through a flag found below every YouTube video. We have teams
charged with reviewing flagged content 24/7 in multiple languages and
countries around the world. We also work closely with members of our
Trusted Flagger program, which is comprised of NGOs and government
agencies with specific expertise who are provided a bulk-flagging tool
to alert us to content that may violate our policies. And of course we
rely upon our technology, which has always been a critical part of our
solution. Our video-matching techniques, for example, can prevent the
dissemination of violative content by catching re-uploads of known bad
content before it is public.
Nonetheless, given the evolving nature of the threat, it is
necessary for us to continue enhancing our systems. We know that no
enforcement regime will ever be 100 percent perfect. Over the past year
in particular, we have taken several steps to build on our efforts:
The first is an investment in machine learning technologies
for the detection and removal of violent extremist videos. We
have been working on machine learning for years, and recently
deployed classifiers that detect terrorist material and flag it
for review. Since June, our teams have manually reviewed
approximately two million videos to improve this flagging
technology by providing large volumes of training examples.
Machine learning is now helping our human reviewers remove
nearly five times as many videos in violation of our policies
than they were previously. Last June, only 40 percent of the
videos we removed for violent extremism were identified by our
algorithms. Today, that number is 98 percent. Our advances in
machine learning let us now take down nearly 70 percent of
violent extremism content within 8 hours of upload and nearly
half of it in 2 hours.
Second, we are focused on improving and expanding our
expertise and resources on these issues. We expanded our
Trusted Flagger Program to an additional 50 NGOs in 2017,
including to groups like Anti-Defamation League and several
counter-terrorism experts such as the Institute of Strategic
Dialogue and International Centre for the Study of
Radicalization. Working with these organizations helps us to
better identify emerging trends and understand how these issues
manifest and evolve. In 2018, we will have 10,000 people across
Google working to address content that might violate our
policies. This includes engineers and reviewers who work around
the world, 24/7, and speak many different languages.
We are taking a tougher stance on videos that may be
offensive, but do not violate our policies. Our Community
Guidelines prohibit hate speech that either promotes violence
or has the primary purpose of inciting hatred against
individuals or groups based on certain attributes. Some
borderline videos, such as those containing inflammatory
religious or supremacist content without a direct call to
violence or a primary purpose of inciting hatred, may not cross
these lines for removal. But we understand that these videos
may be offensive to many and have developed a new treatment for
them. Identified borderline content will remain on YouTube
behind an interstitial, won't be recommended, won't be
monetized, and won't have key features including comments,
suggested videos, and likes. Initial uses have been positive
and have shown a substantial reduction in watch time of those
videos.
Greater Transparency. We understand that people want a
clearer view of how we're tackling problematic content. That's
why in 2018, we will be creating a report to provide more
aggregate data about the flags we receive and the actions we
take to remove videos and comments that violate our content
policies.
Finally, we are creating programs to promote counterspeech
on our platforms. We are expanding our counter-extremism work
to present counternarratives and elevate the voices that are
most credible in speaking out against terrorism, hate, and
violence.
For example, our Creators for Change program supports
creators who are tackling social issues, including
extremism and hate, by building empathy and acting as
positive role models. There are 60 million video views of
Creators for Change content to date; 731,000 total watch
time hours of Creators for Change content; and through
`Local chapters' of Creators for Change, creators tackle
social challenges specific to different markets.
Google's Jigsaw group, an incubator to tackle some of
the toughest global security challenges, has deployed the
Redirect Method, which uses Adwords targeting tools and
curated YouTube videos uploaded to disrupt online
radicalization. It focuses on the slice of ISIS's audience
that is most susceptible to its messaging and redirects
them towards YouTube playlists of videos debunking ISIS
recruiting themes.
We also collaborate across the industry. In 2016, we created a
hash-sharing database with Facebook, Twitter and Microsoft, where we
share hashes (or ``digital fingerprints'') of terrorist content to stop
its spread across platforms. Using other companies to give us notice is
effective because of the counter-terrorism research showing the pattern
of cross-platform abuse and the particularly dangerous nature of this
content. We added 7 companies to this coalition in 2017 and our shared
database contains over fifty thousand videos and image hashes. Last
summer, we announced the Global Industry Forum to Counter Terrorism
(GIFCT) to formalize industry collaboration on research, knowledge
sharing, and technology. The GIFCT also set a goal of working with 50
smaller tech companies in 2017 to help them better tackle terrorist
content on their platforms--and we exceeded that goal. To date, we've
hosted 68 small companies at workshops through the Tech Against
Terrorism Initiative, our partners under the UN Counter Terrorism
Executive Directorate. We've held workshops for smaller companies in
San Francisco and New York, Paris, Jakarta, London, and Brussels.
No single component can solve the problem in isolation. To get this
right, we must all work together. Since June, YouTube has removed over
160,000 violent extremist videos and has terminated approximately
30,000 channels for violation of our policies against terrorist
content. We achieved these results through tougher policies, enhanced
enforcement by machines and people, and collaboration with outside
experts. That has become the blueprint for how we tackle this
challenge.
While Google's services can provide real benefits to our users, we
recognize that detecting and preventing misuse of those services is
critically important. We are deeply committed to working with law
enforcement, government, others in the tech industry, and the NGO
community to protect our protect our services from being exploited by
bad actors. We will only make progress by working together to address
these complex issues at their root. That is why forums like this are so
important to underscoring our shared goals and commitments. We look
forward to continued collaboration with the Committee as it examines
these issues.
Thank you for your time. I look forward to taking your questions.
The Chairman. Thank you, Ms. Downs.
Mr. Monje.
STATEMENT OF CARLOS MONJE, JR., DIRECTOR, PUBLIC POLICY AND
PHILANTHROPY, U.S. AND CANADA, TWITTER
Mr. Monje. Thank you, Chairman Thune, Ranking Member
Nelson, distinguished members of the Committee, and staff.
I'm here on behalf of Twitter, an open communications
platform that allows more than 330 million users to see what's
happening in the world and to share viewpoints from every side.
Each day, we serve 500 million tweets. We have about 3,700
employees around the world.
Twitter has been at the forefront of preventing terrorist
exploitation of the internet. Our work in this area will never
be complete as the threats we face constantly evolve. As new
challenges emerge, we will continue our efforts to both ensure
terrorists don't have a place on Twitter while also giving
voice to those who promote a positive message for the future.
Twitter has a zero tolerance policy for terrorist content.
This includes not only specific threats of violence, but also
promoting terrorism, affiliating with violent extremist groups,
and glorifying violence. Our job is to enforce this policy
globally, at scale, to evolve to stay one step ahead of the
terrorists. We have dramatically improved our ability to
implement these rules and have suspended more than 1.1 million
terrorist accounts since mid 2015.
Our progress fighting terrorist content is due to our
commitment to innovation. While there is no ``magic algorithm''
for identifying terrorist content, we have increasingly
improved the effectiveness of our in-house proprietary
technology. Our technology supplements user reports, human
review, and it significantly augments our ability to identify
and remove bad content from Twitter.
At the beginning of 2015, our in-house technology detected
roughly a third of the terrorist accounts that we pulled down
at that time. Last year, that number increased dramatically. We
identified more than 90 percent of suspensions for terrorism by
our internal tools, and 75 percent or three-quarters of those
accounts were suspended before they had a chance to tweet even
once. Let me repeat that because it's important. We spot more
than 90 percent of terrorist accounts before anyone else does,
and we stop 75 percent of those accounts before they can spread
any of their deplorable ideology.
Of course, like any determined adversary, as we make it
harder for terrorists to use Twitter, their behavior evolves.
To stay in front of this, we continue to invest in technology
to prevent new accounts from being opened to replace those that
we suspend while also developing further the tools that prevent
the distribution of propaganda in the aftermath of attacks.
Because this is a shared challenge, our industry has
established the Global Internet Forum to Counter Terrorism,
which is focused on learning and collaboration, on technical
cooperation, and research. Twitter sees the forum as a
substantial opportunity to ensure that smaller companies are
not soft targets for terrorists. We have engaged with 68
smaller companies over the past several months to share best
practices and learnings, and we plan to grow on that work.
Removing a tweet doesn't eliminate the ideology behind it,
so we invest heavily in alternative narratives. Twitter has
participated in more than 100 NGO trainings and events around
the world since 2015. We work with respected organizations to
empower credible, non-governmental voices against violent
extremism.
As part of a continuing effort to make Twitter a safe place
for open democratic debate, late last year, we broadened our
rules to prohibit accounts affiliated with violent extremist
groups and to make hateful imagery much harder to find on our
platform. We also stepped up our enforcement of abuse reported
by witnesses and increased transparency about our enforcement
decisions to further educate our users about our terms of
service.
Twitter has also devoted significant resources to combat
disinformation and election interference by foreign state
actors. To prepare for the U.S. midterm elections this year, a
cross-functional elections task force is prepared to verify
major party candidates as a hedge against impersonation, to
maintain open lines of communication with Federal and State
election officials, to continually improve and apply our
technology to address networks of malicious automation, and to
monitor trends and spikes in conversations related to the
elections.
The companies here today have both shared and unique
challenges, and while we are competitors in the marketplace, we
are close partners in combating the threats of extremism and
those who would harm our democratic process.
Thank you. Thank you for your leadership on these issues. I
look forward to this discussion.
[The prepared statement of Mr. Monje follows:]
Prepared Statement of Carlos Monje, Jr., Director, Public Policy and
Philanthropy, U.S. and Canada, Twitter
Thank you Chairman Thune, Ranking Member Nelson, and distinguished
members of the Committee and staff. Twitter has been at the forefront
of responding to the evolving challenge of preventing terrorist
exploitation of the Internet. Our work in this area will never be
complete, as the threats we face constantly evolve. As new challenges
emerge, we will continue our efforts to both ensure terrorists don't
have a place on Twitter while giving voice to those who promote
positive messages for the future.
The Twitter Rules
To be clear, terrorist organizations have no place on Twitter and
the promotion of terrorism is against our Rules. The Twitter Rules make
clear:
You may not make specific threats of violence or wish for
the serious physical harm, death, or disease of an individual
or group of people. This includes, but is not limited to,
threatening or promoting terrorism. You also may not affiliate
with organizations that--whether by their own statements or
activity both on and off the platform--use or promote violence
against civilians to further their causes.
Moreover, our Rules prohibit content that glorifies violence or the
perpetrators of a violent act. This includes celebrating any violent
act in a manner that may inspire others to replicate it or any violence
where people were targeted because of their membership or inclusion in
a protected group.
Terrorist Content Removals
Beyond having a clear policy against the promotion of terrorism, we
have been tackling the issue of terrorist content on the Twitter
platform for many years. As our biennial Twitter Transparency Reports
indicate we have made steady progress in this area:
In 2015, 67,069 accounts suspended
In 2016: 569,202 accounts suspended
In 2017: 574,070 accounts suspended
In total, we have suspended more than 1.1 million terrorist
accounts since mid-2015.
Technology
Our progress fighting terrorists on Twitter is due to the
commitment we have made internally to harness innovation to address the
tactics employed by terrorist organizations on our platform. While
there is no ``magic algorithm'' for identifying terrorist content on
the Internet, we have increasingly tapped technology in efforts to
improve the effectiveness of our in-house proprietary anti-spam
technology. This technology supplements reports from our users and
dramatically augments our ability to identify and remove violative
content from Twitter. Through these efforts we have found success in
preventing recently suspended users from coming back onto Twitter.
At the beginning of 2015, this technology was being used to detect
roughly one-third of the terrorist accounts we suspended at that time.
Last year, that number increased to more than 90 percent of suspensions
being flagged by our internal tools. Three-quarters of those
suspensions were flagged before the account had a chance to Tweet even
once. As is the case with a determined adversary, as we make it harder
for terrorists to use Twitter, their behavior evolves. To stay in front
of this, we continue to invest in technology to prevent new accounts
being opened to replace those we suspend, while also developing further
the tools that prevent the distribution of propaganda in the aftermath
of attacks.
Industry Collaboration
Because this is a shared challenge, our industry has established
the Global Internet Forum to Counter Terrorism (GIFCT), which has
focused on learning and collaboration; technical cooperation; and
research. This builds on previous work undertaken by the EU Internet
Forum and follows constructive discussions held at the UN General
Assembly and the G7 Interior Ministers meeting. Twitter sees the GIFCT
as a substantial opportunity to ensure that smaller companies are not
soft targets for terrorists and that the learnings that we have
developed are shared and built upon. The GIFCT's initial goal for 2017
was to work with 50 smaller tech companies to share best practices on
how to disrupt the spread of violent extremist material. We have
exceeded that goal, engaging with 68 companies over the past several
months.
In the coming months, we plan to deepen this collaboration with
smaller companies, working directly to educate them about potential
technological approaches, sharing expertise from our own operational
teams and allowing them to develop a peer network across industry to
support their work.
Twitter Countering Violent Extremism Trainings
The GIFCT, through its work with the Tech Against Terrorism and
ICT4Peace projects, is a further avenue through which best practices
can be shared and our existing company efforts can be further scaled-
up. Twitter has participated in more than 100 CVE trainings and events
since 2015, including events in Beirut, Bosnia, Belfast and Brussels
and summits at the White House, the United Nations and in London and
Sydney.
We work with respected organizations such as Parle-moi d'Islam
(France), Active Change Foundation (UK), Wahid Foundation (Indonesia),
The Sawab Center (UAE), and True Islam (US) to empower credible non-
governmental voices against violent extremism. We also continue to play
an active role in the task force created by the French Interior
Ministry and have attended government-convened summits on CVE hosted by
the French Interior Ministry and the Indonesian National
Counterterrorism Agency.
We supported the Institute for Strategic Dialogue's ``Against
Violent Extremism'' project, the results of which were published in a
report, ``The Impact of Counternarratives.'' The project used pro bono
Twitter advertising to increase the reach of key NGOs. The campaigns
yielded real results. One NGO participant, Average Mohamed, doubled its
number of Twitter followers and another, ExitUSA, tripled its Twitter
followers.
We also are a member of the Anti-Defamation League's Cyberhate
Problem-Solving Lab, which works collaboratively to counter hate speech
online.
Extremism
Late last year we broadened our rules to encompass accounts
affiliated with violent extremist groups and to cover violent content
or hateful imagery displayed in profile information. Our prohibition on
the use of Twitter's services by violent extremist groups--i.e.,
identified groups subscribing to the use of violence as a means to
advance their cause--applies whether the purpose or cause of any such
group is a political, religious, or social objective.
Accounts affiliated with groups or organizations in which violence
is a component of advancing their cause risk having a chilling effect
on opponents of that cause who may want to comment on Twitter. In
addition, the violence that such groups promote online could also have
dangerous consequences offline, potentially putting the physical safety
of Twitter users and others in jeopardy.
The broadening of our policies against violent extremism also
includes covering any account that abuses or threatens others through
their profile information. In other words, if an account's profile
information includes a violent threat or multiple slurs, racist or
sexist tropes, or incites fear or otherwise dehumanizes another person,
it will be removed. Further, hateful imagery will now be included in
the category of ``sensitive media'' under our rules. This change means
that logos, symbols, or images whose purpose is to promote hostility
and malice to others based upon their race, religion, disability,
sexual orientation, or ethnicity will now be actionable and we will
require accounts displaying such imagery to remove such violative media
content.
Misinformation
As we have previously described, Twitter has also devoted
significant resources to the issue of misinformation and interference
in the election context by foreign state actors. We have sought through
our Information Quality initiative to enhance our ability going forward
to detect and stop such activity and to do our part to protect the
democratic process from interference and abuse. We have also undertaken
a retrospective review to further the public's understanding of what
happened in the 2016 election. As we explained last year, we expect to
keep Congress updated on the latest results of that ongoing review as
our work progresses. And we made the decision last year not only to
offboard both RT and Sputnik as advertisers on our platform, but also
to commit to donate the revenue we received from those sources to
research into elections and civic engagement on Twitter. We have begun
to scope such research needs and are in dialogue with several academic
researchers and NGOs in this area. We take these issues seriously and
our efforts to address them remain among our highest priorities.
Preparing for the U.S. Midterms
Since 2016 we've had additional elections around the world--such as
in France, Germany, and South Korea during 2017--and we have midterm
elections approaching this November in the United States.
To prepare for the U.S. midterm elections, we have organized
internally to ensure that our teams are working to address election-
related issues as they may arise. Our cross-functional elections task
force will be prepared to:
Verify major party candidates for all statewide and Federal
elective offices, and major national party accounts, as a hedge
against impersonation;
Maintain open lines of communication to Federal and state
election officials to quickly escalate issues that arise;
Address escalations of account issues with respect to
violations of Twitter Rules or applicable laws;
Continually improve and apply our anti-spam technology to
address networks of malicious automation targeting election-
related matters;
Monitor trends and spikes in conversations relating to the
2018 elections for potential manipulation activity; and
Implement our Ads Transparency Center to bring transparency
to voters about political ads they see on Twitter.
The companies here today have both shared and unique challenges.
And while we are competitors in the marketplace, we are close partners
in combating the threat of extremism or those who would do harm to our
democratic process. We value the collaboration with our industry peers,
and coordinated efforts are driving further progress to degrade the
presence of content promoting terrorism.
Thank you, and I look forward to this discussion.
The Chairman. Thank you, Mr. Monje.
Mr. Watts.
STATEMENT OF CLINT WATTS, ROBERT A. FOX FELLOW,
FOREIGN POLICY RESEARCH INSTITUTE; SENIOR FELLOW, CENTER FOR
CYBER AND HOMELAND SECURITY, THE GEORGE WASHINGTON UNIVERSITY;
AND NON-RESIDENT FELLOW, ALLIANCE FOR SECURING DEMOCRACY,
GERMAN MARSHALL FUND OF THE UNITED STATES
Mr. Watts. Chairman Thune, members of the Committee, thanks
for having me here today.
Ten years ago, it was al Qaeda in Iraq videos on YouTube. A
few years later, al Shabaab's deadly rampages played out on
Twitter. Shortly after, Facebook groups and Twitter feeds
brought the Islamic State to the world's attention and into the
homes of new recruits before they scurried off to other social
media platforms like Telegram. And 4 years ago, amongst global
jihad's social media storm, I stumbled into Russian influence
campaigns, their reboot of an old playbook called ``Active
Measures,'' which they've deployed across nearly every social
media platform with devastating effect.
Social media at its height offered a platform for
discussion across diverse audiences and led to uprisings
toppling dictators during the Arab Spring. But bad actors with
motivation, money, manpower, and know-how will always come to
these information gateways to pursue their objectives. Lesser
educated populations around the world predominately arriving in
cyber space via mobile phones will be particularly vulnerable
to the social media manipulation of terrorists and
authoritarians.
American focus on the Islamic State social media
recruitment or Russian meddling in the Presidential election of
2016 overlooks other indicators of damaging activity. American
companies have suffered and remain particularly vulnerable to
smear campaigns launched by foreign state actors through
malicious false narratives, pushing their stock prices down and
decreasing sales through reputational damage.
Beyond just smear campaigns and character assassinations,
this committee should take seriously the ability of foreign
nations to mobilize violence inside the U.S. through an
evolution I would call ``Anwar Awlaki Meets PizzaGate.'' Just a
few years ago, Anwar Awlaki, al Qaeda's external operations
leader in Yemen, recognized the power of the Internet to
recruit and mobilize terrorists in America to conduct violence
in the U.S. homeland.
The Islamic State took this to another level with their
spokesman, Abu Muhammad al-Adnani, calling on supporters to
conduct attacks at home and then further enabling those e-
recruits by using a social media battalion to guide those plots
remotely. A little over a year ago, America saw an individual
consume a false conspiracy on the internet and social media,
known as PizzaGate, and then travel to Washington, D.C., to
investigate those claims. He arrived at a falsely implicated
restaurant and discharged a weapon before being arrested.
Surely, a foreign adversary of the United States sees an
opportunity in combining these two scenarios. The greatest
concern moving forward might likely be a foreign intelligence
service posing as Americans on social media, infiltrating one
or both political extremes in the U.S., and then recruiting
unwitting Americans to undertake violence against a target of
the foreign power's choosing. Social media companies will be
better positioned to stop this potential scenario from
occurring than U.S. intelligence or Homeland Security that are
blind to the technical signatures behind this manipulation.
Social media companies realize the damage of these bad
actors far too late. They race to implement policies to prevent
the last information attack, but have yet to anticipate the
next abuse of their social media platforms by emerging threats.
I've offered a range of recommendations for how to counter bad
actors using social media in previous testimony. I'll focus on
a few issues here today.
The first and most pressing challenge comes in the debate
over social media account anonymity versus authenticity.
Anonymity of social media accounts has in many cases allowed
the oppressed and the downtrodden to speak out about injustice.
But over time, anonymity has empowered hackers, extremists, and
authoritarians. Under the veil of anonymity, they spread hate,
recruit members, and advance divisions in American society.
Social media companies can and should protect the public
anonymity of account holders if their user chooses, but they
must be able to determine a real person resides behind each
persona. Social media companies have better advanced tools
recently to certify authenticity. However, the current level of
authenticity on the Twitter platform is suboptimal. I'd
encourage Twitter to rapidly expand its verification to as many
users as possible as quickly as possible.
Closely connected to the issue of account authenticity is
the rise of computational propaganda. The negative effects of
social bots far outweigh any benefits. The anonymous
replication of accounts that routinely broadcast high volumes
of misinformation can pose a serious risk to public safety and,
when employed by authoritarians, a direct threat to democracy.
Last, social media companies continue to get beat in part
because they rely too heavily on technologists and technical
detection to catch bad actors. Artificial intelligence and
machine learning will greatly assist in cleaning up nefarious
activity, but will for the near future, fail to detect that
which hasn't been seen before. Those who understand the
intentions and actions of criminals, terrorists, and
authoritarians must work alongside technologists to sustain the
integrity of these social media platforms.
I know it is unreasonable to think that every social media
company can and should hire threat analysts for every possible
emerging threat. But a variety of rapid outreach approaches
with external social media analysts and threat experts
positioned outside social media companies could easily be
developed or even be collectively sponsored by social media
companies. Several models from counterterrorism and
cybersecurity could be adopted by Silicon Valley in this
regard. I've made other recommendations in the past which I can
address during the Q and A.
But, in conclusion, some social media companies have done
more than others to improve the safety and integrity of their
platforms. Others have a lot of work to do to improve their
platforms against bad actors. Ultimately, the American consumer
will decide whether the benefits of using these services
outweigh the risks. Many are walking away from social media
applications because they can't trust the information being
shared or tolerate the vitriolic user experience.
Social media companies should move aggressively to thwart
terrorists and authoritarians exploiting their systems not only
because it's what's best for their users and society, but
because it's good for business as well.
Thank you for having me.
[The prepared statement of Mr. Watts follows:]
Prepared Statement of Clint Watts, Robert A. Fox Fellow, Foreign Policy
Research Institute; Senior Fellow, Center for Cyber and Homeland
Security, the George Washington University; and Non-Resident Fellow,
Alliance For Securing Democracy, German Marshall Fund of the United
States
Ten years ago, it was al Qaeda in Iraq videos on YouTube. A few
years later, al Shabaab's deadly rampages played out on Twitter.
Shortly after, Facebook groups and Twitter feeds brought the Islamic
State to the world's attention and into the homes of new recruits,
before they scurried off to other social media applications like
Telegram. And four years ago amongst global jihad's social media storm,
I stumbled into Russian influence campaigns, their reboot of an old
playbook called ``Active Measures'', which they've deployed across
nearly all social media platforms with devastating effect.
Today, disinformation spread on Facebook propels deadly violence in
Myanmar against the minority Ronhingya population.i The
Duterte regime in the Philippines uses social media groups to suppress
domestic political opponents.ii LTG H.R. McMaster, our
National Security Advisor, noted just last week the Kremlin is again
using its cyber influence just across our southern border seeking to
push their preferred party and politicians to the forefront in
Mexico.iii
---------------------------------------------------------------------------
\i\ Hannah Beech. ``Across Myanmar, Denial of Ethnic Cleansing and
Loathing of Rohingya.'' New York Times. 24 October 2017. Available at:
https://www.nytimes.com/2017/10/24/world/asia/myanmar-rohingya-ethnic-
cleansing.html?_r=0.
\ii\ Lauren Etter. ``What happens when the government uses Facebook
as a weapon?'' Bloomberg. 7 December 2017 Available at: https://
www.bloomberg.com/news/features/2017-12-07/how-rodrigo-duterte-turned-
facebook-into-a-weapon-with-a-little-help-from-facebook.
\iii\ David Alire Garcia and Noe Torres. ``Russia meddling in
Mexican election: White House aide McMaster.'' Reuters. 7 January 2018.
Available at: https://www.reuters.com/article/us-mexico-russia-usa/
russia-meddling-in-mexican-election-white-house-aide-mcmaster-
idUSKBN1EW0UD.
---------------------------------------------------------------------------
Social media, at its height, offered a platform for discussion
across diverse audiences and led to uprisings usurping dictators during
the Arab Spring. But bad actors with motivation, money, manpower and
know--how will always come to these information gateways to pursue
their objectives. Criminals, terrorists and authoritarians see the
Internet and social media as a cost effective open doorway into the
very heart of their adversaries. Authoritarians worldwide now recognize
the power of the Kremlin's social media manipulation, and if left
unchecked, will copy and deploy Russia's playbook against their
enemies. Lesser--educated populations around the world predominately
arriving in cyberspace via mobile phones will be particularly
vulnerable to the social media manipulation of terrorists and
authoritarians.
American focus on the Islamic State's social media recruitment or
Russian meddling in the 2016 Presidential election overlooks other
indicators of damaging activity. American companies have suffered and
remain particularly vulnerable to smear campaigns launched by foreign
state actors through malicious, false narratives pushed by bogus social
media personas. These campaigns can cause serious reputational damage
sending stock prices plummeting and decreasing sales.
Beyond just smear campaigns and character assassination, this
committee should take seriously the ability of foreign nations to
mobilize violence inside the U.S. through an evolution I would call
``Anwar Awlaki meets PizzaGate''. Just a few years ago, Anwar al--
Awlaki, al Qaeda in the Arabian Peninsula's leader of external
operations, recognized the power of the Internet to recruit and
mobilize terrorists in America to conduct violence in the U.S.
homeland. The Islamic State took this to another level with their
spokesman abu Muhammad al--Adnani calling on supporters to conduct
attacks at homeiv and then further enabling e--recruits by
using a social media battalion to guide plots remotely--connecting
with, coaching and directing terrorists in the West to specific
targets.v A little over a year ago, America saw an
individual consume a false conspiracy on the Internet and social media,
known as PizzaGate, and then travel to Washington DC to investigate
these bogus claims. He arrived at a falsely implicated restaurant and
discharged a weapon before being arrested.vi
---------------------------------------------------------------------------
\iv\ Bulos, Nabih. ``Islamic State's taunting speech calls for
killing civilians.'' LA Times. 22 September 2014. Available at: http://
beta.latimes.com/world/middleeast/la-fg-islamic-state-taunts-20140922-
story.html.
\v\ Rukmini Callimachi, ``Not `Lone Wolves' After All: How ISIS
Guides World's Terror Plots From Afar.'' New York Times. 4 Feb 2017.
Available at: https://www.nytimes.com/2017/02/04/world/asia/isis-
messaging-app-terror-plot.html?_r=0.
\vi\ Grace Hauck. ``Pizzagate shooter sentenced to 4 years in
prison.'' CNN. 22 June 2017. Available at: http://www.cnn.com/2017/06/
22/politics/pizzagate-sentencing/index.html.
---------------------------------------------------------------------------
Surely a foreign adversary of the United States sees an opportunity
in combining these two scenarios. The greatest concern moving forward
might likely be a foreign intelligence service, posing as Americans on
social media, infiltrating one or both political extremes in the U.S.
and then recruiting unwitting Americans to undertake violence against a
target of the foreign power's choosing. Social media companies will be
better positioned to stop this potential scenario from occurring than
U.S. intelligence or homeland security that are blind to the technical
signatures behind this manipulation.
The U.S. Government's response to terrorist social media use has
been sustained and significant, and their response to state sponsored
influence on Americans disjointed and perplexing. In both cases,
government officials have pointed to social media companies asking why
they would allow their platforms to be used for nefarious purposes.
Social media companies realize the damage of these bad actors far
too late. They race to implement policies to prevent the last
information attack, but have yet to anticipate the next abuse of their
social media platforms by emerging threats seeking to do bad things to
good people. In previous testimony to the Senate Homeland
Security,vii Intelligence,viii Armed Services
ix and Judiciaryx committees, I've offered a
range of recommendations for how to counter bad actors using social
media in the pursuit of violence and nefarious influence. Today, I'll
focus and reiterate a few of these recommendations.
---------------------------------------------------------------------------
\vii\ Clint Watts. ``Terror in Europe: Safeguarding U.S. Citizens
at Home and Abroad.'' Statement prepared for the Senate Committee on
Homeland Security and Government Affairs, 5 April 2016. Available at:
https://www.fpri.org/article/2016/04/terror-europe-safeguarding-u-s-
citizens-home-abroad/
\viii\ Clint Watts. ``Disinformation: A Primer In Russian Active
Measures and Influence Campaigns.'' Statement prepared for the Senate
Select Committee on Intelligence, 30 March 2017. Available at: https://
www.intelligence.senate.gov/sites/default/files/documents/os-cwatts-
033017.pdf.
\ix\ Clint Watts. ``Cyber-enabled Information Operations.''
Statement prepared for the Senate Committee on the Armed Services,
Subcommittee on Cybersecurity, 27 April 2017. Available at: https://
www.armed-services.senate.gov/download/watts_04-27-17
\x\ Clint Watts. ``Extremist Content and Russian Disinformation
Online: Working with Tech to Find Solutions.'' Statement prepared for
the Senate Committee on the Judiciary, Subcommittee on Crime and
Terrorism. Available at: https://www.judiciary.senate.gov/download/10-
31-17-watts-testimony.
---------------------------------------------------------------------------
The first and most pressing challenge comes in the debate over
social media account anonymity and authenticity. Anonymity of social
media accounts has in many cases allowed the oppressed and the
downtrodden to speak out about injustice. It's given the weak a voice
against the strong, powerful, and corrupt. But over time, anonymity has
empowered hackers, extremists and authoritarians to inflict harm on the
public. Under the veil of anonymity, they spread hate, recruit members
and advance divisions in American society.
All people, real humans and their virtual personas, have the right
to free speech, but this right to free speech does not permit them to
endanger society. Account anonymity today allows nefarious social media
personas to shout the online equivalent of ``fire'' in a movie theater.
Bad actors and their fictitious and/or anonymous social media accounts
can and have created a threat to public safety. This is not protected
free speech and many social media companies offer no method to hold
these anonymous personas accountable.
Social media companies can and should protect the public anonymity
of account holders if the user chooses, but they must be able to
determine a real, authentic person resides behind each persona
accountable for their actions on the platform. Some social media
companies have advanced better methods to certify account authenticity.
However, the current level of authenticity on the Twitter platform is
sub-optimal. I'd encourage Twitter to rapidly expand its verification
to as many users as possible, as quickly as possible.
Closely connected to the issue of account authenticity is the rise
of computational propaganda. The negative effects of social bots far
outweigh any benefits. The anonymous, replication of accounts that
routinely broadcast high volumes of misinformation can pose a serious
risk to public safety and when employed by authoritarians a direct
threat to democracy. Social bots should be ceased immediately. For non-
automated accounts, reasonable limits on the number of posts any
account can make during an hour, day or week should be developed. Even
further, human verification systems (CAPTCHA) should be employed by all
social media companies to reduce automated broadcasting.
Federal laws governing attribution of political ads and
solicitations in television, radio and print should immediately be
extended to social media advertising conducted by political campaigns
and political action committees. Social media political advertising
will continue to grow in every election cycle and U.S. citizens must
know the source of the information they consume in any medium--print,
radio, television or social media.
Social media companies continue to get beat in part because they
rely too heavily on technologists and technical detection to catch bad
actors. Artificial intelligence and machine learning will greatly
assist in cleaning up nefarious activity, but will for the near future,
fail to detect that which hasn't been seen before. Threat intelligence
proactively anticipating how bad actors will use social media platforms
to advance their cause must be used to generate behavioral indicators
that inform technical detection. Those that understand the intentions
and actions of criminals, terrorists and authoritarians must work
alongside technologists to sustain the integrity of social media
platforms. Some social media companies have already moved in this
direction.
I'd note it's unreasonable to think that every social media company
can and should hire threat analysts for every possible emerging threat.
But a variety of rapid outreach approaches with external social media
analysts and threat experts positioned outside social media companies
could easily be developed or even be collectively sponsored by social
media companies. Several models from counterterrorism and cybersecurity
could be adopted by Silicon Valley in this regard.
I've made many other recommendations in the past but will close for
now and can elaborate further on them during the question and answer
session. In conclusion, some social media companies have done more than
others to improve the safety and integrity of their platforms. Others
have a lot of work to do to improve their platforms against
bad actors. Ultimately, the American consumer will decide whether
the benefits of using these services outweigh the risks. Many are
walking away from social media applications because they can't trust
the information being shared or tolerate the vitriolic user experience.
Social media companies should move aggressively to thwart terrorists
and authoritarians exploiting their systems not only because its what's
best for their users and society, but because it's good for business as
well.
The Chairman. Thank you, Mr. Watts.
Senator Nelson. Mr. Chairman.
The Chairman. I recognize Senator Nelson for an
introduction.
Senator Nelson. We have a new member of the Committee.
The Chairman. We do, indeed.
Senator Nelson. Senator Jon Tester of Montana. Senator
Tester has been wanting to get on this Committee for quite a
while, and we are so happy that he finally was able to be
appointed to the Committee. He brings a wealth of experience as
a senior member of the Appropriations Committee to this
committee.
So, welcome, Senator Tester.
The Chairman. Senator Tester.
Senator Tester. Mr. Chairman and Ranking Member, I just
very quickly--first of all, thanks for the welcome. I look
forward to working on this Committee, although I will tell you
I have flashbacks to 2007 right now. Thank you.
The Chairman. Well, I also want to thank and welcome
Senator Tester for joining the Committee. It's always nice to
have another rural Senator on this Committee, somebody who is
my neighbor and represents a state like mine where there are
more cattle than there are people, and, obviously, an area
where there are still cell phone and broadband free areas. But
we're hoping to change that. Senator Tester can probably
remember the days like I can when there were party phone lines.
So we've come a long way, but we've got a long way to go, and a
lot of the issues that we address on this committee are issues
that impact the daily lives of people in his state just like
they do with so many members of this Committee.
Senator Tester, welcome. It's good to have you here.
We're going to start with some questions, and I want to
direct this to Mr. Monje and Ms. Bickert. As you mentioned,
both of you, I think, in your testimony, Google, Facebook,
Twitter, and Microsoft announced the formation of a hashing
coalition to better identify and ultimately remove offending
content. The question is: Is there any shared standard for what
constitutes extremist or terrorist content in your coalition?
Mr. Monje. Thank you, Chairman, for that question. Our
companies are constantly working with one another and with
civil society and with smaller companies to address the issues
that change and evolve and new things that we see around the
world. We are constantly adapting how we attack the challenge,
and we do rely on the advice and good counsel not only of our
peer companies but also of academics and NGOs.
The Chairman. There's no standard definition, though, that
you've agreed upon?
Ms. Bickert. That's right, Mr. Chairman. I would just add
that the companies--we launched the Global Internet Forum in
June 2017, but we've actually been working together for a
number of years informally. Part of those meetings is
discussing what the appropriate standards are, recognizing, of
course, that these different products work differently.
But the two types of policies I think you most commonly see
are first directed toward the groups having any presence on the
platform. So, for instance, at Facebook, if you are Boko Haram,
you cannot have a page on Facebook even if you're just talking
about the lovely weather. You simply can't be on the platform.
And then the other types of policies that you see certainly
across the major companies is banning any praise or support of
these groups or their actions.
The Chairman. Ms. Downs, according to the Counter Extremist
Project, one single bomb-making video used to instruct the
Manchester suicide bomber has been uploaded to YouTube and
removed 11 times but continues to resurface as recently as this
month. How is it possible for that to happen? Why aren't your
hashing efforts working to keep this video off your platform
permanently?
Ms. Downs. Thank you, Chairman. As I mentioned, we have
strict policies against terrorist content, including videos
that instruct people on how to commit terrorist attacks. This
certainly includes bomb-making videos, videos instructing
people on how to drive vehicles into crowds, and so on. This
particular video was caught by our systems. We have used it in
our hash-sharing database, and we are catching re-uploads of
the video quickly and removing it as soon as those uploads are
detected.
The Chairman. Are your companies--and anybody can answer
this. But are your companies, as you start to roll out some of
these new counterterrorism programs--how do you have ways of
measuring their effectiveness? What is sort of the metric or
the standard?
Mr. Monje. Chairman, at Twitter, we've really doubled down
on the technology, on the machine learning, to try to identify
and remove content as quickly as possible. So our metric is how
many accounts are we taking down, how many accounts are we
identifying proactively, and how many are we able to take down
before they're able to tweet. And we've seen steady progress in
that. We started--we were taking down about a third of our
content proactively with our machine learning. Today, that's
north of 90 percent, with 75 percent of that coming down before
anybody gets a chance to tweet. So that's how we--that's our
main metric.
The Chairman. So it's been reported that ISIS surrogates
are using AI bots to carry out recruiting and messaging
campaigns, and as you all become more sophisticated in how to
prevent and root out, the bad people also become more
sophisticated in how to get around, and the threat evolves. So
are you seeing that level of sophistication, and, if so, what
are you doing to mitigate it--the use of AI against you by
these groups?
Does anybody want to take a stab at that?
Ms. Downs. In addition to our policies against terrorist
content, we have very aggressive and proprietary spam detection
systems which would catch massive re-uploads of AI-generated
videos. So our long history in fighting spam across our
services is an effective technique to get at that behavior.
The Chairman. Anybody else?
Mr. Monje. I would just agree with you, Chairman, that it
is a cat and mouse game, and we are constantly evolving to meet
the challenge. When we, often in the past, would ban an
account, suspend an account, they would try to come back and
then brag about the fact that they were banned. That became a
very strong signal for us which resulted in them being taken
down even quicker. So they've stopped doing that.
The Chairman. My time has expired.
Senator Nelson.
Senator Nelson. Mr. Watts, I'd like you to take my time and
inform the Committee with your expertise what the Russians--for
that matter, anybody else--can do to us in this coming
election?
Mr. Watts. Thank you, Senator. I think I would start off
with--there has been no response from the U.S. Government with
regards to Russian influencing campaigns with social media. So,
therefore, they have stayed on course with their operations.
During non-election years, they tend to focus on social issues
and what I would say is audience infiltration. So any
organization, entity, social media group that they really want
to be able to move or influence later, they begin infiltrating
that by just sharing the same content back with that audience
and trying to develop their own content within it.
Beyond just the United States and this Presidential
election, I think we should look at all elections worldwide.
They've realized that this playbook works very well. It's
extremely cost-effective, and there has been almost no
downside, at least to this point, of doing it. So you've seen
it in Europe, where they continue to seed audience bases.
Anywhere that they can break up a union--so the European
Union or NATO--they will continue to seed in those populations.
So I would tell you to look at Catalonia or even Scotland,
places like that where they see an opportunity to break up an
alliance, to create divisions within a democracy--they are
moving there. I think Lieutenant General McMaster last week
pointed to Mexico as another example of where they've seen some
sort of audience infiltration.
The key trigger I always look for is hacking. When they
launch widespread hacking against a target, they are making a
strategic decision to go after an objective, and that's one
thing I would tell everyone to look for on the horizon.
Beyond that, if you want to know where the Russians are
going with their influence, you should always look at where
they are putting up new state-sponsored outlets. To infiltrate
an audience, you have to have a base of content to launch your
campaign. So when they add an additional language or for their
wire service--let's say Sputnik News--or an RT YouTube channel,
that is an audience that they're going to reach for. And I will
tell you right now, they are looking very heavily into Latin
America. I think they would like to build a capability more in
the Middle East moving forward.
Beyond just Russia, they will focus on social issues to win
over audiences during non-election years to then be able to
pivot them toward whichever candidate or party they want to
support moving forward. The goal isn't one party or the other
and their victory. The goal is to create divisions inside the
United States, and that will be their predominant focus moving
forward, further polarizing the information landscape.
I would also note that everyone is adopting this technique.
You see it in Myanmar. You see it in the Philippines. Any low-
level educated population around the world that's on social
media, particularly through mobile applications, is highly
vulnerable to this. They have not built up the ability to
assess information or how to avoid being influenced, and so
they're highly vulnerable to this influence technique.
And, last, I would say it's political campaigns and the
companies that are going to be hired. If there's not some sort
of regulation put around ads in social media, every political
campaign, whether it's in the U.S. or around the world, will
have to use a dark social media campaign through either Super
PACs or candidates to keep up with their competitors, and it
will further--it will not only harm the societies in which it's
in, but it will actually harm the social media companies and
their platforms. They will actually make the environment so
terrible and so polarized, as we've seen over the past few
years, that it will create just a nasty sense for democracy.
If you want to look at how this has effected in Russia,
Russia did this to their own people first before they came, you
know, across the ocean. It creates widespread apathy in
democracies. It dilutes the line between fact and fiction, and
when that happens, you actually cannot keep a democracy moving
forward. I think that's what's most dangerous about this entire
system, is it's agnostic of party or candidate. Ultimately,
it's about breaking our system and turning it against each
other.
Senator Nelson. So when you see them dive deep into the
instrumentalities of government, such as the example that I
gave, that there were half a million comments on the recent FCC
rule, and when you see that--you read the public press that
they're in 20 states' elections divisions, sketch out what are
some of the dastardly things that they could do to undermine
America.
Mr. Watts. The one big thing that they would try and do is
an information attack on the integrity of a democratic
institution. That's really played out in both of those
scenarios. With the FCC, it's ``you can't trust the FCC. We
need to get rid of these regulatory bodies. You can't trust
them. They're trying to mind control you.''
The other part is the elections, the integrity of an
election. The second campaign they launched in the run-up to
2016 was voter fraud, election rigged. They didn't really care
what candidate won. They wanted the American people to think
that their vote did not count. The hacking campaign against
voter databases--it was to sow doubt such that when you see the
narrative of voter fraud, election rigged, you might think,
``Oh, maybe my candidate didn't really get elected because my
vote didn't count.''
So it's about destroying democratic institutions and
confidence in the U.S. Government or democratic institutions to
govern properly, that the system is always rigged and you can't
trust anyone. That's really the focal point of all of those
efforts the Russians might run or any authoritarian regime that
wants to run a campaign against the U.S. Government.
Senator Nelson. Thank you.
The Chairman. Thank you, Senator Nelson.
Senator Wicker.
STATEMENT OF HON. ROGER WICKER,
U.S. SENATOR FROM MISSISSIPPI
Senator Wicker. So it wasn't so much an attempt to get one
candidate elected over the other. It was knowing there was
going to be loser and that relatively half of the population
who supported that loser would think their vote hasn't counted.
Mr. Watts. That was their second campaign. They ran four
narratives during the first one which were specific to the
candidates. Then in October 2016, they really shifted to the
integrity of democratic institutions. So it was twofold. They
were running to try and get a candidate they wanted up to
election day, and then beyond election day, it was to create
mass chaos inside the United States.
Senator Wicker. Thank you for clearing that up. Let me move
to Twitter.
Is it Monje? Am I pronouncing it correctly?
Mr. Monje. Thank you for asking, sir. It's Monje.
Senator Wicker. Monje. OK. Good. Well, let me ask you,
then--and is that a Cajun name? You told me you're from New
Orleans.
Mr. Monje. I am from New Orleans, sir. I'm a Saints fan.
But my family is from Argentina.
Senator Wicker. Very interesting. Let me ask you about
aggregate user data. There are data analytics companies who
purchase aggregate user data from all of you, Twitter. Is that
correct?
Mr. Monje. Yes, sir.
Senator Wicker. So, for example, if I am an analytic
company, and I want to work for the NFL, for example, I would
purchase aggregate user data from Twitter and, using keywords,
develop information that might be helpful to the National
Football League.
Mr. Monje. It depends on what it is they plan to use it
for. A lot of times, what our data is used most often for is to
target advertising.
Senator Wicker. To target advertising. OK. Let me ask you
this. Is that same ability to purchase aggregate data available
to Federal law enforcement? Is it available to Federal anti-
terrorism agencies?
Mr. Monje. It depends on what the purpose of the use of
data is, and we have rules about how any entity, regardless of
whether governmental or not, anywhere in the world, can use our
data.
Senator Wicker. And what are those rules with regard to
terrorism?
Mr. Monje. With regards to terrorism. I'd have to get back
to you on the exact rule--on the exact language of that, sir.
Senator Wicker. OK. Well, because this is pretty important.
If a data analytics company wants to purchase data from
Twitter, you're willing to sell that to them. What I want to
know is if that company is going to supply information to
agencies that are seeking information about terrorist
activities and that activity is part of this aggregate user
data. Will you sell that data to them? Because, frankly, I'm
informed that you will not do so.
Mr. Monje. Well, let me tell you a little bit about what we
do with our data, sir, which is we--on our side, on the Twitter
side of the equation--are very data-focused and use that data
to inform the machines that help us fight the terrorists. We
work on a daily basis with law enforcement.
Senator Wicker. That's within the Twitter organization.
Mr. Monje. Within the Twitter organization. We work on a
daily basis with law enforcement, particularly with the FBI,
and will respond to any request that they have, as long as they
give us the proper legal process, and we are on a first name
basis with our counterparts at the FBI.
Senator Wicker. And what would that proper legal process
be?
Mr. Monje. It depends on what they're looking for. It could
be a warrant--depending on whether they're looking for private
or non-private information on--where they're looking for direct
messages.
Senator Wicker. Has Twitter told these data analytic
companies that the purchases of this data cannot be used for
counterterrorism purposes?
Mr. Monje. They cannot be used for persistent surveillance
of individuals.
Senator Wicker. They can be used to target advertising and
products and sales, but they cannot be used to help our anti-
terrorism agencies. Is that correct?
Mr. Monje. We do help our anti-terror agencies,
particularly the Federal Bureau of Investigation on a daily
basis. And in response----
Senator Wicker. But if a third----
Mr. Monje. I'm sorry, sir. After you.
Senator Wicker. Go ahead, please.
Mr. Monje. No and to respond to their request, we have a
very fast system that is an input--any time they have
information to us, we turn it around as quickly as we can,
within hours. We do not allow persistent surveillance of our
users. We protect the privacy of our users.
Senator Wicker. You protect the privacy of your users, even
if a Federal agency wants to surveil that public information
for anti-terrorism purposes?
Mr. Monje. If an agency comes to us with the right process,
and it's according to Federal law, the ECPA----
Senator Wicker. But that's not what I'm talking about. I'm
talking about an independent data analytics company.
Mr. Monje. Yes, sir. So----
Senator Wicker. You will sell that data to them, but you
tell that company they can't use it for anti-terrorism
purposes. Is that correct?
Mr. Monje. We're not going to allow any company, whether
they're selling cars or cereal or anything, the NFL, to
persistently figure out where somebody is in a given time. But
we do have news products, data alerts, for law enforcement, for
the FBI, that they use.
Senator Wicker. Ms. Bickert, is that the policy of
Facebook?
Ms. Bickert. Thank you, Senator. We don't sell user data
outside the context of allowing people to target audiences in
their advertisements. That is a capability that is equally
available to law enforcement, as it would be to anybody else.
Law enforcement can provide us--if they want to find out
specifics about an individual user, they can provide us with
legal process, and we will respond.
Senator Wicker. What is the privacy concern that supersedes
the need to surveil terrorist organizations that participate in
Facebook?
Ms. Bickert. Senator, we absolutely respond to valid law
enforcement requests. If it's part of an investigation, and
they give us that process, we do respond.
Senator Wicker. Thank you, Mr. Chairman.
The Chairman. Thank you, Senator Wicker.
Senator Klobuchar.
STATEMENT OF HON. AMY KLOBUCHAR,
U.S. SENATOR FROM MINNESOTA
Senator Klobuchar. Thank you very much, Mr. Chairman.
I came in, and three senators told me, Mr. Monje, that you
were a Saints fan. Is that correct?
[Laughter.]
Senator Klobuchar. I would just like to note my scarf and
who won the game.
Who won the game, Mr. Monje?
Mr. Monje. It was an excellent game, and it was a
spectacular ending for the Vikings.
Senator Klobuchar. Thank you very much. Now, let's get to
some serious matters.
So we've had a hearing focused on the election piece of the
internet, and I note that, while I want to get to some
questions about terrorism as well, there are many ways we can
undermine our country and undermine our democracy. One is
obvious, with violent attacks. Another is if Americans aren't
able to make their own decisions about who they were voting for
because they get false information. So that's why Senators
Warner, McCain, and I have introduced the Honest Ads Act. We
had an entire hearing about this over in Judiciary.
But I would just start with you, Mr. Watts. As you know,
right now, there are disclosure rules so radio, TV, print--they
all have to keep on file ads of national and political
importance, legislative importance, as well as candidate ads,
so that opposing campaigns' press can see these ads, as well as
disclaimer requirements. Do you think those should apply to
social media ads, paid ads as well?
Mr. Watts. Absolutely. If it does not happen, I mean, both
from society and social media's perspective, the conversation
will continue to get more polarized and more negative, and
people won't be able to trust information on the platform,
regardless. So I think it's essential that the ad regulation
extend to social media because that's where all advertising is
going in the future.
Senator Klobuchar. Exactly. We had $1.4 billion in the last
election, and there are projections it's going to go to $3
billion or $4 billion and things like that, and there are
literally no rules. We do appreciate that a number of the
companies, including the ones here, have stepped up to start
putting in place some of their own guidelines and changes. But
I do believe that this won't work unless we have guidelines
like we have for media. Do you agree with that?
Mr. Watts. I do. If we don't, it can have a very
devastating effect and force all political campaigns
essentially to try to do social media manipulation that's maybe
not entirely authentic.
Senator Klobuchar. Thank you. Terrorist online recruiting--
my state has had its share of that recruiting, especially
related--some from ISIS, but in past years, al Shabaab. We've
had dozens of prosecutions out of our U.S. Attorney's office,
successful ones, where people have actually been recruited to
go overseas to fight on behalf of terrorist groups.
What kind of recruiting activity are you able to detect on
your platforms, and what can you tell us about the trends? How
are they changing their strategies? I remember the FBI showing
me the ads targeted at Minnesota with literally airplane
tickets from Minnesota to Somalia for terrorists. So tell me
what you're doing now and what you see in terms of recruiting
and what you can do about it.
Anyone?
[No verbal response.]
Senator Klobuchar. OK. Should I call on people?
Mr. Watts. Well, I'm sure they don't want to answer as much
as I do, so I'll go first.
[Laughter.]
Mr. Watts. What I would say is that what we should note is
that these social media companies here were the forerunners,
but they're also the dinosaurs of the social media era, meaning
that they're the largest platforms and they have the greatest
capability to actually deter this activity. But in the future,
if I were a terrorist or an extremist group trying to mobilize,
I would go to the smaller social media applications that have
the greatest encryption, the largest dissemination capability,
and I would focus there and then move to other social media
platforms, because there would be less ability for them to
deter my activity on the platform.
With that, in terms of the extremists, I think you need to
look at what are the social media applications essentially
being used by language--language is the key for actually doing
recruitment--and where are the populations in each of your
states and cities that are refugee populations, immigrant
populations, and then how does that sort of play out, and who
are they interfacing with overseas.
Senator Klobuchar. One last question here. Throughout the
2016 cycle, Russians worked to influence the U.S. electorate,
as I mentioned, and part of it was they did it by searching
algorithms to promote misinformation. In the current news era,
information is consumed rapidly, and algorithms play a
significant role in deciding what content consumers see.
Mr. Monje and Ms. Bickert, what are Twitter and Facebook
doing to help ensure the information appearing in search
results and on consumers' feeds is free from that kind of
exploitation?
Mr. Monje. Thank you very much, Senator, for that question,
and we do quite a bit to protect our search, in particular.
More than 95 percent of our users as a default setting have
safe search as part of their experience on Twitter. So what we
do is when we identify a bot, malicious automation, which is a
lot of the ways that this kind of information has promulgated
on the internet, is that is severely down ranked so it's very
hard to find.
Senator Klobuchar. Ms. Bickert?
Ms. Bickert. Thank you, Senator. We are increasingly
finding new ways to disrupt false news and help people connect
with authentic news. We know that's what they want to do. We're
also investing in efforts to help people distinguish between
the two, which includes basic education and public outreach.
As far as disrupting the false news, oftentimes--because we
have a requirement that people have to use Facebook with their
authentic name--if we can identify inauthentic accounts--and
we're getting much better at that--we can remove those accounts
and the false news goes away. The majority of the actors that
we see trying to spread disinformation are financially
motivated. So that goes a long way.
We're also working with our community to flag false news,
send it to third-party fact checkers, and make that content
less visible and put it in context. So now, if you come to
Facebook and you see a story in your news feed that is an
article that has been flagged as potentially false by our
community, we will also show you some related articles
underneath it so that you have a sense of where this story sits
in the broader spectrum of news.
We're working with responsible publishers to make sure that
they know how to most effectively use social media, and then
we're also working on user education campaigns.
Senator Klobuchar. Thank you very much.
The Chairman. Thank you, Senator Klobuchar.
Senator Moran.
STATEMENT OF HON. JERRY MORAN,
U.S. SENATOR FROM KANSAS
Senator Moran. Mr. Chairman, thank you. Thank you to you
and the Ranking Member for conducting this hearing. I think
it's one of the most interesting and potentially valuable
hearings we will have had, and what a great development it
would be if we could reduce the military necessity and the loss
of life that comes from military action in fighting terror if
we can keep it from occurring in the first place.
So thank you for being here. Thank you for your testimony.
Let me ask this. Some of you covered in your testimony
collaborative efforts among multiple businesses and groups that
involved a shared industry database, which eventually led to
the formulation of the Global Internet Forum to Counter
Terrorism. I want to know more about that collaboration.
Part of the reason for that question is that my guess is
that as larger social media companies become more innovative
and effective in what you're attempting to accomplish,
preventing terrorism, it would seem to me that other smaller
platforms may become the platform of choice in this space. So
if you're successful in your efforts, what prevents terrorists
from moving to a different platform, and, therefore, what's
those smaller platforms' engagement in what you're doing? It's
directed at anyone who desires to answer. Or maybe if no one
does----
Ms. Bickert. Thank you, Senator. That is exactly what we
were thinking as the large companies, was that we needed to
make sure that this movement was industry-wide. With that in
mind, we reached out to a number of small companies several
years ago. I think we reached out to 18 companies initially.
All 18 said yes, they wanted to meet to talk about best
practices to counter terrorism.
We then met for more than a year before we ultimately
launched the Global Internet Forum. Through that forum, which
we launched in June, we've since had five international working
group sessions with 68 smaller companies based around the
world, and this is an opportunity for us to share expertise and
learnings from the larger companies.
Senator Moran. And let me take that a step further. So what
are the smaller companies, smaller platforms, doing? They're a
participant in this collaboration? They're doing something
similar to what you're telling us that your companies are doing
today?
Ms. Bickert. Yes, Senator. Often, they are learning from
what we are experiencing as the larger platforms in terms of
the conduct that we see from bad actors, the policies we've put
in place, and how we're thinking about using technology and
people to combat those threats.
Senator Moran. Anyone else?
Mr. Monje. I would just add that, you know, we've been
extremely successful at taking terrorist content off of
Twitter. It's a tremendous success for Twitter, but it doesn't
eliminate the terrorists and them moving to other platforms
like Telegram. It doesn't help everybody. You know, Twitter is
a smaller company among the giants, and so we often--because
we've had to be creative and innovative in our use of
technology--can help be a bridge to the smaller companies and
tell them you can make significant progress. You just have to
invest in the technology.
Senator Moran. What evidence do you see that terrorist
organizations are changing their behavior as a result of what
you're doing?
Mr. Watts. There is open--you know, in some of their forums
right now--they're trying to find a platform where they can go
in a secure fashion, communicate, and push their propaganda
around the world. So they're actively seeking new platforms.
And I think your question is a great one, which is how do
we help these small companies that are developing new social
media applications, who don't have the capabilities in terms of
security, ward this off, and I don't think there's a good
answer for that question. But they are seeking a new home. They
just haven't found it yet.
Senator Moran. Mr. Watts, is the response by terrorist
organizations to seek a new home, or are they finding ways to
hinder your efforts, or both?
Mr. Watts. Both. They're looking for a place where they can
communicate and organize, but they have to be able to push
their propaganda globally in order to recruit and gain
resources, so they need some way to do that. They will continue
to try and exploit these small applications, but it's tougher
for them on these small applications because, globally, there
are not as many people on them.
So it's a better problem to have than what we've had in the
past, but it really begs the question that, ultimately, one of
these social media platforms that's popular overseas will start
to gain traction with them, either due to its encryption
capabilities or how they can connect with audience or how they
can load and share videos. I think this is important across all
extremist groups. If you look at some of the platforms that are
out there, Reddit, 4Chan, these anonymous platforms, they also
can be great tools, and it would be great to see them
integrated with the bigger companies that have way more
capability to detect that activity.
Senator Moran. Thank you very much.
Thank you, Mr. Chairman.
The Chairman. Thank you, Senator Moran.
Senator Schatz.
STATEMENT OF HON. BRIAN SCHATZ,
U.S. SENATOR FROM HAWAII
Senator Schatz. Thank you, Mr. Chairman.
Thank you all for being here. Context--Facebook had $10.3
billion in revenue last quarter, about 23,000 employees, if I'm
not mistaken; Twitter, $590 million in revenue last quarter,
3,700 employees. So my question for Facebook and Twitter is: In
dollars, as a percentage of revenue, however you want to
calculate it, and in terms of employee count, both part-time
and full time, how many people, how many dollars are you
devoting to this problem?
Ms. Bickert?
Ms. Bickert. Thank you, Senator. This has been a
significant area of investment for us, not just now, but over
the past years. But I do want to point to a recent announcement
from our CEO, Mark Zuckerberg, after we released earnings last
quarter where he specifically pointed to the fact that as we
invest in safety and security, people should expect to see
Facebook's costs go up. That's reflected in the fact that we
now have more than 7,500 people who are working to review
terror content and other potential violations.
We have 180 people who are focused specifically on
countering terrorism. So these are people like the former
academics, like Brian Fishman, formerly with the West Point
Counterterrorism Research Center, and others.
Senator Schatz. So 180 full time; 7,500, it's part of their
job?
Ms. Bickert. Seventy-five hundred are content reviewers. In
the area of safety and security, more generally, we have 10,000
people currently. We are looking to be at 20,000 by the end of
the year.
Senator Schatz. Mr. Monje?
Mr. Monje. It is fewer than that. But I can tell you that
our entire engineering, product, and design team at various
stages are all working on this. We're a small team. We have to
be supple. We have to be able to shift as the challenges move.
The numbers that are really important also to look at are 2
billion users, 400 hours of video every minute, and, for us,
350,000 tweets every minute.
This isn't--in order to make progress on this issue, you do
need to have humans, and we have former law enforcement; we
have experts; we partner with contractors, consultants,
academics----
Senator Schatz. I want to give you an opportunity to set
the record straight about fake accounts. I've been reading a
lot about this. I saw anywhere from 9 percent to 15 percent
fake. I saw another USC study that said it's actually 48
million out of your nearly 300 million users. What's the
number? How many fake accounts do you have?
Mr. Monje. We believe that fewer than 5 percent of the
accounts on Twitter are fake accounts.
Senator Schatz. Now, if you've kind of zeroed in on--let's
stipulate that it's 5 percent of almost 300 million, right? If
you know they're fake, what's the issue here?
Mr. Monje. We are--they keep coming back, and they try
different methods to get back on the radar screen, actually,
and so we are, as a matter of course, consistently fighting
malicious automation. We are now challenging 4 million
malicious automated accounts a week. That means we are
essentially sending them a note saying, ``You're acting weird.
Can you verify you're a human being?'' That's double where we
were last year.
Senator Schatz. Can I just talk to you about bots a little
bit? I know this is a hearing about terrorism, primarily, but I
think it's worth asking what we're doing about active measures.
You know, there was public reporting that the Roy Moore
campaign went from 27,000 to 47,000 Twitter followers over the
weekend, and a substantial portion of those appeared to be
located in Russia. We had the take-a-knee thing where, clearly,
there was an active measure to try to just sow discord. In
other words, you've got bots and bot farms out there that are
taking both sides of the argument.
So when we get into a conversation about active measures
against our country, I don't think we should think of it as
active measures against Democrats, and I don't think we should
assume that it's just Russian active measures. We have to think
of this as undermining democracy itself and undermining our
ability to have our First Amendment rights exercised in any way
that's meaningful.
So my question for you is--I mean, this is relatively
recent, and it doesn't seem to--you can give us the measure of
your activities, you know. Four million accounts are being
challenged, and 500,000 accounts have been taken down. But
based on results, you're not where you need to be for us to be
reassured that you're securing our democracy. To the degree and
extent that elected officials and people who vote and our
adversaries are participating in your platform, how can we know
that you're going to get this right and before the midterms?
Mr. Monje. Yes, sir. Thank you for that question, and
that's exactly the question that we ask ourselves every day. We
think we're better prepared for this election than we've ever
been. We are continually improving our tools, and we're going
to get better, and we're going to report to the American people
the results of our efforts.
Senator Schatz. Thank you.
The Chairman. Thank you, Senator Schatz.
Senator Young's not here.
Senator Markey.
STATEMENT OF HON. EDWARD MARKEY,
U.S. SENATOR FROM MASSACHUSETTS
Senator Markey. Thank you, Mr. Chairman, very much.
Last month, the FCC gutted the net neutrality rules that
protected the Internet as we know it, and as a result, the next
Facebook, the next YouTube, the next Twitter will struggle to
get off the ground. I strongly oppose that FCC decision, which
is why I plan to introduce a Congressional Review Act
Resolution of Disapproval, which will undo the FCC's recent
actions and restore the 2015 Open Internet Order. My resolution
enjoys the support of Democrats. Susan Collins, the Senator
from Maine, has indicated that she will vote for it.
My question to each company here is simple. Do you support
my CRA resolution which would put net neutrality back on the
books?
Mr. Monje.
Mr. Monje. Yes, sir. Thank you for your leadership on this
issue. It's an important issue for our company and for our
users.
Senator Markey. You would support it. Thank you.
Ms. Downs.
Ms. Downs. We support strong enforceable net neutrality
protections. We supported the 2015 rules, and we will support
any effort to put those rules back in place.
Senator Markey. Thank you.
Ms. Bickert.
Ms. Bickert. Thank you, Senator. Same answer. We will
support the CRA, and we also support and will work with anybody
who's interested in working to find a way to put those rules
back in place.
Senator Markey. Thank you. We thank each of you. Thank you
so much.
Next question. Bad actors can and do use the internet and
social media to acquire weapons, including firearms. That's why
in 2016, I wrote a letter to Facebook and Instagram asking why
gun sales continue to take place on their sites, even after
announcement of self-imposed policy changes aimed at
eliminating this type of activity. I was pleased when both
Facebook and Instagram announced they would prohibit individual
users from buying and selling firearms on their sites.
Yet recent media reports indicate that users are still able
to gain access to deadly weapons on social media. Just last
month, the Chicago Police Department arrested 50 people in a
case involving the sale of illegal guns in Facebook groups.
Ms. Bickert, it appears that gun sales on your platform may
have moved into private Facebook groups. How is Facebook
working to stop the sale of firearms in that corner of your
platform? Notably, the Chicago Police Department said it did
not receive cooperation from Facebook during its 10-month
investigation. Law enforcement officials reported that Facebook
hampered their investigation by shutting down accounts that
officers were using to infiltrate the group in question.
Ms. Bickert. Thank you, Senator. It's certainly an issue
that we take seriously, and as a former Federal prosecutor
based in Chicago, our relationship with law enforcement
authorities is very important to us. We have cooperated with
law enforcement and will continue to do so in that case.
We do not allow firearm sales. Enforcement has presented
challenges for us, and to get better, one of the things we're
doing is working on our technology. Anybody in the community
can report gun sales to us, and we will take action, and that's
important, and that does happen even in private groups. But we
know we need to do more, and that's why we're now using things
like image prediction technology to help us recognize when
those sales might be taking place.
Senator Markey. So since Instagram can turn into Instagun,
you know, for someone who intends on using it for nefarious
purposes, the answer that you would give to the Chicago Police
Department when it said it did not receive cooperation from
Facebook during its 10-month investigation is that you did
cooperate or that you have now established a policy of
cooperation with the Chicago Police Department and every police
department across the country?
Ms. Bickert. Thank you, Senator. I believe they clarified
their statement afterwards. We have been cooperative with them
from the beginning, and I would be happy to follow up
afterwards with you on that.
Senator Markey. That would be helpful. And in terms of the
private Facebook groups that this type of activity has migrated
to, you are saying as well that you are working to shut that
down as well?
Ms. Bickert. That's right, Senator. This is an area where
we recognize enforcement can be challenging, and we have to be
proactive in looking for solutions. So we're trying to make it
easy for people to report, but also going further to look for
this content.
Senator Markey. Thank you. And that's why this hearing is
so important.
I would thank you, Mr. Chairman, because the internet can
be used to spread hate, but it can also be used to spread
weapons of war into the hands of those who are the haters and
do enormous harm in all of our communities across the country.
So thank you.
We thank each of you for your testimony.
The Chairman. Thank you, Senator Markey.
I would just ask the three, too, that you all, I assume,
would support legislation that would put in place rules for an
open Internet as well. Would that be true?
Mr. Monje. Twitter has long been a supporter of net
neutrality, and, hopefully, Congress can develop good rules.
Ms. Downs. Same answer.
Ms. Bickert. Same answer, Mr. Chairman.
The Chairman. Very good.
Next up is Senator Baldwin.
STATEMENT OF HON. TAMMY BALDWIN,
U.S. SENATOR FROM WISCONSIN
Senator Baldwin. Thank you, Mr. Chairman and Ranking
Member, for this important hearing.
Much of the conversation today has been focused on
addressing foreign terrorist organizations' use of your
platforms as tools to recruit and radicalize individuals both
here and abroad. I'd like to turn to how you are addressing the
use of social media to further domestic extremism. Whether it's
the vehicular attack on counter-protestors in Charlottesville
this summer or the 2012 shooting at a Sikh temple in my home
state of Wisconsin, we've seen numerous individuals subscribing
to racist ideologies turning to violence.
Beyond that, there's a disturbing increase in hate crimes
in this country, as documented by FBI's limited collection of
data from state and local law enforcement. As with other forms
of extremism, social media is undoubtedly playing a role in
spreading these ideologies and channeling these individuals
into violent action. How are your companies working to address
the role of social media in furthering domestic extremism,
particularly white nationalist or white supremacist violence?
I'd like to start with you, Ms. Bickert.
Ms. Bickert. Thank you, Senator. I want to be clear that
our policies prohibit any group that is either a violent
organization--and that's regardless of ideology. So if it is a
domestic terror organization, if it's a foreign terror
organization, no matter what the ideological underpinning is,
they are not allowed on Facebook.
But we also prohibit hate organizations, and these are
groups that are propagating hate based on a protected
characteristic, like race, religion, gender, gender identity,
and so forth. The same consequences under our policies apply.
They're not allowed to be on our platform. People cannot praise
or support them.
Senator Baldwin. Ms. Downs.
Ms. Downs. Thank you, Senator. Our violent extremism
policies apply to violent extremism in all its forms, including
white supremacy and other forms of hatred, and we apply our
policies against incitement to violence and violent ideology
consistently across violent extremism in all its
manifestations.
Senator Baldwin. Mr. Monje.
Mr. Monje. That's a very similar answer for Twitter as
well. We don't allow violent extremist groups. We don't allow
glorification of violence. I think it's also when--you know,
Charlottesville was a hard day for a lot of folks, and I think
what you saw not only online was the very small minority of
folks who were saying terrible things, but the vast majority of
folks who were coming out to reject it.
Senator Baldwin. I'm going to turn to a different topic.
I'm concerned by President Trump's and Secretary of State
Tillerson's reluctance to support, fund, and staff the State
Department's Global Engagement Center, which is tasked with
coordinating U.S. efforts to counter extremist propaganda and
recruitment as well as Russian active measures like
disinformation.
I'd like to hear from each of the companies about their
experiences working with the Center and how cooperative efforts
could be improved.
And, Mr. Watts, what are the national security impacts of
this administration's failure to prioritize the Center,
especially in the context of Russia?
Why don't we again go right down the line?
Ms. Bickert.
Ms. Bickert. Thank you, Senator. We are committed to
working with governments around the world in promoting and
finding counter-speech solutions. We have worked with the
Global Engagement Center and others in the U.S. Government. We
have found that collaboration to be effective. Often, what we
find is that government can be very effective as a convening
power for bringing together civil society stakeholders and then
industry and researchers to get together and share their
knowledge. That's something that we hope to continue in the
future.
Ms. Downs. Thank you, Senator. Our efforts to combat
terrorism on our product obviously start with making sure we're
removing the most egregious content. But an equally important
part of the strategy is our investment in counter-speech, to do
the hearts and minds work to address these issues at their
root. So we meet regularly with NGOs and government actors,
including the State Department and the Global Engagement
Center, to talk about counter-speech and the importance of
investing in that work.
Senator Baldwin. Mr. Monje.
Mr. Monje. A very similar answer as well, in that, you
know, government does have an important role in combating this
issue and not only investing in counter-speech but investing in
the groups that are authentic voices in their communities.
Senator Baldwin. Mr. Watts.
Mr. Watts. I'm absolutely baffled as to why the Global
Engagement Center--they received that mission, from what I
understood, in 2016 before the election. Senator Portman, if I
recall, was one of the leaders of that, and I had actually
communicated with their staff on the Russia issue. At a bare
minimum, the U.S. Government needs to have a real-time
understanding of what Russia is doing in social media.
The Hamilton 68 platform I've tried to provide to the U.S.
Government directly through multiple agencies. I have briefed
the U.S. Government since 2014 in different contexts on Russian
active measures. I sit here today and I have no answer for you.
I don't understand why we wouldn't, at a minimum,
regardless of the outcome of the election in 2016, want to
equip our intelligence agencies, our law enforcement agencies,
and the Department of Defense with just an understanding--we
don't even have to counter--just an understanding of what
Russian active measures are doing around the world. There's no
excuse for it. I can't understand it.
The Chairman. Thank you, Senator Baldwin.
Senator Udall.
STATEMENT OF HON. TOM UDALL,
U.S. SENATOR FROM NEW MEXICO
Senator Udall. Thank you very much, Mr. Chairman. I really
appreciate you and the Ranking Member pursuing this very, very
important topic.
Terrorism and social media is a challenging and, I think,
pressing subject, and I recognize that technology companies
cannot solve this alone. But they must do more, and I think
that has been highlighted by the questioning you've seen here
today. I'm focusing--my first question is similar to Senator
Baldwin's. I'm particularly concerned about the explosion of
white supremacists online.
In December, after years of posting fantasies about school
shootings and hate-filled racist rants over many internet
platforms and many other identities, a young man took a gun to
a local high school in Aztec, New Mexico, and killed two
students before taking his own life. And listening to you, I'm
wondering, you know, what can be done in this kind of
situation?
Ms. Bickert, in your testimony, you highlighted the efforts
that Facebook is taking to counteract ISIS and other foreign
terrorists. But can you speak to the efforts Facebook is taking
to fight one of the most and biggest--one of the biggest
threats to us in the United States, domestic terrorists like
white supremacists? I mean, in this kind of situation where you
have an individual under various identities taking positions
and indicating right on the edge of violence, what can be done
in this kind of circumstance, and have you run into situations
like this before?
Ms. Bickert. Thank you, Senator. It's certainly an
important issue. We stand against violence in all its forms,
and we don't allow any violent organization, regardless of
ideology. If we become aware of a threat of violence, credible
threats of eminent harm to somebody, we proactively reach out
to law enforcement authorities, and that is something that we
have done in cases where we've seen a threat like a shooter.
Whatever the ideology is, it doesn't matter. We will
proactively provide that to law enforcement.
Senator Udall. Mr. Watts, do you think more could be done
here based on the answers you hear?
Mr. Watts. In terms of domestic extremism, I side with the
social media companies in the sense that it's difficult to
understand where to fall, because there's not good leadership
from the U.S. Government about what a domestic extremism group
is. We have the luxury----
Senator Udall. Do you think we could do more there, in
terms of the government?
Mr. Watts. Yes. If we delineate more appropriately as a
Federal Government, we can then enable the social media
companies to effectively draw the line. I don't like the social
media companies having to decide what is free speech versus
violent speech or extremist versus norm. It puts them in a
terrible position. I also don't think it's good for business
and their platforms.
At the same time, you know, how do you do that short of a
violent threat or an eminent threat? To do that, we would have
to have the equivalent of an FTO or a foreign terrorist
designation program in the domestic context. I'm not sure how
we get there.
Senator Udall. And that may be something we should
consider, is how to urge the government to be more specific
here and outline areas where we could work with industry in
order to move that along.
Mr. Watts. I think--so it's difficult, even from the FBI's
perspective, that there are two different playbooks. There's
the international terrorist playbook and the domestic terrorist
playbook, and without that formalization of what an extremist
group is or an extremist, individually, is, it's really hard, I
think, for any corporation or company to evenly and
legitimately enforce any sort of regulation on a user or a
group.
Senator Udall. Mr. Monje, in your testimony, you outlined
Twitter rules against terrorism that expressly include that
users--and I'm quoting here--``cannot make specific threats of
violence or wish for serious physical harm, death, or disease
of an individual or group of people,'' end quote. I'm curious,
then, what Twitter's position is on one of the President's
video tweets, where he was body slamming a person with the CNN
logo superimposed on their face. The video appears to promote
serious physical harm to CNN reporters in the context of an
alarming increase in violence against reporters in the U.S.
Mr. Monje. Thank you very much, Senator, for that question.
No Twitter user is above the Twitter rules. As we action
accounts on any given moment, we are looking whether they are
trying to do satire, whether they're trying to do humor, even
if it's not successful humor. We also recognize that world
leaders do have a special voice, and it is in the public
interest for their constituents to hear from them.
Senator Udall. Well, I don't think this was humor, and I
don't think the result--I think if you look at the--what CNN
reporters have said since this, there's more violence toward
them. There's more animosity toward them. I think you need to
look at it in the whole context, and I would encourage all the
companies at this table to take threats to journalists very
seriously. I'm extremely concerned when any threats of
violence-based reporting that the President finds disagreeable
with our President calling U.S. media outlets ``the enemy of
the people.'' I think it is up to all of us to safeguard the
First Amendment.
Thank you very much, Mr. Chairman.
The Chairman. Thank you, Senator Udall.
Senator Tester.
STATEMENT OF HON. JON TESTER,
U.S. SENATOR FROM MONTANA
Senator Tester. Thank you, Mr. Chairman and Ranking Member
Nelson. Thanks for having this hearing, and I want to thank the
witnesses today.
I think we'll start with you, Mr. Watts. Can you tell me
why transparency behind who's paying for political ads and
issue-based ads is important?
Mr. Watts. Yes. I think the number one issue with that is
public safety. We saw with Russian active measures one of the
primary things they sought to do was actually mobilize the
population regardless of the election, mobilize people to
protest or counter protest, which can lead to violent
confrontations. At the same point, those advertisements, when
annotated and noted based on campaign laws, give legitimacy to
those advertisements so that the public actually knows what is
a real political stance versus a false or manipulated truth or
a narrative, they have to stand behind their actions. I think
that's important for the public to restore trust and faith in
the democratic processes.
Senator Tester. Could it also have impacts on election
results?
Mr. Watts. Yes. It makes it more difficult for a foreign
adversary or even a social media manipulator with a lot of
resources and an ax to grind to do character assassination or
to tear down social movements.
Senator Tester. OK. So--and this goes to any one of the
other three that wants to answer this. Can you tell me why you
don't tell us who's paying for the ads, whether they're
political ads or whether they're issue-based ads? Who wants to
answer that?
Mr. Monje. Thank you, Senator. Twitter is very proud that
we last year announced industry-leading transparency practices
for political advertising.
Senator Tester. So do you tell people who's paying for the
ads?
Mr. Monje. For electioneering ads, yes, sir.
Senator Tester. How about issue-based ads?
Mr. Monje. Issue-based ads are a harder-to crack. It's
harder to determine--and we are working with our colleagues,
with our peer companies, to try to figure out what the right
way to address those issues are.
Senator Tester. OK. How about the other two? Do you want to
talk about political ads versus issue-based ads and if you're
telling us who's paying for them?
Ms. Downs. We're working to put more transparency into the
election-based advertising system and are taking four steps in
advance of the 2018 midterms. The first is verification. We
will require advertisers to identify who they are and where
they're from before purchasing advertisements. We'll also
launch in-ad disclosures where we notify users of who is
running an election-based ad. We'll release a transparency
report on election advertising purchased through Google, and
we'll also release a creative library to the public where all
of those advertisements are made public.
Senator Tester. And will that release of the transparency
report have who's paid for the ads?
Ms. Downs. I believe it will, yes, sir, Senator.
Ms. Bickert. Thank you, Senator. Our answer is
substantially similar to my peer companies on the issue of
Federal election related ads. And like Mr. Monje, political
ads, broadly, is a little bit more complicated, but certainly
an area where we think increased transparency is important.
Senator Tester. Political ads are more complicated than
issue-based ads, or the other way around?
Ms. Bickert. The issue-based--I'm sorry, Senator. The
issue-based ads--they're hard to define. But that said, we're
very interested in how we can increase transparency, and we
look forward to talking to yourself and other policymakers
about it.
Senator Tester. Well, I would just tell you this as an
editorial comment. I would agree with Senator Schatz. I don't
think this is a Democrat-Republican issue. I think this is a
democracy issue, and you guys are smart guys, and just about
everybody I read writings of tell me that it's not that
difficult, and they're smart people, too. So I would hope that
you guys really would put pen to paper, if that's what you do
these days, and figure out how you can let people know who's
paying for ads. And I think issue-based ads, by the way, are
just as important as political-based ads, because those fall
into the political category, and I would just say that's
important.
Every one of you said that you did not like the FCC
decision on net neutrality that came out a month or two ago.
During that debate, we had learned that there were bots that
dropped comments into the hopper that distorted the whole
public comment period. How is that going to be stopped the next
time we have a public comment period on a rule that's written
by an agency?
Anybody want to answer that? And I'm out of time, so make
it quick.
[No verbal response.]
Senator Tester. I'll tell you what. We'll not occupy the
time of the Committee. Give me an answer to that in writing
when you go back to your folks.
This is a really important issue. I just want to say this
is a really important issue, from a terrorist standpoint, from
all the questions that were asked before. But our democracy is
at risk here. We've got to figure out how to get this done and
get it done right and get it done very quickly, or we may not
have a democracy to have you guys up to hear you out.
The Chairman. Thank you, Senator Tester.
Senator Young.
STATEMENT OF HON. TODD YOUNG,
U.S. SENATOR FROM INDIANA
Senator Young. Well, thank you, Mr. Chairman, for holding
this hearing on terrorism and social media.
YouTube went from having 40 percent of its post takedowns
last June, being identified by algorithms, AI, and machine
learning, to 98 percent today. Twitter went from roughly 33
percent detection of terrorist accounts in 2015 to more than 90
percent of the detections today, again attributable to
algorithms, AI, or machine learning. Facebook has stated that
nearly 99 percent of ISIS and al Qaeda related content is
detected and removed before anyone even reports it.
So what is--Ms. Bickert, Ms. Downs, Ms. Monje, what's
responsible for the recent increase in the use of AI and
machine learning for this purpose of taking down posts? Is it
primarily because of a new commitment to take down posts by
your companies, or is it simply that the technology is finally
at a place to be effective, or some combination thereof? We'll
start with Ms. Bickert.
Ms. Bickert. Thank you, Senator. It's definitely a timely
question. These innovations have been happening over the years.
We have seen a lot of improvement, particularly over the past
one to two years at Facebook. A lot of these efforts have been
in place since I joined the company 6 years ago, such as still
image hashing, but it has gotten better.
In the fall of 2016 is when we finally found video hashing
to be sufficiently reliable, where we could use it to detect
these terror propaganda videos. And for some of them, like a
beheading video, that we know violates our policies regardless
of how it's shared, we could actually accurately identify it
and stop it at the time of upload. That's something we've been
trying to do for a while and had not been able to do.
Another area where we've gotten better is in detecting
recidivists. So we take down the bad account. They try to come
back. That's something that for a variety of reasons has been
important to the company for years, but an area where we've
made significant progress in the past one to two years.
And then the final advance I'll point to before turning to
my colleagues is in the area of natural language understanding.
This is hard. We have many different languages that we support
on Facebook, and when you train these models, they have to be
trained on sufficient data. So this process takes a long time,
but we are making progress here, and we're now using it in the
area of terrorism where we couldn't before.
Senator Young. Thank you.
Ms. Downs.
Ms. Downs. Thank you, Senator. We've always used a mix of
technology and humans to enforce our policies, and as
technology gets better, we see it doing more of the heavy
lifting in detecting the content that violates our policies and
needs to be removed. These are a reinforcing loop, where as
humans make judgments about what content violates our policies,
that feeds back into the training set of data to teach the
classifiers and algorithms what they're looking for. So the
more content we review over time, the better and better these
classifiers get and the more they're able to detect the content
that needs to be removed.
Senator Young. Before I turn to--my apologies--Mr. Monje--I
just stepped into the room--I would just note you referenced
human judgments and how that feeds into an algorithm to help
make more informed decisions moving forward. There won't be
time to explore it here, but one of the things I really want to
learn more about is what parameters are used to determine, by a
human, what is an appropriate or an inappropriate post, and is
there transparency, or will there be transparency about that
decisionmaking process? But, again, that's for another day
since I have 47 seconds left.
Mr. Monje.
Mr. Monje. Yes, sir. I'd just very briefly--we approach it
very similarly to our peer companies and are constantly trying
to figure out ways that we can use our technology and feeding
it the input so that it can tackle--the AI can tackle
increasingly more difficult and more nuance challenges.
Senator Young. OK. I'll just note in the remaining time
here that I really enjoyed visiting with Yasmin Green, Director
of R and D at Google's Jigsaw group. I'll say that Alphabet is
doing some really great work there, and I look forward to
working with all of you to improve how we remove this horrible
content from the Internet and keep Americans more safe and
secure.
Thank you, Mr. Chairman.
The Chairman. Thank you, Senator Young.
Next up is Senator Blumenthal.
STATEMENT OF HON. RICHARD BLUMENTHAL,
U.S. SENATOR FROM CONNECTICUT
Senator Blumenthal. Thanks, Mr. Chairman. Thank you for
holding this hearing and thank you to this really all-star
panel for being here today.
Mr. Watts, I find your testimony absolutely chilling. The
Internet is a potential monster when it comes to extremists and
terrorism, and it requires the kind of inventive and robust
investment attitude that, in fact, created the internet. I've
been reading a book called The Innovators by Walter Isaacson,
and it is an inspiring account of how we came to have the
Internet and social media, involving heroes whose names have
been largely lost to history, including some Nobel Prize
winners.
But the point that he makes that I think is so relevant to
this discussion is that the Internet itself is the result of a
partnership between private industry and inventors, government,
and academia, and those partners are as necessary for this
effort in combating terrorism and extremism as they were in
inventing the platforms themselves.
I want to join in thanking you for your commitment to net
neutrality. I also want to thank you for the commitments that
your companies have made with varying degrees of enthusiasm to
our anti-sex trafficking efforts, most especially SESTA, which,
hopefully, will come to a vote. I encourage you to enlist more
of your colleagues in that effort.
I want to highlight the importance of the nationalist hate
groups and extremist groups that have come to pose a very dire
extremist threat. You received a letter signed by 19 civil
rights groups, including Muslim advocates, on October 30, 2017.
It was co-signed to Facebook, but it's equally applicable to
all of your companies. I ask that it be made a part of the
record, and I see the Chairman has stepped off, but I'm
assuming there will be no objection and it will be made part of
the record.
[The information referred to follows:]
October 30, 2017
Mr. Mark Zuckerberg, Chief Executive Officer
Ms. Sheryl Sandberg, Chief Operating Officer
Facebook, Inc.
1 Hacker Way
Menlo Park, CA 94025
Dear Mr. Zuckerberg and Ms. Sandberg,
We, the undersigned civil rights, interfaith, and advocacy
organizations write to express our deep concern regarding ads, pages,
and hateful content on your platform used to divide our country, and in
particular, to promote anti-Muslim, anti-Black, anti immigrant, and
anti-LGBTQ animus. We thank you for recent meetings with some of our
organizations representing communities that were directly affected by
the material on your platform. We appreciate that senior members of
your team--including you, Ms. Sandberg--have facilitated these
meetings, and we hope that these conversations are the beginning of a
serious and ongoing dialogue. Now, it is necessary for Facebook to take
critical steps to address the bigotry and discrimination generated on
your platform.
As you know, we do not yet have access to all the divisive content
targeting communities we represent; therefore, we are only able to cite
to the few examples that were leaked to the media.
For example, Russian operatives set up misleading accounts
impersonating or posing as American individuals and groups on Facebook
to promote Russian propaganda during the American election season.
Reports indicate that a Russian Facebook account called
``SecuredBorders'' posed as a group of U.S. citizens concerned about
the increased number of refugees in America. This fake account not only
promoted anti-immigrant messaging online, but also managed to organize
an in-person anti-refugee rally in Twin Falls, Idaho in August 2016.\1\
---------------------------------------------------------------------------
\1\ Geoffrey Smith, ``Russia Orchestrated Anti-Immigrant Rallies in
the U.S. via Facebook Last Year,'' Fortune, Sept. 12, 2017, available
at http://fortune.com/2017/09/12/russia-orchestrated-anti-immigrant-
rallies-in-the-u-s-via-facebook-last-year/.
---------------------------------------------------------------------------
In addition, a Facebook page entitled ``United Muslims of America''
was an imposter account traced back to Russia \2\--the real United
Muslims of America is a California-based interfaith organization
working at the local level to promote dialogue and political
participation.\3\ The imposter account smeared political candidates and
promoted political rallies aimed at Muslim audiences.\4\ In another
example, the Internet Research Agency in Russia promoted an anti-Muslim
rally thousands of miles away in Houston, Texas where individuals
protested outside of a mosque.\5\ Additional reports indicate that
Facebook offered its expertise to a bigoted advocacy group by creating
a case study testing different video formats, and advising on how to
enhance the reach of the group's anti-refugee campaign in swing states
during the final weeks of the 2016 election.\6\ These examples of
content on Facebook were not only harmful, but also used to rile up
supporters of President Trump.
---------------------------------------------------------------------------
\2\ Dean Obeidallah, ``How Russian Hackers Used My Face to Sabotage
Our Politics and Elect Trump,'' The Daily Beast, Sept. 27, 2017,
available at https://fwww.thedailybeast.com/how-russian-hackers-used-
my-face-to-sabotage-our-politics-and-elect-trump.
\3\ United Muslims of America ``About'' page, available at http:/
fwww.umanet:orgfabout-us.
\4\ Obeiallah, supra note 1.
\5\ Tim Lister & Clare Sebastian, ``Stoking Islamophobia and
secession in Texas--from an office in Russia,'' CNN Politics, Oct. 6,
2017, available at http://www.cnn.com/2017/10/05/politics/heart-of-
texas-russia-event/index.html.
\6\ Melanie Ehrenkranz, ``Facebook Reportedly Used Anti-Muslim Ad
as Test Case in Video Formats,'' Gizmodo, Oct. 18, 2017, available at
https://gizmodo.com/facebook-reportedly-used-anti-muslim-ad-as-test-
case-in-1819645900.
---------------------------------------------------------------------------
Furthermore, it has been reported that Russian operatives purchased
Facebook ads about Black Lives Matter--some impersonating the group and
others describing it as a threat.\7\ This included ads that were
directly targeted to reach audiences in Ferguson, Missouri and
Baltimore, Maryland. CNN reports that the Russian Internet Research
Agency used these ads in an attempt to amplify political discord and
create a general atmosphere of incivility and chaos.\8\ This included a
fake ad containing an image of an African-American woman dry-firing a
rifle, playing on the worst stereotypes regarding African-Americans as
threatening or violent.\9\
---------------------------------------------------------------------------
\7\ Adam Entous, Craig Tim berg, & Elizabeth Dwoskin, ``Russian
operatives used Facebook ads to exploit America's racial and religious
divisions,'' The Washington Post, Sept. 25, 2017, available at https://
www.washingtonpost.comfbusiness/technology/russian-operatives-used-
facebook-ads-to-exploit-divisions-over-black-political-activism-and-
muslims/2017/09/25f4a011242-a21b-11e7-ade1-
76d061d56efa_story.html?tid=sm_tw&utm_term=.e49cecc1a834.
\8\ Dylan Byers, ``Exclusive: Russian-bought Black Lives Matter ad
on Facebook targeted Baltimore and Ferguson,'' CNN Media, Sept.
28,2017, available at http://money.cnn.com/2017/09/27fmedia/facebook-
black-lives-matter-targeting/index.html.
\9\ Adam Entous, Craig Timberg, & Elizabeth Dwoskin, ``Russian
Facebook ads showed a black woman firing a rifle, amid efforts to stoke
racial strife,'' The Washington Post, Oct. 2, 2017, available at
https://www.washingtonpost.com/business/technology/russian-facebook-
ads-show
ed-a-black-woman-firing-a-rifle-amid-efforts-to-stoke-racial-strife/
2017/10/02/e4e78312-a785-11
e7-b3aa-c0e2e1d41e38_story.html?utm_term=.aa2267a2f46c.
---------------------------------------------------------------------------
We were alarmed to see your platform being abused to promote
bigotry, and especially disappointed that it has taken media exposure
and congressional oversight to give a degree of transparency into your
practices. It is important to keep in mind that pervasive bigotry has
long existed on your platform, and the Russian operatives simply
exploited the hateful content and activity already present. We are
concerned about how a platform like Facebook's could operate without
appropriate safeguards that take into account how it could be
manipulated to further sow divisions in our society.
As a company and social network platform whose mission is ``to give
people the power to build community and bring the world closer
together,'' \10\ we hope that you understand the gravity of this
hateful rhetoric and behavior. During a time when anti Muslim, anti-
Black, anti-LGBTQ, and anti-immigrant sentiment has swept the nation,
it is more important than ever for companies like yours to take an
unequivocal stance against bigotry.
---------------------------------------------------------------------------
\10\ Facebook ``About'' page, February 4, 2004, available at
https://www.facebook.com/pg/face
bookf/about/?ref=page_internal.
---------------------------------------------------------------------------
Over the years, many of us have raised concerns about how your
platform may have a negative impact on our communities, with
disappointing results. For example, we have requested that you address
attacks on African Americans and Muslims, organizing by hate groups,
and the censorship of Black, Arab, Muslim, and other marginalized
voices. As a result of the pervasive presence and organizing by hate
groups on your platform-some could not exist as national level entities
without it we have repeatedly requested that you convene a gathering
with civil rights organizations to discuss appropriate and strategic
responses. While you were unable to sufficiently respond to the
concerns raised above, Facebook participated in and organized events
that stigmatized Muslims and other communities such as a recent
convening called ``Tech Against Terrorism.''
Though in the past you have displayed a willingness to listen to
our concerns, we have yet to see meaningful change. It is our hope that
recent developments will mark a new chapter in Facebook's commitment to
protecting the rights of all who use your platform.
As we continue this important dialogue, we urge you to:
1. Fully disclose to the public all of the ads, pages, events,
accounts, and posts you have traced back to Russian operatives
targeting African American, LGBTQ, and Muslim communities. In
particular, we believe that Facebook has a special
responsibility to notify those individuals and organizations
who have been impersonated or misrepresented.
2. Bring on an independent third-party team to conduct a thorough
and public audit of the civil rights impact of your policies
and programs, as well as how the platform has been used by hate
groups, political entities, and others to stoke racial or
religious resentment or violence. Other leading companies in
the industry like Airbnb have made the decision to conduct such
an assessment, and we hope you will follow their lead.
3. Regularly convene a new working group of a diverse group of civil
rights organizations working to counter bigotry, and solicit
input on policies and processes from this group. And, integrate
addressing hate into Facebook's corporate structure by:
a. Assigning a board committee with responsibility for assessing
management efforts to stop hate groups, state actors, and
individuals engaged in hate from using your platform and
tools;
b. Assigning a senior manager who is a member of Facebook's
Executive Team with authority to oversee addressing hate
company-wide and name that person publicly and employing
staff with expertise in this area to vet advertisements and
develop process and procedures the address this issue; and,
c. Creating a committee of outside advisors with expertise in
identifying and tracking hate who will be responsible for
producing an annual report on the effectiveness of steps
taken by Facebook.
4. Develop, with input from diverse civil rights groups and experts,
and make public a clear process for how Facebook:
a. Reviews content constituting hate speech;
b. Reviews efforts to use Facebook as a platform to stoke
identity-based, racial, or religious resentment or violent
actions; and,
c. Responds to complaints about content that reasonably creates
fear and chills speech on Facebook.
5. Make public detailed information regarding training and support
for anti immigrant, anti-Muslim, anti-black, and anti-LGBTQ
organizations, including the monetary value of these services;
and establish a fund to provide grants to organizations
combating hatred and bigotry.
Thank you in advance for your consideration. Please contact Naheed
Qureshi at [email protected] with any questions.
We look forward to your reply.
Sincerely,
Arab American Institute (AAI)
Asian Americans Advancing Justice/AAJC
Center for Media Justice
Center for New Community
Color of Change
CREDO
Human Rights Campaign (HRC)
The Leadership Conference on Civil and Human Rights
League of United Latin American Citizens (LULAC)
MoveOn.org
Muslim Advocates
NAACP
NAACP Legal Defense and Educational Fund, Inc. (LDF)
National Center for Lesbian Rights
National Hispanic Media Coalition
National LGBTQ Task Force
National Sikh Campaign
Sikh Coalition
Southern Poverty Law Center
Senator Cortez Masto. No objection.
[Laughter.]
Senator Blumenthal. And I'm not willing to yield a part of
my time to address that objection.
The Southern Poverty Law Center has warned that social
media has been instrumental to the growth of the alt-right
movement, allowing legions of anonymous Twitter users to use
the hashtag alt-right to push far right extremism. On YouTube,
for example, it's easy to find anti-Semitic content. All of
these forms of extremism, often white supremacist extremism,
have been allowed to flourish, and they pose a real and present
danger.
In the time that I have left, I want to ask about a letter
that I wrote to Facebook, Google, and Twitter calling on these
companies to individually inform all users who are exposed to
false, misleading, and inflammatory posts generated by Russian
agents. I'm assuming that none of you have any doubt that the
Russians meddled in our 2016 election and attacked our
democracy. Any question?
[No verbal response.]
Senator Blumenthal. None. And that the investigation of
those efforts is not a hoax, or a witch hunt, that this danger
is continuing, as Mr. Watts has so dramatically and powerfully
outlined, and that they will continue to do it unless they're
made to pay a price, and those who colluded and cooperated with
them are made to pay a price.
I want to thank Facebook for its substantive response in
terms of its commitment to providing consumers with an online
tool to inform users if they have interacted with Russian-
sponsored pages or accounts. I'm hopeful that Facebook will do
even more with more robust steps to further increase
transparency in the future, but I am very, very grateful for
your beginning.
And I just want to be blunt. I am disappointed by Google's
written response. It essentially blew off my concerns by saying
the nature of the platform made it difficult to know who has
viewed its content. I look forward to responses from Twitter
and others. If you want to respond now, I would be eager to
hear what your response is to the letter that I wrote.
Mr. Monje. Thank you, Senator, and we have briefed your
staff on our plans, and we'll be rolling out the fulsome
response shortly.
Senator Blumenthal. And what will that response be?
Mr. Monje. We will be working to identify and inform
individually the users who may have been exposed to the IRA
accounts during the election.
Senator Blumenthal. Thank you. I think it's so tremendously
important that we have all hands on deck in dealing with this
threat, not only the companies that are represented here, but,
again, as Mr. Watts said, some of the smaller actors, some of
the newer ones. And there will be others coming that provide,
in effect, platforms for hate, extremism, terrorism, division,
chaos. In some ways, they are the biggest threat to our
democracy today, those groups that want to foster hate.
And, of course, the Russians will continue. They have an
asymmetric advantage here. It's an absolutely wondrous
investment for Vladimir Putin. He gets more return on the
dollar than any other investment he can make in sowing chaos
and discord in our democracies, and we must be as inventive as
the innovators were, the inventers of the internet, in
combating this threat to our democracy.
Thank you, Mr. Chairman.
The Chairman. Thank you, Senator Blumenthal.
Senator Cortez Masto.
STATEMENT OF HON. CATHERINE CORTEZ MASTO,
U.S. SENATOR FROM NEVADA
Senator Cortez Masto. Thank you.
Welcome and thank you for this conversation. I wanted to
start with, unfortunately, a horrific tragedy that occurred in
my home town on October 1, 2017. Las Vegas experienced the
worst tragedy that we have ever seen, and it is the worst mass
shooting in American history. As we were dealing with the
horrific tragedy of the situation and trying to gain
information, particularly for law enforcement purposes,
unfortunately, a lot of misinformation was being spread after
that tragedy on some of your platforms and on the internet,
and, particularly, misinformation about the shooter was
highlighted on both Google and Facebook.
Obviously, that's incredibly unhelpful for law enforcement,
particularly as we move through an unfolding potentially
dangerous situation. I know both Facebook and Google cited the
need to make algorithm improvements to fight the spread of fake
news during a crisis. What do you see as your companies' roles
in fighting fake news, especially during a crisis such as a
mass shooting or a terrorist attack, and what specific and
verifiable metrics can you provide us to ensure our trust in
these remedies? I'll start with Ms. Bickert.
Ms. Bickert. Thank you, Senator. What happened in Las Vegas
was horrific, and there were false news stories that we saw
that we did address, but not fast enough, and it's an area
where we're trying to get faster. We've changed the way that
our crisis center operates so that we can make sure that that
type of false news story does not appear in the headlines that
people are seeing. The crisis center can be incredibly useful
during times like this.
In Las Vegas, we saw people using not only our safety
check, which allows people to say that they're safe, but also
coordinating help, offers of housing and assistance to people
throughout the city. So we want to make sure that's working
effectively. Things we're doing: removing the bad accounts that
are propagating this false news, making algorithmic changes to
make news that is likely to be false less visible on the site,
providing related articles when people see a news story that
has been flagged as something that might be false so that they
can see the broad spectrum of information across the internet,
and then working with responsible publishers to make sure that
they know how to use our tools to get their stories out there.
Senator Cortez Masto. Thank you.
Ms. Downs.
Ms. Downs. Thank you, Senator, and my heart goes out to the
City of Las Vegas and all the victims of that senseless
tragedy.
We take misinformation on our platforms very seriously, and
we've made a lot of efforts in our products, from improvements
to our ranking algorithms to highlight authoritative sources
and to demote low-quality or less reliable sources,
particularly when users are seeking news content. We also have
strict policies in place against the monetization of news sites
that are misrepresenting themselves in order to remove the
financial incentive to create and distribute fake news.
Senator Masto Cortez. Thank you.
Mr. Monje. And a very similar answer for us. I'd only add
that one of Twitter's great advantages in the world is that
it's fast. It's faster than television news often. We try to
arm emergency responders with the knowledge of how to use that
as a strength, and so it's one of our key pieces.
During the hurricanes in the Gulf Coast, we were actively
working with folks who were responding--they were actually
folks in Texas and Houston who were using our platform to
identify people to rescue. And so it's one of the strengths of
our platform, and, like everyone, it's a continuing challenge
to address misinformation.
Senator Cortez Masto. Thank you.
Mr. Watts, would you like to address this, or is there
anything else that can be done?
Mr. Watts. I don't know in terms of the technical things
that could be done. But I do think the spread of misinformation
so quickly like that--the first thing that you see is what you
tend to believe over time. That which you see the most is what
you tend to believe as well. It really empowers social media
manipulators if you can do amplification through social bots,
or if you can generate other systems to push the news quicker
than everybody else, and so you see a lot of gaming in terms of
trending hashtags and things like that.
I think there has to be some sort of trip that you can put
in technically over time--and I'm sure that all these companies
are trying to develop--that will tamp that out. When you see an
artificial spike in any one of those trends, you should be able
to detect it, and I think they're advancing on that. But it's a
huge public safety issue, regardless of the threat actor that's
employing it.
Senator Cortez Masto. Thank you. I know I'm running out of
time, but let me just say this. I had the opportunity to work
with Facebook on our Internet Crimes Against Children Task
Force in Nevada when I was Attorney General, and I will tell
you that for every company that we reached out to, whether it
was YouTube, Google, they were willing to work with law
enforcement.
So I know there has been a lot of discussion on that
interaction that you've had with law enforcement, but I've seen
it from one side of it. I know now there is this balance we
need to find to figure out how we continue to work together to
address these evolving crimes and activity that's happening on
the internet, and I'm grateful that you're here, and I look
forward to figure out how we can continue to evolve that
relationship as well. So thank you.
The Chairman. Thank you, Senator Cortez Masto.
Senator Lee.
STATEMENT OF HON. MIKE LEE,
U.S. SENATOR FROM UTAH
Senator Lee. Thank you, Mr. Chairman.
Thanks to each of you for being here. We live in an
exciting world. We live at a time when the companies
represented at this table today 15 years ago were just ideas,
and today, they've changed the way we interact with the world
around us. Today, these companies have made it possible in ways
never imagined just a couple of decades ago for a few people
with very little money to have an impact, not only in their
community, but across the country and throughout the world. But
with that comes a lot of challenges, and those challenges are
the reason why we're here today at this hearing.
In some parts of the world, there has been a suggestion
that I can summarize only as an effort to make public utility
companies out of social media enterprises that would rather
comprehensively attempt to regulate social media, imposing
escalating fines and other penalties on companies that fail to
report certain types of information to the government. Some of
these recommendations for policies like this have been made in
the United Kingdom and in the European Union.
To me, this is kind of distressing, in part because I worry
about what that would do to private property, what that would
do to these thriving businesses that have given so many people
so much of an opportunity to be heard. I also worry about what
it would do to public safety, the very end sought to be
achieved by these proposals. Sometimes when government gets
involved and it sets a certain standard in place, that becomes
both the floor and the ceiling. Understandably, I would worry
about that.
So I'd like to--we'll start with you, Ms. Bickert. Tell me
what you think about proposals like that and what some of the
risks might be to starting to treat social media companies like
public utilities?
Ms. Bickert. Thank you, Senator. I think whenever we think
about regulation, there often are unforeseen consequences, and
those can impede our ability to provide services to the people
that trust and need our products. I think the big thing for us
is that our incentives are often aligned with those of
government in terms of creating a safe community.
On this issue, absolutely, there is no question that the
companies here do not want terrorists using their platforms.
The long-term business interest for Facebook is we need people
to have a good experience when they come to Facebook. We need
them to like this community and want to be a part of it, and
that means keeping them safe and removing bad content. So the
incentives are there. These companies are working together to
address these challenges, and that's how we think it can work
best.
That said, we will continue to have a productive dialog
with government. The concerns that you face and what you're
hearing from your constituents matter to us very much, and we
want to make sure that we're considering that in responding to
that.
Senator Lee. In light of the fact that your company and
others have--that the progress that your company and others
have made in this area does not suggest that some of these
proposals are unnecessary, in any event?
Ms. Bickert. Thank you, Senator. Because our incentives are
aligned, the kind of progress that you're going to see is going
to happen regardless of what we're seeing from governments,
what we're hearing from governments. It's still important to
have that dialog. We learn every time that we engage with
policymakers. But the incentives exist independently.
Senator Lee. Ms. Downs, would you agree with that?
Ms. Downs. Yes. The security and integrity of our products
is core to our business model, and that includes the expedient
enforcement of all of our content policies. So we are already
sufficiently motivated to invest the necessary resources and
people in addressing this threat.
Senator Lee. And how might treating you more like a public
utility change that dynamic?
Ms. Downs. I think the risks that you outlined are
important things for policymakers to remain cognizant of.
Obviously, the tech industry is incredibly innovative, has
created tremendous economic opportunity, and anything that
slows down that innovation will cause damage to the ability of
the industry to continue to thrive.
Senator Lee. Mr. Monje.
Mr. Monje. I'd agree with that. We take our responsibility
extremely seriously, and it is important to our business to get
it right. We measure progress in matters of weeks and months.
We move very quickly. So I'd agree with everything that was
said.
Senator Lee. I've got one second remaining, if I can just--
how do you determine--we'll just go with you, Mr. Monje, since
we're already on you. How do you determine what constitutes
terrorist or extremist content? For example, do you make this
determination internally within your staff? Do you have certain
subject matter experts that help you decide that?
Mr. Monje. Yes, sir. We have former law enforcement
officials who are on our team. We also interact with and
communicate with governments and NGOs to determine that on an
individual basis.
Senator Lee. My time has expired. Thank you very much.
Thank you, Mr. Chairman.
The Chairman. Thank you, Senator Lee.
Senator Hassan.
STATEMENT OF HON. MAGGIE HASSAN,
U.S. SENATOR FROM NEW HAMPSHIRE
Senator Hassan. Thank you very much, Mr. Chairman, and
thank you to our panelists today.
I want to talk about the See Something, Say Something
campaign. The campaign is simple, and many of the thwarted
terrorist attacks in the U.S. were stopped because everyday
people alerted authorities to suspicious behavior. So I'd like
to get a better sense of whether your companies fully embrace
this See Something, Say Something campaign. While I understand
that most of you shut down accounts that espouse violent
extremist propaganda, it's not clear that you proactively
report those accounts to law enforcement.
Here's an example of why that makes us less safe. In 2012,
Tamerlan Tsarnaev, one of the Boston Marathon bombers, posted
on YouTube several videos espousing al Qaeda propaganda under
the name Muaz. At the time, the FBI was unaware of this
account. However, the FBI had previously investigated Tamerlan,
thanks to a tip from the Russians, but found nothing to
corroborate the Russians' claims.
In September 2012, Tamerlan applied for U.S. citizenship
with DHS. As part of the vetting process, DHS instructed the
FBI to run a check on the application, which came up all clear.
However, in his application, Tamerlan revealed that he tried to
change his legal name to Muaz, the same name as his YouTube
account. Eight months later, Tamerlan orchestrated a terrorist
attack with his brother that resulted in the death of four
people and almost 300 injured.
In hindsight, if YouTube had reported Muaz's troubling
social media account to the FBI, then maybe the FBI would have
been able to link Tamerlan to Muaz's extremist YouTube account
when Tamerlan was applying for citizenship. That could have
prompted the FBI to re-open a closed terrorism investigation
just weeks before Tamerlan carried out this awful tragic
bombing.
So to Mr. Monje and Ms. Bickert and Ms. Downs, I'd like to
understand how and when your companies report extremist
accounts to law enforcement, and has it changed since the days
of the Boston Marathon bomber?
Mr. Monje. Thank you, Senator, for that question. When
we're aware of an imminent threat, we absolutely do proactively
reach out to law enforcement. Whenever they come to us and ask
for information, as long as they have the right process, which
we are very good with working with them to figure out, we will
respond as quickly as we can.
Senator Hassan. Thank you.
Ms. Downs?
Ms. Downs. Thank you, Senator. We also cooperate with law
enforcement pursuant to the legal process, including the
emergency disclosure provisions, where if we detect any content
on our services that poses a threat to life, we proactively
report it to law enforcement.
Senator Hassan. And Ms. Bickert?
Ms. Bickert. Thank you, Senator. The same answer.
Senator Hassan. Well, I thank you. I will say that the See
Something, Say Something campaign is premised on something a
little bit different than what you all just said, because it's
premised on ``if you think,'' not ``does this meet my
definition of eminent danger.'' But we ask members of the
public if they see something suspicious to step up, and what
you're all saying is that if it meets certain criteria or if
you're asked, and I think that's a little bit different.
So let me follow up with Mr. Watts. As a former Federal law
enforcement officer, how would you grade these companies'
performance in addressing violent extremist accounts? Do you
think they can do more to actively support Federal law
enforcement and counterterrorism officials?
Mr. Watts. Over the last decade or so, they've all done
better. Facebook and Google have outpaced Twitter. Twitter, in
my opinion, relies too much on technical signatures and doesn't
staff on the threat intelligence level to the extent that they
should.
Senator Hassan. Thank you very much.
Thank you. That's all the questions I have, Mr. Chair.
The Chairman. Thank you, Senator Hassan.
Senator Peters.
STATEMENT OF HON. GARY PETERS,
U.S. SENATOR FROM MICHIGAN
Senator Peters. Thank you, Mr. Chairman.
And to each of our witnesses, thank you for being here
today. It's an important topic, and we appreciate your active
involvement in this.
My question concerns the extent that algorithms are used
and play a role in the problem and how algorithms can also be
used as a solution to this problem that we're dealing with. I
was pleased to read Ms. Bickert's post in the Facebook newsroom
that Facebook has started using artificial intelligence to help
counter terrorist threat efforts on your platforms.
The speed and breadth of the Internet certainly makes it
nearly impossible for humans to keep track of all this. We need
to have AI systems to do that, and they need to continually
evolve if we're going to be effective in using them. However,
it is likely that algorithms may be partly responsible for
getting extremist material in front of users, whether it be in
search results through Facebook's news feed or YouTube's up
next list or elsewhere.
So my question is these algorithms are under your direct
control, as all platform providers can control that. What are
you specifically doing to learn more about whether and how your
algorithms may be promoting extremist content?
I'll start with you, Ms. Bickert.
Ms. Bickert. Thank you, Senator. The first thing that we
need to do is make sure we're removing the terror content, and
then it doesn't matter--once you take it out of the equation,
then the algorithm has no role in promoting it because the
content is simply not available on Facebook. That's something
that we do by, as you pointed out, using technology to find the
content.
But we don't stop there. After we find an account that is
associated with terrorism, if we remove that account, we also
fan out from that account. We look at associated content,
associated accounts, and we remove those as well. If we can get
better in that space, then we can make sure that the content is
not appearing before our community.
Senator Peters. Ms. Downs.
Ms. Downs. Thank you, Senator. Absolutely correct that the
first priority is making sure that none of this content is on
the platform in the first place. At the same time, we also have
teams that are protecting our algorithms from being gamed.
Obviously, this is a threat to our services and to our users'
experience on our services across many issues, and so we have
dedicated teams to make sure that people aren't manipulating
our systems and that they're working as intended to serve
relevant information to users who come to YouTube.
Mr. Monje. Very similar answer from Twitter. We've been
able to use our machine learning, our algorithms, to help
identify more than 90 percent of the terrorist content that
we've taken down before anybody else brings it to our
attention, 75 percent of those before they get to tweet once.
And, also, we protect our trends against manipulation. We've
done that since 2014, and we continually improve our processes
to protect our users' experience.
Senator Peters. Mr. Watts?
Mr. Watts. I would just note that any sort of algorithm
detection technique is only as good as what's already been seen
out in the world, which is part of the reason why the Russians
have been more successful in terms of social media
manipulation. They understand the terms of service. They have
the capabilities to actually beat those systems, and they play
within the rules.
The smarter, better-resourced, higher computational people
around the world that want to use it will do better. It's kind
of like zero day viruses in cybersecurity speak. Cybersecurity
protections, anti-virus, is only as good as what has already
been seen before in terms of malware, and so the only way to
get in front of that is to combine really smart threat analysts
on whatever threat actor it is that's out there with the
technologists, and those companies that do that do better in
terms of getting in front of these actions.
Senator Peters. What's your assessment of the companies
here and others in the United States?
Mr. Watts. I think Facebook and Google--I've seen massive
increases and much more success in that space. I think Twitter
gets beat oftentimes and can continue to get beaten because
they rely too heavily on technology, and I don't think they
have the partnerships they need to adequately get out in front
of it.
Mr. Monje. If I could respond to that, because----
Senator Peters. Absolutely.
Mr. Monje.--because he said it twice, and I disagree. I
think there are many external researchers who said that a lot
of this terrorist content doesn't--has moved off of our
platform. The average ISIS account in 2014 had 177 followers,
and now they have 14. They measure their life on Twitter in
minutes and hours. We are extremely effective at taking them
out. We do have the resources in place and the technology in
place to fight the fight.
Senator Peters. Mr. Watts?
Mr. Watts. They get beat by a new terrorist group every few
years. I mean, al Shabaab--we watched the entire Westgate
attack go down on Twitter, monitoring it. We had a key
monitoring list that we watch on that. With ISIS and al Nusra
in Syria, we were able to build that list of anywhere from
3,000 to 4,000 terrorist accounts at any given time.
They do better after the fact, once they pick up what the
signatures are. But the problem is you're always trailing
whatever the threat actor is. You're not staying out in front
of it, which is why in the cybersecurity space--or even some
social media companies are taking this on now--you employ the
threat analysts so they can work with the technologists.
Otherwise, the technologists are always behind the curve. They
have to wait until the group creates enough signatures that
they can detect it, and then they can weed it out.
They're getting better all the time, but AI and machine
learning, even with its advancements, can only detect what's
already been seen before. And what humans are very good at, at
least up until now and until they become autonomous machines
out there--they're good at gaming systems and figuring out ways
around it.
So I think in the case of the Russians, for example--and
I've seen the takedowns of their accounts by Twitter, and they
are hardly making a dent in what I'm seeing in terms of flows.
I can't confirm all those accounts that are out there, but, you
know, I hear about troll farms. Why do we think there's only
one?
So I think in terms of moving forward, there has got to be
a much bigger focus for those social media companies on putting
threat analysts and pairing them together--and I know both
analysts that have gone to Facebook and Google in that space,
and they haven't--you know, some have been there longer than
others, but I think that's the right approach moving forward.
Senator Peters. Thank you. I'm out of time. Thank you so
much.
The Chairman. Thank you, Senator Peters.
Senator Cruz.
STATEMENT OF HON. TED CRUZ,
U.S. SENATOR FROM TEXAS
Senator Cruz. Thank you, Mr. Chairman.
Welcome to each of the witnesses. I'd like to start by
asking each of the company representatives a simple question,
which is do you consider your companies to be neutral public
fora?
Ms. Bickert.
Ms. Bickert. Thank you, Senator. The mission of our company
is to connect people. We do not look at ideology or politics.
We want people to be able to connect and share who they are.
Senator Cruz. I'm just looking for a yes or no, whether you
consider yourself to be a neutral public forum.
Ms. Bickert. We do not have any policies about political
ideology that affect our platform.
Senator Cruz. Ms. Downs.
Ms. Downs. Yes, our goal is to design products for
everyone, subject to our policies, and on occasions they impose
on the types of content that people may share on our product.
Senator Cruz. So you're saying you do consider YouTube to
be a neutral public forum.
Ms. Downs. Correct. We enforce our policies in a
politically neutral way. Certain things are prohibited by our
community guidelines, which are spelled out and provided
publicly to all of our users.
Senator Cruz. Mr. Monje.
Mr. Monje. Yes, sir.
Senator Cruz. Well, let me focus for a minute, Mr. Monje.
As you know, there have been several videos that were released
in recent weeks that I and a lot of other people thought were
highly troubling, and so I want to give you an opportunity to
respond to that.
One individual, Abhinav Vadrevu, described as a former
Twitter software engineer, was captured on video saying the
following, quote, ``One strategy is to shadow ban so you have
ultimate control. The idea of a shadow ban is that you ban
someone but they don't know they've been banned, because they
keep posting and no one sees their content. So they just think
that no one is engaging with their content, when, in reality,
no one is seeing it.'' Is that a practice that occurs at
Twitter?
Mr. Monje. No, sir. We do not shadow ban users.
Senator Cruz. Why would this individual described as a
former Twitter software engineer say that?
Mr. Monje. Thank you for the opportunity to respond,
Senator, about this. These folks were caught on video. They
weren't speaking on behalf of the company. They were speaking
in their personal capacity. We do not shadow ban folks. What we
do do is if an account is spamming, meaning engaging in
malicious automation, we will hide--make it harder for them to
find--to be found on our platform.
If I could continue, sir, that was one of the reasons why
the efforts that we saw with the Russian misinformation didn't
hit as big a mark as they were hoping for. We were able to stop
that in real time. The other thing, sir--I'm sorry sir.
Senator Cruz. Another individual named Mo Norai, a Twitter
content review agent, was quoted on a video as saying ``On
stuff like that, it was more discretion on your viewpoint, I
guess how you felt about a particular matter. Yes, if they said
this is, quote, `pro-Trump,' I don't want it because it offends
me, this, that, and I say I ban the whole thing, and it goes
over here and they're like, 'Oh, you know what? I don't like
it, too. You know what? Mo's right. Let's go. Let's carry on.
What's next?' ''
Is that individual describing a practice that occurs at
Twitter?
Mr. Monje. No, sir. We use algorithms as a way to--if we
see an account that is being abusive, that also will be down
ranked. If they're engaging in targeted abuse against
minorities, if they're being--if they're consistently violating
our terms of service but they haven't crossed the line into
being suspended, we'll make it less visible. But what we won't
do is make--your followers will always be able to see you, and
we are not--we ensure that--if you go on Twitter at any moment,
you can see--you can see arguments on all sides of the issue.
Senator Cruz. Wait. I want to make sure I'm understanding
you right. You're saying for some people who are posting, you
will restrict viewership only to those who are actively
following them?
Mr. Monje. If we believe that they're engaged in malicious
automation, if we believe that they're violating our terms of
service when it comes to abuse.
Senator Cruz. So is it your position that the individuals
that are subject to this form of censorship are extremists or
fringe? Is that what you're telling us?
Mr. Monje. It depends on the user. I can tell you that this
is not something that we hide from the public. This is out in
the open, the fact that we will reduce the visibility of tweets
that are abusive or that are engaged in malicious automation.
Senator Cruz. Well, let me ask, what about Congresswoman
Marsha Blackburn? Is she someone you would consider somehow
abusive or fringe or otherwise?
Mr. Monje. No, sir.
Senator Cruz. Well, then, why did Twitter restrict and
censor her announcement video announcing as a candidate for the
U.S. Senate?
Mr. Monje. I want to be very clear about that, sir, and
thank you for the question. We never removed her tweet, and
what she did do is advertise on our platform. We do, like many
platforms, have a higher standard when it comes to advertising,
because we are putting in front of people things they didn't
ask to see. Her video was reported to us. There was a decision
that was made that was later reversed because of some of the
language that was used in her account. It was a mistake, and we
acknowledged it.
Senator Cruz. So her announcement was censored because it
was pro-life content. Has Twitter ever censored anyone for pro-
choice content?
Mr. Monje. She was never censored.
Senator Cruz. So you're saying nothing happened to her
tweet?
Mr. Monje. Her tweet got a lot of attention on the organic
side. We action our accounts, and we take our terms of service
very seriously. Sometimes we make the wrong decision. We have
action on all sides of issues, and we strive to be better every
day.
Senator Cruz. Let me ask a final question, because my time
has expired.
Ms. Downs, I'd like to know--what is YouTube's policy with
respect to Prager University and the allegations that the
content Prager University is putting out are being restricted
and censored by YouTube?
Ms. Downs. As I mentioned, we enforce our policies in a
politically neutral way. In terms of the specifics of Prager
University, it's a subject of ongoing litigation, so I'm not
free to comment on the specifics of that case.
Senator Cruz. Well, I will say the pattern of political
censorship that we are seeing across the technology companies
is highly concerning, and the opening question I asked you,
whether you're a neutral public forum--if you are a neutral
public forum, that does not allow for political editorializing
and censorship, and if you're not a neutral public forum, the
entire predicate for liability immunity under the CDA is
claiming to be a neutral public forum. So you can't have it
both ways.
Thank you.
The Chairman. Thank you, Senator Cruz.
I think we've exhausted all the questions. Thank you all
for being here. It has been a very informative session. We all
know that the Internet is an incredibly powerful tool which
offers enormous benefits to people globally. But we also
realize we live in a dangerous world, and that there are people
out there who want to do harm and do bad things and are looking
for any means in order to accomplish those. Of course, we know
that in the modern world, cyber has become increasingly the
tool of choice for a lot of bad actors.
So we appreciate your informing us about steps that you're
taking to try and police some of that bad behavior. As I said
earlier, you know, we have constitutional protections and a
Bill of Rights, and we also have--I think we want to make sure
that we have a light touch when it comes to regulating the
internet, and that's certainly something that I hope that this
Committee will continue to support, and that those at
regulatory agencies will adopt as well. But we also want to
make sure that we are doing what we can to keep our country
safe.
So we appreciate the efforts that you have undertaken
already and those that you--as you continue to develop and look
at ways to combat some of these threats that we face, and we
hope that, working together as partners, that we can do a
better job, and there's always room for improvement. So thank
you for what you've done and for what you continue to do, and
we'll look forward to discussing, I'm sure, in the future, as
the threats continue to evolve, things that we can do better.
Thank you all for being here.
I'm going to just say that before we close, I've got a
letter from the Consumer Extremism Project highlighting its
work on combating radicalization online, and I'm going to enter
that into the record, and also enter a piece by the Wall Street
Journal authored by the Counter Extremism Project's Senior
Advisor, Dr.--let me see if I can say this right here--Hany
Farid, underscoring his work on this important issue.
[The information referred to can be found in the Appendix.]
The Chairman. We'll keep the hearing record open for a
couple of weeks. Senators are encouraged to submit any
questions that they have for the record, and upon receipt of
those questions, we ask our witnesses to submit their written
responses to the Committee as quickly as possible.
Thank you all for being here. This hearing is adjourned.
[Whereupon, at 12:25 p.m., the hearing was adjourned.]
A P P E N D I X
Prepared Statement of the Counter Extremism Project
Thank you Chairman Thune, Ranking Member Nelson and distinguished
members of the Committee for holding this hearing to examine the
commitment and performance of Google/YouTube, Facebook and Twitter in
permanently removing persistent and dangerous extremist and terrorist
content on their platforms.
The Counter Extremism Project (CEP) is a not-for-profit, non-
partisan, international policy organization formed in 2014 to combat
the growing threat from extremist ideologies. Since its inception, CEP
has pioneered efforts to combat extremists' radicalization and
recruitment tactics online, and has persistently called upon Internet
and social media companies, to take definitive action and adopt
policies to stop the misuse of their platforms that has cost many lives
around the world.
The Internet promised to democratize access to knowledge, spread
great ideas, and promote tolerance and understanding around the globe.
This promise, however, is being poisoned by the rise of trolling,
cyber-bullying, revenge porn, fake news, child exploitation, hate,
intolerance, and extremist and terrorist propaganda.
The horrific aftermath of extremists' weaponization of the Internet
and social media platforms stretches from Paris, to Brussels, to
London, Orlando, San Bernardino, Istanbul, Beirut, Cairo, and New York
Only a few years ago, big technology companies flatly denied the
existence of this problem. And while their tone has undoubtedly
changed, CEP remains concerned about the level of commitment,
consistency and transparency that will be required to overcome the
systematic misuse by these platforms. While big social media platforms
acknowledge the existence of radicalizing content that violates their
stated terms of service, their response to date has followed a familiar
pattern utilized in response to other discoveries of abuse: denial,
followed by half-measures and attempts to spin the issue in the media,
and finally, reluctant action when faced with threats to their bottom-
line or possible regulatory action.
Make no mistake. There is no question that reigning in online
abuses is challenging. There is also no question, however, that we can
and must do more than we are to mitigate the harm that is being seeded
and fueled online, while maintaining an open and free Internet where
ideas can be shared and debated.
To cite but a few examples of ongoing problems of moderation. After
it was determined that Manchester suicide bomber Salman Abedi, who
killed 22 people on May 22, 2017 relied in part on ISIS bomb-making
instructional videos on YouTube to build his explosive device, Google
declared that bomb-making videos had no place on the platform. However,
that same video was still on YouTube almost two months after the
suicide bomb attack CEP has determined that the bomb making video has
been uploaded (and removed) from Google platforms at least 11times
since, most recently on January 9, although the actual number is most
likely much higher.
The ISIS video ``The Religion of Kufr is One,'' which shows
multiple executions by firearms and a hanging-clear violations of
YouTube's terms of service-has been uploaded and removed from YouTube
at least six times since May 30, 2016.
Google/YouTube's process for detecting and removing terror content
is still allowing prohibited content to be repeatedly uploaded and stay
live for a sufficient period of time for hundreds people to view and
download. Experience has shown that if most of the sharing of a video
can take place in the first few hours it is available, meaning if the
content is not removed quickly and uploads prevented, the moderating
process has failed.
The Committee will no doubt hear today from Google, Facebook and
Twitter about improvement they have made and pledges of more action.
CEP notes that in November, Google/YouTube removed the lectures and
sermons from al-Qaeda operative Anwar ai Awlaki from its platform. That
decision, however, followed a multi-year CEP campaign that included
direct outreach to the leadership of Google, a sustained effort to
highlight issues in the media, including via op-eds in USA Today and
Fox News, and a series of original reports detailing Awlaki's
ubiquitous presence on YouTube and other Internet platforms, as well as
his radicalizing influence on U.S. and European terrorists. For
example, CEP researchers identified 90 extremists in the U.S. and
Europe with ties to al Awlaki, including Said and Cherif Kouachi, who
carried out the Charlie Hebdo attacks; Omar Mateen, who killed 49
people in Orlando; Ohio State car attacker Abdul Razak Ali Artan;
Boston Marathon bombers Dzhokhar and Tamerlan Tsarnaev; and many
others.
One action does not constitute a lasting solution. Industrywide
standards are needed to ensure the timely and permanent removal of
dangerous content, especially when produced by groups and individuals
on the State Department's Foreign Terrorist Organizations list, the
Treasury Department's Specially Designated Nationals and Blocked
Persons list, and the United Nations Security Council Sanctions list,
and individuals with demonstrable links to violence. There is no
shortage of extremists online--Turki ai-Binali, Abdullah Faisal, Yusuf
al-Qaradawi and Ahmad Musa Jibril are notable examples. They must be
subject to the same treatment and their content should be swiftly and
permanently removed.
Existing technology can also assist with the enforcement of new
polices and prevent the re-upload of material from known extremists
such as Mr. Awlaki. Dr. Hany Farid, a professor of computer science at
Dartmouth College who advises our organization, developed an algorithm
called eGLYPH that quickly and accurately identifies for removal known
extremist material on the Internet and social media platforms. This
technology is based on software developed by Dr. Farid and Microsoft
almost a decade ago called PhotoDNA. In 2016 alone PhotoDNA was
responsible for the take-down of over 10 million child pornography
images around the world, based on known images as determined by the
National Center for Missing and Exploited Children (NCMEC).
It has already been proven that technology exists to effectively,
aggressively, and consistently filter content that is either illegal or
an explicit violation of a company's terms of service. While there is
no question that reasonable people can disagree about the extremist-
nature of some content, we can all agree that videos of a murder,
videos of beheadings, videos with explicit calls to violence, or videos
on how to build a suicide vest are extremist in nature--the worst-of-
the-worst--and violate terms of service of all major tech companies. It
is important to understand that technologies like eGLYPH, which was
offered to all three of the companies before you today for free, simply
allow companies to effectively, consistently, and transparently enforce
their own terms of service, their standards for what is and what is not
allowed on their networks.
There is no technological, economic, or legal reason why we cannot
purge major online platforms of the worst-of-the-worst extremist
content that grows more pernicious each year. There is no reason why we
can't significantly disrupt global online radicalization and
recruitment by hate and extremist groups. And, there is no doubt that
this can be accomplished in a thoughtful, effective, and transparent
manner, while respecting the privacy and rights of every user.
Lawmakers and the public should demand that tech finally implement
industry-wide standards and policies that ensure the timely and
permanent removal of dangerous extremist and terrorist material,
establish measurable best practices and transparently deploy proven
technologies to prevent there-upload of materials already determined to
violate company policies. If tech fails to act, then it is time for
regulators to promulgate measures to force the industry to take
necessary action to protect the public.
______
How Algorithms Can Help Beat Islamic State
Hany Farid `changed the world' by combating child porn. Now his
software could suppress terrorists online.
By Joseph Rago, Hanover, N.H.
You can't blame the message on the medium, not exactly. But maybe,
all things considered, arming everyone with pocket supercomputers, and
then filtering most of human experience through social-media feedback
loops, wasn't the greatest idea.
America recently endured the most electronic and media-saturated
presidential campaign in memory, with its hacks, private servers,
secret videotapes, fake news, troll armies and hour-by-hour Internet
outrage across all platforms. And however glorious modern
communications may be, they've also empowered a cast of goons, crooks
and jihadists to build audiences and influence worldwide.
A technological solution, at least to that last problem, may lie
2,600 miles east of Silicon Valley, in a computer-science laboratory at
Dartmouth College. Prof. Hany Farid, chairman of the department,
creates algorithms that can sweep digital networks and automatically
purge extremist content--if only the tech companies will adopt them.
``If you look at recent attacks, from Orlando to San Bernardino to
Nice to Paris to Brussels,'' Mr. Farid says, ``all of those attackers
had been radicalized online. They weren't going to Syria. They watched
YouTube videos.''
He continues: ``The dark side of the open Internet is that truly
fringe and harmful ideas now are mainstream, or at least accessible to
7\1/2\ billion people.'' Yet ``whenever we have one of these attacks,
we just wring our hands for a few weeks and then wait for the next one
to happen.''
Social networks have created ``a new environment for radicalization
and recruitment,'' says David Ibsen, executive director of the Counter
Extremism Project, a nonprofit research and advocacy organization to
which Mr. Farid is a senior adviser. Terror groups weaponized Twitter,
Google, Facebook and other forums to plan or encourage violence; to
discover the vulnerable or disaffected; and to publish professional,
sophisticated and carefully presented propaganda.
Islamic State is basically a digital-first media startup. (By
comparison, al Qaeda was MySpace.) ISIS content is beamed out globally
and becomes refractory across the viral web. Some videos show vignettes
of ISIS bureaucrats delivering social services or its fighters talking
about the battle between belief and unbelief. Others are more savage--
beheadings, stonings, drownings, other torture and combat operations.
Mr. Farid slipped into this world slant-wise. He's a founder of the
computer-science field known as digital forensics. In the late 1990s as
a postdoctoral researcher, he was among the first to recognize that
mathematical and computational techniques to authenticate digital
images and other media would be useful to society.
Because images so powerfully change what we are willing to believe,
the modern era requires a scientific method to ensure we can trust
them. How can we prove, for example, that digital photographs aren't
forgeries so they are admissible as evidence in court? Images are
increasingly important in cellular, molecular and neurological
medicine, Mr. Farid notes, and tampering has led to more than one
research-and-retraction scandal. Unscrupulous stringers sometimes file
doctored photos with news organizations, and unscrupulous motorists
sometimes photoshop pictures to exaggerate fender-benders for insurance
claims.
Mr. Farid explains how image authentication works: ``We think about
how light interacts in the physical world; what happens when that light
hits the front of the lens and gets focused and goes through an optical
train; what happens when that light hits an electronic sensor and gets
converted from an analog to a digital and then goes through a
postprocessing and gets saved as .jpeg and then gets posted on
Facebook.'' By identifying ``statistical and geometrical and physical
regularities'' in this life cycle, software can search for
inconsistencies to expose manipulation.
In 2008 this research pulled Mr. Farid into another underworld--
child pornography. In 2002 the U.S. Supreme Court struck down a ban on
``virtual'' child porn--computer-generated images that ``appear to
depict minors but were produced without using any real children.'' Mr.
Farid is sometimes brought in as an outside expert when a defendant
claims the material at issue is virtual.
The child-porn industry was nearly defunct by the 1990s, because
negatives and videotapes can be confiscated and destroyed. ``Then the
Internet came,'' Mr. Farid says, ``and all hell broke loose.''
Supply can create its own demand. Much like jihadists, deviants
formed a global community, finding each other online and sharing what
are really crime-scene photos. Like ISIS agitprop, material is
continuously copied, cut, spliced, resized, recompressed and otherwise
changed, in part to evade detection as it is retransmitted again and
again.
Mr. Farid worked with Microsoft to solve both problems--detection
and replication. He coded a tool called Photo DNA that uses ``robust
hashing'' to sweep for child porn. ``The hashing part is that you reach
into a digital image and extract a unique signature. The robust part is
if that image undergoes simple changes, the fingerprint shouldn't
change. When you change your clothes, cut your hair, as you age, your
DNA stays constant,'' he says. ``That's what you want from this
distinct fingerprint.''
The algorithm matches against a registry of known illegal
signatures, or hashes, to find and delete photographs, audio and video.
Photo DNA is engineered to work at ``internet scale,'' says Mr. Farid,
meaning it can process billions of uploads a day in microseconds with a
low false-positive rate and little human intervention.
Monitoring by Photo DNA, which is licensed by Microsoft at no cost
and now used in most networks, revealed that the nature of the problem
was ``not what we thought it was,'' says Ernie Allen, the retired head
of the National Center for Missing and Exploited Children. Child
pornography was far more widely circulated than law enforcement
believed. ``Hany Farid changed the world,'' Mr. Allen adds. ``His
innovation rescued or touched the lives of thousand of kids, and
uncovered perpetrators, and prevented terrible revictimization as
content was constantly redistributed.''
Mr. Farid linked up with the Counter Extremism Project to apply the
same robust-hashing method to extremist propaganda. But this effort has
encountered resistance. ``The pushback from the tech companies has been
pretty strong,'' the project's Mr. Ibsen says dryly.
U.S. law immunizes Internet companies from criminal and civil
liability for content that travels over their transoms. Their terms of
service forbid abusive content, but they rely on users instead of
algorithms to police violations. ``It's a very slow and tedious
process: You wait for it to get reported, somebody has to review it,
they make mistakes,'' Mr. Farid says. ``They take down the Vietnam
napalm girl on Facebook.''
Liability aside, what about their moral obligations to help prevent
death, injury and destruction? ``In my mind, we're not asking them even
to do something that they haven't said they want to do already. We're
saying, hey, would you please do the thing that you promised you would
do?'' he explains. ``I am simply saying, look, for free, you can
automate this and make it really efficient and really fast and save you
money on the side.''
But the ``ethos'' of Silicon Valley doesn't include becoming the
censors of the internet, and tech firms fear a slippery slope. ``The
concern they have is, OK, first they came for the child porn, then they
came for the extremism, next they're going to take the kitten videos,''
Mr. Farid says. ``I think that's a bit of a hysterical leap. We are
talking about content with very clear and well-defined harm. These are
not abstract notions--`I don't want people to be mean to me.' We're not
talking about bullying. We are talking about things with very immediate
consequences and very real harm.''
One question is how to distinguish support for terrorism from the
merely inappropriate or objectionable. What about Islamic State's
black-flag brand, or a declaration of a caliphate, or the sermons of
Anwar al-Awlaki? Maybe you know it when you see it.
``Is an ISIS fighter saying `Death to the West' extremism? I don't
know. I don't want to have that conversation,'' Mr. Farid replies.
``I'm talking about explicit acts of violence, explicit calls to
violence, explicit glorification of violence, depravity, the worst of
the worst of the worst.''
His point is that tech companies can make judgment calls about the
middle ground, wherever it might be, for themselves: ``You decide: Yes,
no, yes, no, yes, no, and then we'll build a cache and eliminate that
content from your networks.''
Mr. Farid concedes that there are dangers: ``This type of
technology is agnostic in what it's looking for. It can be used in ways
we would not approve of, such as stifling speech. You can't deny that.
This is what we've learned about technology over the years--it can be
used for good and for bad. Social media platforms can be good and
bad.''
There has been some progress. Twitter has deleted hundreds of
thousands of handles associated with terrorism since 2015, and late
last year Twitter, Facebook, Microsoft and YouTube announced an
industry antiterror consortium. But Mr. Farid's robust hashing remains
a hard sell.
The irony is that algorithms increasingly govern the world.
Networks are perpetually scanned for spam, malware, viruses; Google
reads your e-mail to target ads; credit-card companies monitor your
financial transactions to prevent fraud. Facebook's Mark Zuckerberg
even promises to use algorithms to distinguish truth from falsehood. As
a scholar of the differences between the two, Mr. Farid has a few
thoughts.
In the backwash of 2016, Mr. Zuckerberg published a 5,800-word
manifesto that promised Facebook's artificial intelligence would soon
learn to sort real news from hoaxes and misinformation, break up
``filter bubbles,'' and draw a line between free speech and suborning
terror. The goal, he wrote, is to preserve ``our shared sense of
reality.''
Mr. Farid is a skeptic: ``As somebody who worked for a long time in
this space, I think he's underestimating what a hard problem this is.''
Mr. Zuckerberg ``paints this picture like machine learning is going to
be fully automatic--basically you'll be able to set criteria on your
page, ``I don't want to see violence, I don't want to see bad words,''
and it'll just work.
``Even as a technologist, and despite all the advance of
technology, the human brain is astonishing at what it does. Our ability
to make these types of assessments that are really hard for these AI
algorithms is humbling. I don't think we'll get it in the next five or
10 years.''
Meantime, Mr. Farid has developed a technology that could work
today to contain a growing threat. While we await the Facebook utopia,
perhaps our digital lives--and our real lives--would be healthier if it
were widely deployed.
Mr. Rago is a member of The Wall Street Journal's editorial board.
Appeared in the March 11, 2017, print edition.
______
Response to Written Questions Submitted by Hon. Jerry Moran to
Monika Bickert
Question 1. Your written testimony emphasized the importance of the
credibility of the speaker as it relates to Facebook's efforts to
prevent recruitment through ``counterspeech.'' How have your strategic
partnerships with non-governmental organization and community groups
bolstered Facebook's ``counterspeech'' efforts?
Answer. We believe that a key part of combating extremism is
preventing recruitment by disrupting the underlying ideologies that
drive people to commit acts of violence. That's why we support a
variety of counterspeech efforts. Although counterspeech comes in many
forms, at its core it includes efforts to prevent people from pursuing
a hate-filled, violent life or convincing them to abandon such a life.
Our efforts are focused on empowering counterspeech creators and
amplifying local voices by building awareness, educating communities,
encouraging cohesion, and directly countering hateful narratives. We
have partnered with non-governmental organizations and community groups
around the world to empower positive and moderate voices. For example,
in the U.S., we have worked with EdVenture Partners to develop a peer-
to-peer student competition called the Facebook Global Digital
Challenge (P2P). This is a semester-long university course during which
students build a campaign to combat extremism in their area, launch it,
track its success, and then submit the results as part of a global
competition. As part of P2P, a team of communications students from the
University of Central Oklahoma ran an amazing program called uDefy that
reached over one million people in 85 countries using Facebook and
other social media platforms. The team behind uDefy encouraged
participants to recognize and challenge their own beliefs and
stereotypes by taking a four-step pledge: (1) face your truth; (2) get
the facts; (3) commit to defy; and (4) spread the word. The goal of the
campaign is to channel fear and misconception into truth and
understanding one individual at a time. Those who complete the four-
step pledge become uDefy ambassadors and take the campaign back to
their own campuses. In less than three years, these P2P projects have
reached more than 56 million people worldwide through more than 500
anti-hate and extremism campaigns created by more than 5,500 university
students in 68 countries.
We have also partnered with the Institute for Strategic Dialogue to
launch the Online Civil Courage Initiative, a project that has engaged
with more than 100 anti-hate and anti-extremism organizations across
Europe. Similarly, we work with Affinis Labs to host hackathons in
places like Manila, Dhaka, and Jakarta, where community leaders joined
forces with tech entrepreneurs to develop innovative solutions to
challenge extremism and hate online.
By fanning out and removing content, and supporting counterspeech
efforts, we can limit the audience and distribution of terrorist
propaganda.
Question 2. Your written testimony stated: ``In the first half of
2017, [Facebook] provided information in response to more than 75
percent of the 1,864 requests for emergency disclosures that [the
company] received from U.S. law enforcement agencies.'' Do you have a
company policy when deciding how to respond to the 1,864 requests for
emergency disclosures?
(a) Does Facebook do their own assessment as to whether the content
constitutes an emergency?
Answer. As part of official investigations, government officials
sometimes request data about people who use Facebook. We disclose
account records in accordance with our terms of service and applicable
law, and we may voluntarily disclose information to law enforcement
where we have a good faith reason to believe that the matter involves
imminent risk of serious physical injury or death. We have strict
processes in place to handle these government requests. We require
officials to provide a detailed description of the legal and factual
basis for their request, and we push back if the request appears to be
legally deficient or is overly broad, vague, or otherwise inconsistent
with our policies. More information about the requests we have received
from governments around the world can be found at https://
transparency.facebook.com/.
(b) Do your internal policies account for the 25 percent of requests
that are not responded to with information?
Answer. Please see the response to question 2a.
(c) Do you have the resources to deal with these requests?
Answer. Our Law Enforcement Response Team works hard to respond to
legitimate law enforcement requests while fulfilling our responsibility
to protect people's privacy and security. We have a global team that
strives to respond within minutes to emergency requests from law
enforcement. Our effort to make our platform safer and more secure is a
holistic one that involves a continual evaluation of our personnel,
processes and policies, and we make changes as appropriate.
______
Response to Written Questions Submitted by Hon. Ron Johnson to
Monika Bickert
Question 1. Social media companies are increasingly able to remove
terrorist recruitment, incitement, and training materials before it
posts to their platforms by relying on improved automated systems.
Other than content removal, what else can be done to limit the audience
or distribution of these dangerous materials?
Answer. When we find an account that is associated with terrorism,
we use artificial intelligence to identify and remove related material
that may also support terrorism or terrorists. As part of that process,
we utilize a variety of signals, including whether an account is
``friends'' with a high number of accounts that have been disabled for
terrorism, or whether an account shares the same attributes as a
disabled account.
Moreover, we believe that a key part of combating extremism is
preventing recruitment by disrupting the underlying ideologies that
drive people to commit acts of violence. That's why we support a
variety of counterspeech efforts. Although counterspeech comes in many
forms, at its core these are efforts to prevent people from pursuing a
hate-filled, violent life or convincing them to abandon such a life. We
have partnered with non-governmental organizations and community groups
around the world to empower positive and moderate voices. For example,
in the U.S., we have worked with EdVenture Partners to develop a peer-
to-peer student competition called the Facebook Global Digital
Challenge (P2P). This is a semester-long university course during which
students build a campaign to combat extremism in their area, launch it,
track its success, and then submit the results as part of a global
competition. As part of P2P, a team of communications students from the
University of Central Oklahoma ran an amazing program called uDefy that
reached over one million people in 85 countries using Facebook and
other social media platforms. The team behind uDefy encouraged
participants to recognize and challenge their own beliefs and
stereotypes by taking a four-step pledge: (1) face your truth; (2) get
the facts; (3) commit to defy; and (4) spread the word. The goal of the
campaign is to channel fear and misconception into truth and
understanding one individual at a time. Those who complete the four-
step pledge become uDefy ambassadors and take the campaign back to
their own campuses. In less than three years, these P2P projects have
reached more than 56 million people worldwide through more than 500
anti-hate and extremism campaigns created by more than 5,500 university
students in 68 countries.
We have also partnered with the Institute for Strategic Dialogue to
launch the Online Civil Courage Initiative, a project that has engaged
with more than 100 anti-hate and anti-extremism organizations across
Europe. Similarly, we work with Affinis Labs to host hackathons in
places like Manila, Dhaka, and Jakarta, where community leaders joined
forces with tech entrepreneurs to develop innovative solutions to
challenge extremism and hate online.
By fanning out and removing content, and supporting counterspeech
efforts, we can limit the audience and distribution of terrorist
propaganda.
Question 2. Terrorist how-to guides are protected by the First
Amendment in the United States, but violate the content policies of
many social media companies as well as the laws of some international
partner nations. What countries have laws that go beyond your company's
content policies and can you give examples of how you have worked with
those countries to de-conflict those differences?
Answer. A number of countries around the world have laws that limit
content that might otherwise be allowed by our Community Standards or
U.S. law. In Germany, for example, laws forbid incitement to hatred. In
the U.S., on the other hand, even the most vile speech may be legally
protected under the U.S. Constitution. There are times when we may have
to remove or restrict access to content because it violates a law in a
particular country, even though it does not violate our Community
Standards. Further, when governments believe that something on the
Internet violates their laws, they may contact companies like Facebook
and ask us to restrict access to that content. When we receive such a
request, it is scrutinized to determine if the specified content does
indeed violate local laws. If we determine that it does, then we make
it unavailable in the relevant country or territory. For example,
Holocaust denial is illegal in Germany, so if it is reported to us, we
will restrict this content for people in Germany.
Question 3. The long-term business interests of social media
platforms are aligned with the public safety concerns of this
committee: users want to feel safe while engaging with the online
community. To this end, Facebook is developing a way to identify users
at higher risk of suicide and urgently pass posts from any user in
danger to a community operations team, as well as provide that user
with a menu of options to reach out to their own friends or other
suicide prevention partners. Is Facebook developing any similar tool to
identify users at higher risk of terrorist activity? If so, what off-
ramp options would Facebook consider offering those users?
Answer. We are using similar automated tools to identify users who
are posting content that violates our policies against terrorism,
including promoting terror groups, sharing their propaganda, and
planning or coordinating violence. We reach out to law enforcement
whenever we see a credible threat of imminent harm.
We are eager to partner with government and civil society to
develop off-ramp options for users at a higher risk of terrorist
activity. A critical part of providing an off-ramp is being able to
link people to appropriate, effective, and responsible services. We are
exploring ways of partnering with such services.
______
Response to Written Questions Submitted by Hon. Maria Cantwell to
Monika Bickert
Question 1. Facebook prohibits individuals and organizations that
promote hate from having a presence on its site. However, I think we
can do much more to promptly communicate threats of hate-based violence
to the relevant law enforcement agencies and internally police hate-
promoting individuals and organizations. Will you commit Facebook to
exploring and implementing a more aggressive effort to report hateful
images and threats to law enforcement? If not, can you explain why you
would not commit to this important request?
Answer. Facebook is opposed to hate speech in all its forms, and we
are committed to removing it from our platform any time we become aware
of it. We carefully review reports that we receive from the public,
media, civil society, and governments, and we remove content that
violates our policies. We are committed to improving our approach to
addressing these issues, and regularly evaluate our hate speech
policies to determine whether they need to be updated. We are also
working to enhance our review process so that we are able to respond
quickly and accurately to community reporting. We also remove credible
threats of physical harm to individuals and specific threats of theft,
vandalism, or other financial harm. We have a long history of working
successfully with law enforcement to address a wide variety of threats
to our platform, and we work with law enforcement when we believe there
is a genuine risk of physical harm or direct threats to public safety.
Our effort to make our platform safer and more secure is a holistic one
that involves a continual evaluation of our personnel, processes, and
policies, and we make changes as appropriate.
Question 2. We have strong principles of freedom of speech, but at
the same time, we need to balance that freedom with the need to protect
against bad actors who would leverage that freedom to plan and promote
illegal acts. How can we use artificial intelligence to help us achieve
a balance between our American ideal of free speech and the need to
protect against extremist acts of terror?
Answer. We already use artificial intelligence (AI) to help us
identify threats of real world harm from terrorists and others. We
reach out to law enforcement whenever we see a credible threat of
imminent harm. The use of AI and other automation to stop the spread of
terrorist content is showing promise. Today, 99 percent of the ISIS and
Al Qaeda-related terror content we remove from Facebook is content we
detect before anyone in our community has flagged it to us, and in some
cases, before it goes live on the site. We do this primarily through
the use of automated systems like photo and video matching and text-
based machine learning. Once we are aware of a piece of terror content,
we remove 83 percent of subsequently uploaded copies within one hour of
upload.
We believe technology can be part of the fight against terrorism.
But deploying AI for counterterrorism is not as simple as flipping a
switch. For example, a photo of an armed man waving an ISIS flag might
be propaganda or recruiting material, but could be an image in a news
story. Ultimately, the use of AI must be reinforced with manual review
from trained experts. To that end, we tap expertise from inside the
company and from outside, partnering with those who can help address
extremism across the internet.
Question 3. Outside of artificial intelligence, what other
technologies could be used to combat potential radicalization on social
media platforms? What does the implementation of those technologies
look like?
Answer. We are constantly updating our technical solutions, but our
current efforts include image matching technology and language
understanding. When someone tries to upload a terrorist photo or video,
our systems look for whether the image matches a known terrorist photo
or video. This means that if we previously removed an ISIS propaganda
video, for example, we can work to prevent other accounts from
uploading the same video to our site. We also have started
experimenting with using AI to understand text that potentially
advocates for terrorism. We are working to develop text-based signals
to detect praise or support of terrorist organizations. These signals
will be incorporated into an algorithm that is in the early stages of
learning how to detect similar posts.
We understand that simply working to keep terrorism off Facebook is
an inadequate solution to the problem of online extremism, particularly
because terrorists are able to leverage a variety of platforms. We
believe our partnerships with others--including other companies, civil
society, researchers, and governments--are crucial to combating this
threat. To this end, we have partnered with our industry counterparts
to more quickly identify and slow the spread of terrorist content
online. For example, in December 2016, we joined with Microsoft,
Twitter, and YouTube to announce the development of a shared industry
database of ``hashes''--unique digital fingerprints for photos and
videos--for content produced by or in support of terrorist
organizations. The database now contains more than 60,000 hashes, and
the consortium has grown to include thirteen companies.
We believe that computer algorithms and machine learning are
necessary but not sufficient to address these problems. That's why we
are also using specialized human review, industry cooperation, and
counter-speech training. We will also be doubling the number of people
who work on safety and security at Facebook by the end of this year--
from 10,000 to 20,000 people.
______
Response to Written Questions Submitted by Hon. Richard Blumenthal to
Monika Bickert
Question 1. On October 30, 2017, nineteen civil rights groups,
including Muslim Advocates, Leadership Conference on Civil and Human
Rights, NAACP, Southern Poverty Law Center, and many others, co-signed
a letter to Facebook to express concern about the hateful content on
the social media platform used to divide the country, and in
particular, to promote anti-Muslim, anti-Black, anti-immigrant, and
anti-LGBTQ animus.
Ms. Bickert, please provide a copy of Facebook's response to this
letter.
Answer. Hate speech and discriminatory advertising have no place on
our platform, and we remove such content as soon as we become aware of
it. We also have partnerships with academics and experts who study
organized hate groups and who share information with Facebook on how
organizations are adapting to social media and give feedback on how
Facebook might better tackle these problems. We have reached out to the
groups referenced in the question and are in a dialogue with them,
which has included in-person conversations. We are committed to
continuing our dialogue with them and other third parties to ensure
that our users feel welcome and safe on our platform.
Question 2. Facebook reports that 99 percent of the ISIS and Al
Qaeda-related terror content is detected and removed before it is even
flagged on Facebook. However, a recent investigation by ProPublica
asked Facebook about its handling of forty-nine posts that they thought
might be deemed offensive Facebook's own Community Standards. Facebook
acknowledged and apologized that its content reviewers made the wrong
call in almost half of them.
Ms. Bickert, what steps is Facebook taking to better counter the
use of the platform to spread hateful information and improve the work
of its content reviewers?
Answer. We don't allow hate speech, which we define as anything
that directly attacks people based on race, ethnicity, national origin,
religious affiliation, sexual orientation, sex, gender, gender
identity, or serious disability or disease. However, our policies allow
content that may be controversial and at times even distasteful, but
which does not cross the line into hate speech. This may include
criticism of public figures, religions, professions, and political
ideologies. Our challenge is identifying hate speech across different
cultures, languages, and circumstances for a community of more than 2
billion people.
Nudity and violence, for example, are fairly easy to spot, but hate
speech is often determined by its context. Because of these nuances, we
cannot rely on machine-learning or AI to the same degree that we do
with other types of content like nudity. Technology can help flag the
most blatantly reprehensible language. But it cannot yet understand the
context necessary to assess what is or is not hate speech--though we
are working on tools to help us improve the accuracy of our enforcement
and building new AI to better detect bad content.
We encourage people to report posts and rely on our team of content
reviewers around the world to review reported content. Our reviewers
are trained to look for violations and enforce our policies
consistently and as objectively as possible. We have weekly quality
audits of each reviewer, during which we re-review a subset of their
work and address any mistakes made. We receive millions of reports of
possible content violations every week, so we know that we will
unfortunately make many mistakes even if we maintain an accuracy rate
of 99 percent. We are always working to make our platform safer and
more secure through, among other things, continually evaluating our
processes, policies, and training. Enforcement is never perfect, but we
will get better at finding and removing improper content.
Question 3. As recently as last year, Facebook reportedly allowed
offensive claims that specify a sub-group within a protected class,
such as ``black children'' or ``female drivers'' but would ban attacks
aimed at entire groups, such as ``white men.'' It seems that such
policies could easily be gamed to work-around Facebook Community
Standards.
Ms. Bickert, Is that still the policy of Facebook? Can you explain
the nuances in the new policy and how Facebook is working to make sure
such mistakes don't happen again?
Do you think it would be helpful for Facebook to be more
transparent about how it applies its standards, or bring in an
independent third-party and work with civil rights groups to help it
evaluate its current policies?
Answer. No, this is not our policy. Facebook is opposed to hate
speech in all its forms, and that includes removing content that
targets any of the three groups identified in the question, depending
on the context of the post.
We currently define hate speech as anything that directly attacks
people based on protected characteristics--race, ethnicity, national
origin, religious affiliation, sexual orientation, sex, gender, gender
identity, or serious disability or disease. Such content violates our
Community Standards and will be removed. This includes, for example,
content that attacks ``black children'' or ``white men.'' However,
there may be other content that is controversial or distasteful, but
does not cross the line into hate speech.
We are constantly evaluating--and, where necessary, changing--our
content policies to account for shifts in cultural and social norms
around the world. For example, we recently updated our hate speech
policies to remove violent speech directed at groups of people defined
by protected characteristics, even if the basis for the attack may be
ambiguous. Under the previous hate speech policy, a direct attack
targeting women on the basis of gender, for example, would have been
removed from Facebook, but the same content directed at women drivers
would have remained on the platform. We have come to see that this
distinction is a mistake, and we no longer differentiate between the
two forms of attack when it comes to the most violent hate speech. For
instance, we would now remove a comment that dehumanized ``female
drivers'' by comparing them to animals. We continue to explore how we
can adopt a more granular approach to hate speech, both in the way we
draft our policies and the way we enforce on them.
Question 4. Ms. Bickert, Mr. Monje, and Ms. Downs, please provide
copies (including images, text, dates and timestamps) of all content
identified by your platforms as generated by Russian agents or the
Internet Research Agency.
Answer. We have provided this information to the Senate Select
Committee on Intelligence and the Senate Judiciary Committee and
believe you should have access through those committees.
Question 5. Advocates for preventing gun violence have long been
concerned about the ``private sale'' loophole, which allows individuals
to purchase guns without a background check. So, Facebook's
announcement in January 2016 that it would ban the private sale of guns
and ammo on its site and Facebook was met with great applause.
Unfortunately, it soon became evident that new rules have done little,
if anything, to stop the flow of guns on the social network. If we are
serious about fighting terrorism online, we should be just as serious
about really closing this dangerous loophole, which could very well
enable violent ideology to be translated into horrific acts.
Ms. Bickert, do you agree that successfully closing this loophole
is important?
How would you compare the amount of resources devoted to combatting
terrorism online to the amount of resources devoted to ensuring
Facebook's prohibition on the private sale of guns and ammo?
In what way is Facebook making sure it applies any relevant
technologies, tools, and human resources used to review hate speech to
also enforce Facebook's prohibitions on the private sale of guns and
ammo?
Answer. We do not allow firearm sales on Facebook, and any time we
become aware of content that is facilitating gun sales, we remove it.
We allow our users to report such activity. We also look at associated
groups and accounts by ``fanning out'' to identify and remove other
content that may violate our policies. We will continue to look for
ways to get faster at finding and removing violating content, and we
encourage our community to continue to tell us if they see this
behavior anywhere on our platform.
Question 6. At least one of your peers in the tech industry has
voluntarily initiated an outside assessment of the civil rights impacts
of its policies and programs. In response to concerns regarding
discrimination on the home-sharing platform, AirBNB hired former U.S.
attorney general Eric Holder to help craft an anti-discrimination
policy and has promised to pursue technological innovations to guard
against future discriminatory events.
Mr. Monje, Ms. Bickert, and Ms. Downs, can you each commit to
bringing in an independent entity to conduct a thorough and public
audit of the civil rights impact of your policies and programs,
including how your platform has been used by hate groups to stoke
religious resentment and violence?
Answer. Hate speech and discriminatory advertising have no place on
our platform. Our Community Standards prohibit attacks based on
protected characteristics, including religion, and we prohibit
advertisers from discriminating against people based on religion and
other attributes. Facebook has partnerships with academics and experts
who study organized hate groups and hate speech. These academics and
experts share information with Facebook on how organizations are
adapting to social media and give feedback on how Facebook might better
tackle these problems. We recently hosted several of these academics at
Facebook for multiple days of observation and assessment, during which
the academics attended substantive meetings on our content policies and
the guidance we provide to our reviewers. Further, in the area of hate
speech, there are very important academic projects that we follow
closely. Timothy Garton Ash, for example, has created the Free Speech
Debate to look at these issues on a cross-cultural basis. Susan Benesch
established the Dangerous Speech Project, which investigates the
connection between speech and violence. These projects show how much
work is left to be done in defining the boundaries of speech online,
which is why we will keep participating in this work to help inform our
policies at Facebook. We are committed to continuing our dialogue with
third parties to ensure that our users feel welcome and safe on our
platform.
Question 7. A little over a year ago, Facebook, Twitter, Google,
and Microsoft announced a plan to create a joint industry database of
``content that promotes terrorism.''
Mr. Monje, Ms. Bickert, and Ms. Downs, to what extent does this
joint industry database focus on all forms of terror, including the
real terror threat presented by white supremacists?
Answer. At last year's EU Internet Forum, Facebook, Microsoft,
Twitter, and YouTube declared our joint determination to curb the
spread of terrorist content online. Over the past year, we have
formalized this partnership with the launch of the Global Internet
Forum to Counter Terrorism (GIFCT). The GIFCT is committed to working
on technological solutions to help thwart terrorists' use of our
services, including through a shared industry hash database, where
companies can create ``digital fingerprints'' for terrorist content and
share it with participating companies. The database, which became
operational in the spring of 2017, now contains more than 60,000
hashes. It allows the thirteen member companies to use those hashes to
identify and remove matching content--videos and images--that violate
our respective policies or, in some cases, block terrorist content
before it is even posted. Each company has different policies,
practices, and definitions as they relate to terrorist content. If
content is removed from a company's platform for violating that
platform's individual terrorism-related content policies, the company
may choose to hash the content and include it in the database.
Facebook's policies do prohibit all forms of terror, including
threats by white supremacist organizations.
Question 8. As reported by CNN last August after the events in
Charlottesville, only 58 of over 200 Southern Poverty Law Center-
designated hate groups with Facebook accounts had been suspended for
their hateful actions and rhetoric.
Ms. Bickert, how many of those hate groups with Facebook accounts
are now blocked? What further steps are you taking to further review
the actions and accounts of these groups?
Answer. Facebook is opposed to hate speech in all its forms.
Facebook has partnerships with academics and experts who study
organized hate groups, including the Southern Poverty Law Center. These
academics and experts share information with Facebook on how
organizations are adapting to social media and give feedback on how
Facebook might better tackle these problems. That said, we apply our
own policies about what constitutes hate speech, and our definition of
hate speech may differ from others, including those with whom we
partner. We are constantly evaluating--and, where necessary, changing--
our content policies, and we currently define hate speech as anything
that directly attacks people based on protected characteristics--race,
ethnicity, national origin, religious affiliation, sexual orientation,
sex, gender, gender identity, or serious disability or disease. Such
content violates our Community Standards and will be removed. However,
there may be content that is controversial or distasteful, but does not
cross the line into hate speech.
Further, our own content policy team includes subject matter
experts who are focused on staying ahead of trends in hate speech.
Their work is used to inform our Community Operations team, which
reviews content that our users and automated tools flag as
inappropriate, dangerous, abusive, or otherwise violating our
policies--including our hate speech policy.
Managing a global community in this manner has never been done
before, and we know we have a lot more work to do. We are committed to
improving and to ensuring that hate has no place on Facebook.
______
Response to Written Questions Submitted by Hon. Brian Schatz to
Monika Bickert
Question 1. Please quantify and explain Facebook's progress in
tackling the fake-user account issue. For the most recent full month
available and every month in the two years preceding provide:
number of fake accounts created
number of fake accounts removed
number of accounts hacked
number of hacked accounts restored
number of duplicate accounts created
number of duplicate accounts removed
number of inactive accounts existing
number of inactive accounts removed
number of monthly active users
average number of days a fake account remains on the
platform
Please provide the numbers above for Instagram as well.
Answer. Facebook regularly provides information on the number of
monthly active users (MAUs), false accounts, and duplicate accounts in
its filings with the Securities and Exchange Commission. We define an
MAU as a registered Facebook user who logged in and visited Facebook
through our website or a mobile device, or used our Messenger
application (and is also a registered Facebook user), in the last 30
days as of the date of measurement. MAUs are a measure of the size of
our global active user community. As of December 31, 2017, we had 2.13
billion MAUs, an increase of fourteen percent from December 31, 2016.
We regularly evaluate these metrics to estimate the number of
``duplicate'' and ``false'' accounts among our MAUs. A duplicate
account is one that a user maintains in addition to his or her
principal account. We divide ``false'' accounts into two categories:
(1) user-misclassified accounts, where users have created personal
profiles for a business, organization, or non-human entity such as a
pet (such entities are permitted on Facebook using a Page rather than a
personal profile under our terms of service); and (2) undesirable
accounts, which represent user profiles that we determine are intended
to be used for purposes that violate our terms of service, such as
spamming. The estimates of duplicate and false accounts are based on an
internal review of a limited sample of accounts, and we apply
significant judgment in making this determination. In the fourth
quarter of 2017, we estimate that duplicate accounts may have
represented approximately ten percent of our worldwide MAUs. We believe
the percentage of duplicate accounts is meaningfully higher in
developing markets such as India, Indonesia, and the Philippines, as
compared to more developed markets. In the fourth quarter of 2017, we
estimate that false accounts may have represented approximately three
to four percent of our worldwide MAUs. Our estimation of false accounts
can vary as a result of episodic spikes in the creation of such
accounts. Additional information relating to our estimate of false
accounts is included in our filings with the SEC. We do not maintain
public statistics on the other types of accounts that are referenced in
the question.
We continue to make improvements to our efforts to more effectively
detect and deactivate fake accounts to help reduce the spread of spam,
false news, and misinformation. We continually update our technical
systems to identify, checkpoint, and remove inauthentic accounts, and
we block millions of attempts to register fake accounts every day.
These systems examine thousands of detailed account attributes and
prioritize signals that are more difficult for bad actors to disguise,
such as their connections to others on our platform. As with all
security threats, we have been incorporating new insights into our
models for detecting fake accounts, including information specific to
election issues.
Question 2. How does a user find out if they are being impersonated
on Facebook or Instagram? Do Facebook and Instagram notify users
proactively? Or are users expected to monitor the platforms and report
to the company?
Answer. Claiming to be another person violates our Community
Standards, and we want to make it harder for anyone to be impersonated
on our platform. Users can also report accounts that are impersonating
them. We've developed several techniques to help detect and block this
type of abuse. At the time someone receives a friend request, our
systems are designed to check whether the recipient already has a
friend with the same name, along with a variety of other factors that
help us determine if an interaction is legitimate. Further, we recently
announced new features that use face recognition technology that may
help people learn when someone is using their image as a profile
photo--which can help stop impersonation. This is an area we're
continually working to improve so that we can provide a safe and secure
experience on Facebook.
Question 3. What are the average numbers of days or hours that
Facebook and Instagram take to investigate impersonation complaints
before they are resolved?
Answer. We promptly respond to reports of imposter accounts.
Sometimes, these investigations are complex and require, for example,
that users upload identification to confirm their identities. In
general, the majority of all types of complaints received on Facebook
are reviewed within 24 hours.
Question 4. Do Facebook and Instagram have a separate, expedited
process for resolving impersonation of minors' accounts?
Answer. We take the issue of safety on our platform very seriously,
especially that of our teen users. We want people to connect and share
on Facebook, and it's integral that they feel safe in order to do so.
We do not tolerate impersonation in any way and we remove profiles that
impersonate other people. We have developed several techniques to help
detect and block this type of abuse. At the time someone receives a
friend request, for example, our systems are designed to check whether
the recipient already has a friend with the same name, along with a
variety of other factors that help us determine if an interaction is
legitimate. It's an area we're continually working to improve so that
we can provide a safe and secure experience.
Question 5. According to the current best estimate, approximately
126 million people on Facebook may have been served some piece of
content associated with the Internet Research Agency (IRA) between
January 2015 and August 2017. How many Instagram users were also served
IRA content during the same time period? What was the methodology
behind these estimates?
Answer. Using data analysis and modeling, we found that 11.4
million people in the United States saw at least one of the ads
associated with the IRA between 2015 and 2017, and that as many as 126
million people in the United States may have seen a piece of IRA
content on Facebook. Our data related to Instagram is incomplete, but
we believe that as many as 16 million additional people who did not see
this content on Facebook saw IRA content on Instagram starting in
October 2016.
Question 6. How well did IRA content perform on Facebook and
Instagram? Please provide metrics commonly measured on the platforms
and benchmark against industry standards. This includes but is not
limited to engagement (i.e., time on post), reactions, impressions, and
referral traffic for organic content. For ads, please provide the
click-through rate and cost per reaction or reach.
Answer. As noted above, we found that 11.4 million people in the
United States saw at least one of the ads associated with the IRA
between 2015 and 2017, and that as many as 126 million people in the
United States may have seen a piece of IRA content on Facebook. Forty-
four percent of total ad impressions were before the U.S. election, and
56 percent of total ad impressions were after the election. Roughly 25
percent of the ads were never shown to anyone. That's because
advertising auctions are designed so that ads reach people based on
relevance, and certain ads may not reach anyone as a result. Our data
related to Instagram is incomplete, but we believe that as many as 16
million additional people who did not see this content on Facebook saw
IRA content on Instagram starting in October 2016.
Question 7. Has Facebook shared the content, data, and metadata
associated IRA activity above with researchers who are also looking
into this? With law enforcement? With other companies? If not, then why
not?
Answer. Facebook is providing investigators, including
congressional committees, with information it has regarding the scope
and nature of Russian information operations on our platform that may
be relevant to their inquiries. We have also been working with many
others in the technology industry on these issues.
Question 8. Facebook has several advertising tools and properties--
including Facebook Events, Facebook Audience Network, and Facebook
Canvas. Can you list all of them (including those for advertising on
Instagram), and a succinct summary of what each of them do? Which of
these were used by the IRA?
Answer. The Facebook family supports multiple advertising types.
Each ad has two components: the format (what the ad looks like) and the
placement (where it will be displayed). Ads can be placed on Facebook,
Instagram, Messenger and Audience Network, which allows ads to be
delivered on apps and sites beyond Facebook. Depending on where it is
placed, available formats may include video; image; Collection, or
displays of items from a product catalog; Carousel, or multiple image
or videos within an ad; Slideshow, or video ads that can be seen at
slower connection speeds; Canvas, or full-screen ads on mobile devices;
lead generation ads, which allow advertisers to collect information
from people interested in their business; offer ads that businesses can
use to share discounts on their products; Post ads, which allows
advertisers to have their Page posts appear beyond their Pages; ads for
events; and ads for Page likes. The IRA generally used Page Likes or
Page Post ads, typically with still images. The IRA also created some
ads to promote events.
Facebook also offers three primary types of targeting, or
audiences. Core Audiences are traditional targeting options based on
location, demographics (age, gender education, job status, and more),
interests, behavior, and connections. Custom Audiences are groups of
specific people, like an advertiser's own contacts (Customer File
Custom Audiences), visitors to an advertiser's website or app (Website
Traffic Custom Audiences), or people who have engaged with an
advertiser's content on Facebook services (Engagement Custom
Audiences). Finally, Facebook offers Lookalike Audiences, which enables
advertisers to find Facebook users that have similar characteristics to
another audience. The targeting for the IRA ads that we have identified
was relatively rudimentary, targeting broad locations and interests,
and did not use Customer File Custom Audiences or Customer File
Lookalike Audiences.
Question 9. In 2016, accounts affiliated with RT and Sputnik spent
$5.4 million on Facebook advertising. How much did the same accounts
spend on Facebook advertising in 2017? Does this include Instagram? How
well did they perform? Again, please provide metrics commonly measured
on the platforms and benchmark against industry standards. Were there
any other Russian-linked accounts that heavily promoted RT or Sputnik
content to the U.S. audience?
Answer. We have provided information concerning 2016 spending by RT
and Sputnik in response to unique issues regarding the 2016 election.
We have not conducted a similar analysis for 2017.
Question 10. Advertisers on both Facebook and Instagram generally
pay for the size and quality of the audience that they would like to
reach on the platforms. Did Facebook advertise to fake-user accounts?
If so, how much revenue or profit did Facebook bring in by advertising
to fake-user accounts?
Answer. We regularly evaluate metrics to estimate the number of
``false'' accounts among our monthly active users. We divide ``false''
accounts into two categories. The first category includes user-
misclassified accounts, where users have created personal profiles for
a business, organization, or non-human entity such as a pet (such
entities are permitted on Facebook using a Page rather than a personal
profile under our terms of service). The second category includes
undesirable accounts, which represent user profiles that we determine
are intended to be used for purposes that violate our terms of service,
such as spamming. We estimate that in the fourth quarter of 2017, false
accounts may have represented approximately three to four percent of
our worldwide monthly active users. Our estimation of false accounts
can vary as a result of episodic spikes in the creation of such
accounts, and additional information relating to our estimate of false
accounts is included in our quarterly filings with the Securities and
Exchange Commission. We continually update our technical systems to
identify, checkpoint, and remove inauthentic accounts, which means that
once discovered, these accounts do not remain active or eligible to
view ads. We believe that revenue generated by advertising to false
accounts is immaterial.
Question 11. Of the 20,000 people Facebook plans to employ by end
of 2018 to work on safety and security, how many will be full-time,
permanent employees? How many will be contractors? Will the majority of
that team be located at the Menlo Park campus? Where is the majority of
Facebook's current security and safety team located?
Answer. We have people working around the world on safety and
security at Facebook. We use a combination of employees and contractors
to make Facebook a place where both expression and personal safety are
protected and respected. This allows us to scale globally with coverage
across time zones, languages, and markets.
Question 12. Will Instagram have its own security and safety team
as well? If so, please provide details.
Answer. Our safety and security teams work across Facebook's family
of applications.
Question 13. Facebook recently announced that it will implement
additional verification and disclosure requirements for advertisers
running election ads for Federal elections. How will the ad onboarding
process change for political advertisers on Facebook and Instagram? How
will political ads look like to users of Facebook and Instagram? Please
provide mock-ups for both the onboarding process and users' view of the
ad.
Answer. We support efforts to promote greater transparency in
political advertising online and are taking steps to make advertising
on Facebook more transparent, increase requirements for authenticity,
and strengthen our enforcement against ads that violate our policies.
We will require more thorough documentation from advertisers who want
to run election-related ads. As part of the documentation process,
advertisers may be required to identify that they are running election-
related advertising and verify both their entity and location. Once
verified, these advertisers will have to include a disclosure in their
election-related ads, which reads: ``Paid for by.'' When users click on
the disclosure, they will be able to see details about the advertiser,
and we will maintain a searchable archive of information. Like other
ads on Facebook, they will also be able to see an explanation of why
they saw that particular ad. For more information, see newsroom.fb.com/
news/2017/10/update-on-our-advertising-transparency-and-authenticity-
efforts.
Question 14. For political advertisers who do not self-identify,
will there be any human controls in addition to the automated tools to
identify the ads proactively? Will Facebook and Instagram still publish
the ad before its buyer is identified? If not, how long will the
advertiser have to wait before the ad is published if they did not
self-identify?
Answer. As part of our efforts to promote greater transparency in
political advertising online, we'll require more thorough documentation
from advertisers who want to run election-related ads. For political
advertisers that do not proactively disclose themselves, we are
building machine learning tools that will help us find them and require
them to verify their identity. Once they are found, we will take
appropriate steps to enforce compliance with our policies.
Question 15. Why are Facebook's verification and disclosure
requirements for political advertisers only limited to Federal
elections?
Answer. We are implementing new verification and disclosure
standards on Facebook that will bring greater transparency to political
advertising on our platform in general and make it easier for us to
enforce our policies. We expect these reforms to be in effect by the
2018 U.S. Federal elections and will progress from there to additional
contests and elections in other countries and jurisdictions.
Question 16. The Washington Post reported that Facebook removed the
data that Jonathan Albright, a researcher at Columbia University, used
to study Russia-linked ads. In response, Facebook stated that this was
done in order to correct a bug in its system, and that Facebook policy
requires that inactive content is no longer available across its
platforms. What is Facebook's data retention policy? Is it consistent
across all of Facebook's properties? Does this policy apply to all
parties--such as independent researchers, users, advertisers, and data
brokers--in the same way?
Answer. Facebook generally retains data for as long as it is
commercially reasonable and necessary for our business. We have taken
appropriate steps to retain relevant information related to IRA
activity on Facebook.
Question 17. Given the importance of collaborating with third-party
or independent researchers to prevent further interference by Russia,
will Facebook be updating its data retention policy?
Answer. Our effort to make our platform safer and more secure is a
holistic one that involves a continual evaluation of our personnel,
processes, and policies, and we make changes as appropriate. We have
taken appropriate steps to retain relevant information related to IRA
activity on Facebook.
Question 18. In terms of dollars and percentage of annual revenue,
how much is Facebook now spending on preventing foreign interference
with our elections? What was the figure in the election cycle leading
up to November 2016? What is the projected spend leading up to November
2018?
Answer. We are determined to do everything we can to protect our
platform. We are expanding our threat intelligence team, and more
broadly, we are working now to more than double the number of people
working on safety and security at Facebook, from 10,000 to 20,000, by
the end of 2018. Many of the people we are adding to these efforts will
join our ad review team, and we also expect to add at least 3,000
people to Community Operations, which is the team that reviews content
that our users and automated tools flag as inappropriate, dangerous,
abusive, or otherwise violative of our policies. These investments will
help us to enforce our policies, including our authenticity policy, and
help us to counter threats from malicious actors, including those who
are state-sponsored. We will also significantly expand the number of
people who work specifically on election integrity before the 2018 U.S.
Federal elections, including people who investigate information
operations by foreign actors. The investments that we are making to
address election integrity and other security issues will be so
significant that we have informed investors that we expect that the
amount that we will spend will impact our profitability.
Question 19. Congress will judge success not by Facebook's efforts
but by its results. How will Facebook measure success? Will Facebook be
conducting an audit after November 2018? When will the results be
shared?
Answer. Success would consist of minimizing or eliminating abuse of
our platform and keeping our community safe. We have a number of
specific goals that we will use to measure our progress in these
efforts.
First, we will increase the number of people working on safety and
security at Facebook, to 20,000 by the end of this year. We will
significantly expand the number of people who work specifically on
election integrity, including people who investigate this specific kind
of abuse by foreign actors. Those specialists will find and remove more
of these actors.
Second, we will work to improve threat intelligence sharing across
our industry, including, we hope, by having other companies join us in
formalizing these efforts. This is a fight against sophisticated
actors, and our entire industry needs to work together to respond
quickly and effectively.
Third, we will bring greater transparency to election ads on
Facebook by requiring more disclosure from people who want to run
election ads about who is paying for the ads and by making it possible
to see all of the ads that an advertiser is running, regardless of the
targeting. We believe that these efforts will help to educate our
community and to arm users, media, civil society, and the government
with information that will make it easier to identify more
sophisticated abuse to us and to law enforcement.
______
Response to Written Questions Submitted by Hon. Tammy Baldwin to
Monika Bickert
Question 1. According to press reports, my home state of Wisconsin
was one of the states where voters were targeted by Russian groups with
Facebook advertising and political content that employed a series of
divisive messages on key issues like race relations and immigration. In
your testimony, you state that Facebook ``continues to seek more
effective ways to combat extremism, crime and other threats to our
national security.''
Do you believe that these foreign-directed activities intended to
sew social discord and influence our elections are threats to our
national security?
I request that Facebook provide me a verbal and written briefing
regarding the Russia-linked political ads and content targeting
Wisconsin, to include: a list and description of the ads and content;
the entities responsible for the ads and content; Facebook's assessment
of the intent of such entities; how the ads and content were targeted
geographically within the state, with regard to social or political
issues, and with regard to audience subgroups; and the timing of the
ads and content.
What steps is Facebook taking to ensure that Russia or other
foreign governments cannot repeat this effort? And that Facebook users
understand the source of this type of advertising and content?
Answer. The foreign interference we saw in the 2016 election is
reprehensible and outrageous and opened a new battleground for our
company, our industry, and our society. That foreign actors, hiding
behind fake accounts, abused our platform and other Internet services
to try to sow division and discord--and to try to undermine our
election process--is an assault on democracy and our national security,
and it violates all of our values. At Facebook, we build tools to help
people connect, and to be a force for good in the world. What these
actors did goes against everything Facebook stands for. Our goal is to
bring people closer together; what we saw from these actors was an
insidious attempt to drive people apart.
We're determined to do our part to prevent it from happening again.
One improvement that we believe will help to address more subtle kinds
of abuse is that our ad review team will do more to assess not just the
content, but also the overall context of an ad, including the buyer and
intended audience. We will also significantly expand the number of
people who work specifically on election integrity before the 2018 U.S.
Federal elections, including people who investigate this specific kind
of abuse by foreign actors. Additionally, we have begun testing a
program where people will be able to click ``View Ads'' on a Page and
view advertisements a Page is running on Facebook, Instagram, and
Messenger--whether or not the person viewing it is in the intended
target audience for the ad. All Pages will be part of this effort, and
we will require that all ads be associated with a Page as part of the
ad creation process. We are also taking steps to make political ads on
Facebook more transparent, increase requirements for authenticity, and
strengthen our enforcement of ads that violate our policies. And, we
continue to make improvements to our efforts to more effectively detect
and deactivate fake accounts to help reduce the spread of spam, false
news, and misinformation. We continually update our technical systems
to identify, checkpoint, and remove inauthentic accounts, and we block
millions of attempts to register fake accounts every day. These systems
examine thousands of detailed account attributes and prioritize signals
that are more difficult for bad actors to disguise, such as their
connections to others on our platform. As with all security threats, we
have been incorporating new insights into our models for detecting fake
accounts, including information specific to election issues.
We are determined to do everything that we can to protect our
platform. The investments that we are making to address these issues
and other security issues will be so significant that we have informed
investors that we expect that the amount that we will spend will impact
our profitability. We will continue to work the government, and across
the tech industry and civil society, to address this important national
security matter so that we can do our part to prevent similar abuse
from happening again. That's why we have provided all of the ads and
associated information to the committees with longstanding, bipartisan
investigations into Russian interference, and we defer to the
committees to share as appropriate. We believe that Congress and law
enforcement are best positioned to assess the nature and intent of
these activities.
Question 2. Ms. Bickert, the national security website Just
Security recently published troubling evidence that raises doubts about
Facebook's ability to prevent, monitor, and remove extremist content.
According to the article, in a one month period spanning December 2017-
January 2018, a researcher named Eric Feinberg reported dozens of pro-
ISIS pages to Facebook. That means the material had gotten past your
company's initial means for flagging and removing terrorist content. In
56 percent of those cases, Facebook removed the offending page. But for
the other 44 percent of reported pages, Facebook left the content up,
noting that it didn't violate community standards. This is despite
there being no appreciable difference between content that was removed
and content that was retained. For example, pages that Facebook left up
included: a photo of gunmen in an urban neighborhood with the caption,
``We Will Attack you in Your Home;'' an online publication promoting
ISIS among the Bangladeshi community; and a photo of Omar Mateen,
praising his attack on Orlando's Pulse nightclub.
Retaining this content contradicts Facebook's explicit policies,
internal guidelines, and your testimony. Can you please explain this?
Answer. We immediately remove terrorists' accounts and posts that
support terrorism whenever we become aware of them. When we receive
reports of potential terrorism posts, we review those reports urgently
and with scrutiny. After receiving the article mentioned in the
question, we reviewed the accounts and content identified in the
article and disabled and removed all those that violated our policies.
Managing a global community in this manner has never been done before,
and we know we have a lot more work to do. We are committed to
improving and to ensuring that hate has no place on Facebook.
Question 3. In the context of extremist content, I would like to
learn more about each company's policy for proactively reporting users
to law enforcement. I understand your companies evaluate and respond to
law enforcement requests for information, but what framework do you use
to proactively report terrorist-related content to authorities,
including any identifying information of the user? For example, if you
use a standard of imminent harm, how do you define and apply it,
particularly in a threat environment where terrorist organizations
often call on recruits to attack carefully planned targets of
opportunity, rather than to launch an immediate, indiscriminate attack?
Answer. We have a long history of working successfully with the
DOJ, the FBI, and other government agencies to address a wide variety
of threats to our platform, including terrorist threats. We reach out
to law enforcement whenever we see a credible threat of imminent harm.
We have been able to provide support to authorities around the world
that are responding to the threat of terrorism, including in cases
where law enforcement has been able to disrupt attacks and prevent
harm.
We cooperate with governments in other ways, too. For example, as
part of official investigations, government officials sometimes request
data about people who use Facebook. We have strict processes in place
to handle these government requests, and we disclose account records in
accordance with our terms of service and applicable law. We also have
law enforcement response teams available around the clock to respond to
emergency requests. Further, governments and inter-governmental
agencies also have a key role to play in convening and providing
expertise that is impossible for companies to develop independently. We
have learned much through briefings from agencies in different
countries about ISIS and Al Qaeda propaganda mechanisms.
Question 4. I would like to hear from the companies whether they
support implementing Mr. Watts's recommendations to: first, fully
certify the authenticity of all users--in other words, ensure that each
user is a real person; and second, eliminate social bot networks to
reduce automated broadcasting of disinformation.
Answer. We have always believed that Facebook is a place for
authentic dialogue, and that the best way to ensure authenticity is to
require people to use the names they are known by. Fake accounts
undermine this objective, and are closely related to the creation and
spread of inauthentic communication such as spam and disinformation. We
also prohibit the use of automated means to access our platform. We
rely on both automated and manual review in our efforts to effectively
detect and deactivate fake accounts, including bots, and we are now
taking steps to strengthen both. For example, we continually update our
technical systems to identify, checkpoint, and remove inauthentic
accounts. We block millions of attempts to register fake accounts every
day. These systems examine thousands of detailed account attributes and
prioritize signals that are more difficult for bad actors to disguise.
Question 5. What are the indicators that you use to identify a
Russian disinformation account, whether from the Kremlin's so-called
Internet Research Agency or an associated group of hackers or trolls,
and what thresholds must be met to disable an account?
Answer. We continually update our technical systems to identify,
checkpoint, and remove inauthentic accounts, including accounts used
for state-sponsored information operations. We block millions of
attempts to register fake accounts every day. These systems examine
thousands of detailed account attributes such as location information
and connections to others on our platform.
______
Response to Written Questions Submitted by Hon. Catherine Cortez Masto
to Monika Bickert
Question 1. During my time as the Attorney General for the State of
Nevada, I saw too many instances of sex trafficking cases involving
child victims that were dismissed because the conduct occurred online
or through social media. So that's why I'm a strong supporter of the
Stop Enabling Sex Traffickers Act of 2017 (SESTA), which clarifies the
Communications Decency Act (CDA) to allow state Attorneys General to
retain their jurisdiction to prosecute those who facilitate human
trafficking. We know that trafficking is happening online and on social
media, and SESTA is the only current legislative proposal that provides
sufficient deterrence to traffickers by providing the necessary tools
for successful prosecutions. As a former prosecutor, I know what it
will take to successfully prosecute those who engage in sex trafficking
through social media and other websites, and that's why I believe that
the House version of SESTA doesn't go far enough to give prosecutors
the tools they need to protect sex trafficking victims. I hope that
your organizations all agree that victims of sex trafficking deserve
meaningful protections and justice.
If so, I'd like to hear whether you will continue to support SESTA
over the weaker U.S. House version of the bill.
Answer. Facebook supports SESTA. We look forward to continuing to
work with Congress to pass this important legislation.
Facebook is committed to making our platform a safe place,
especially for individuals who may be vulnerable. We have a long
history of working successfully with governments to address a wide
variety of threats to our platform, including child exploitation. When
we learn of a situation involving physical abuse, child exploitation,
or an imminent threat of harm to a person, we immediately report the
situation to first responders or the National Center for Missing and
Exploited Children (NCMEC). Further, as part of official
investigations, government officials sometimes request data about
people who use Facebook. We have processes in place to handle these
government requests, and we disclose account records in accordance with
our terms of service and applicable law. We also have a global team
that strives to respond within minutes to emergency requests from law
enforcement.
Our relationship with NCMEC also extends to an effort that we
launched in 2015 to send AMBER Alerts to the Facebook community to help
find missing children. When police determine that a case qualifies for
an AMBER Alert, the alert is issued by the NCMEC and distributed
through the Facebook system with any available information, including a
photograph of the missing child, a license plate number, and the names
and descriptions of the child and suspected abductor. Law enforcement
determines the range of the target area for each alert. We know the
chances of finding a missing child increase when more people are on the
lookout, especially in the critical first hours. Our goal is to help
get these alerts out quickly to the people who are in the best position
to help, and a number of missing children have been found through AMBER
Alerts on Facebook.
Further, we work tirelessly to identify and report child
exploitation images (CEI) to appropriate authorities. We identify CEI
through a combination of automated and manual review. On the automated
review side, we use image hashing to identify known CEI. On the manual
review side, we provide in-depth training to content reviewers on how
to identify possible CEI. Confirmed CEI is reported to the NCMEC, which
then forwards this information to appropriate authorities. When we
report content to the NCMEC, we preserve account information in
accordance with applicable law, which can help further law enforcement
investigations. We also reach out to law enforcement authorities in
serious cases to ensure that our reports are received and acted upon.
Question 2. I was glad to hear that the Internet Association
supports SESTA, and I'd like to know what else your organization is
doing to address concerns about sex trafficking occurring on your
platforms and helping us pass this important legislation in the Senate.
Answer. Please see the response to question 1.
Question 3. I am glad that Facebook has acknowledged the practices
of discrimination through employment or housing as problematic, by
committing to roll out programs that screen for bad ads, and requiring
advertisers to certify their compliance with antidiscrimination laws.
Can you provide a brief update on the progress of those programs?
Answer. We have Community Standards that prohibit hate speech,
bullying, intimidation, and other kinds of harmful behavior. We hold
advertisers to even stricter advertising policies to protect users from
things like discriminatory ads. We don't want advertising to be used
for hate or discrimination, and our policies reflect that. For example,
we make it clear that advertisers may not discriminate against people
based on personal attributes such as race, ethnicity, color, national
origin, religion, age, sex, sexual orientation, gender identity, family
status, disability, and medical or genetic condition. We educate
advertisers on our anti-discrimination policy, and in some cases, we
require advertisers to certify compliance with our anti-discrimination
policy and anti-discrimination laws.
We are committed to getting better at enforcing our advertising
policies. We review many ads proactively using automated and manual
tools, and reactively when people hide, block, or mark ads as
offensive. We are taking aggressive steps to strengthen both our
automated and our manual review. Reviewing ads means assessing not just
the content of an ad, but the context in which it was bought--such as
the identity of the advertiser and the landing page--and the intended
audience. We are changing our ads review system to pay more attention
to these signals. We are also expanding our global ads review teams and
investing more in machine learning to better understand when to flag
and take down ads, such as ads that offer employment or credit
opportunity while including or excluding multicultural advertising
segments. Enforcement is never perfect, but we will get better at
finding and removing improper ads.
Question 4. What metrics are in place that you can provide us to be
confident in the facts and figures you provide to address this concern?
In my view, permitting ad targeting on the basis of age, race,
religion, or other protected characteristics, especially without a
robust process to review ads for compliance with applicable
antidiscrimination laws, is likely to facilitate unlawful
discrimination.
Answer. Please see the response to question 3.
Question 5. So, if we are to believe that Facebook is serious about
combatting discrimination other hateful content, why would it choose to
facilitate discriminatory practices for advertisers using Facebook's
services?
Answer. Discriminatory advertising has no place on Facebook. Our
advertising policies prohibit discrimination, and in some cases, we
require advertisers to certify compliance with our anti-discrimination
policy and anti-discrimination laws. We use automated and manual review
to find and remove discriminatory ads, and we constantly seek to
strengthen our ability to enforce our policies and prevent
discrimination on Facebook.
Question 6. Given that Facebook, like other tech giants, may not be
meeting their own goals for workplace diversity, are you confident that
your employment advertising program meets all of its obligations to
avoid facilitating unlawful employment discrimination?
Answer. As noted above, discriminatory advertising has no place on
Facebook, and we are constantly trying to find ways to improve
enforcement of our anti-discrimination policies. To assist us in these
efforts, we have met with policymakers and civil rights leaders to
listen to their concerns and to gather feedback about ways to improve
our enforcement while preserving the beneficial uses of our advertising
tools. We are grateful for the collaboration of many experts who have
worked with us to develop solutions to combat discriminatory ads. We
look forward to finding additional ways to combat discrimination, while
increasing opportunity for underserved communities, and to continuing
our dialogue with policymakers and civil rights leaders about these
important issues.
Question 7. How can you be sure that the program isn't replicating
the same biases and blind spots that have impeded your own diversity
efforts?
Answer. Please see the response to question 6.
Question 8. Over the past few months, our country has been
reckoning with some hard truths about the way that women and minorities
are treated in the workplace. And I think this is a moment for all
types of organizations, including tech giants like the ones represented
here, to take a clear-eyed accounting of their culture and practices,
to take responsibility for what hasn't worked, and to renew their
commitments to make meaningful improvements. The Equal Employment
Opportunity Commission's 2016 report on ``Diversity in High Tech''
found that women, African Americans, and Hispanics are all represented
at significantly lower levels in high tech than in private industry as
a whole. And while recent internal studies at Facebook and Google have
showed some progress in the hiring of women, there has not been equal
improvement in the representation of people of color and other
underrepresented groups.
What technically qualifies as diversity to your organization?
Answer. With a global community of over two billion people on
Facebook, greater diversity and inclusivity are critical to achieving
our mission. Studies have shown that cognitive diversity on teams that
are working on hard problems produces better results. Diversity helps
us build better products, make better decisions and better serve our
community. In order to achieve that, we have developed programming to
attract and retain more people from traditionally underrepresented
groups which include women, people of color, veterans and people with
disabilities.
We are not where we would like to be, but we are encouraged that
representation for people from underrepresented groups at Facebook has
increased. We've grown Black and Hispanic representation by 1 percent
each (2 percent combined) between our first report in 2014 and our most
recent report in 2017:
Black Representation: from 2 percent to 3 percent
Hispanic Representation: from 4 percent to 5 percent
Black Non-Tech: from 2 percent to 6 percent
Hispanic Non-Tech: from 6 percent to 8 percent
Black Leadership: from 2 percent to 3 percent
Hispanic Leadership: from 4 percent to 3 percent
Black and Hispanic Tech have stayed at 1 percent and 3
percent
As of August 2017, the number of women globally increased from 33
percent to 35 percent and the number of women in tech increased from 17
percent to 19 percent. Women made up 27 percent of all new graduate
hires in engineering and 21 percent of all new technical hires at
Facebook.
We seek to promote diversity in a variety of ways, and we want to
highlight three programs in particular. First, we have adopted our
Diverse Slate Approach (DSA) to interviewing job candidates. The more
people that hirers interview who don't look or think like them, the
more likely they are to hire someone from a diverse background. To
hardwire this behavior at Facebook, we introduced our DSA in 2015 and
have since rolled it out globally. DSA sets the expectation that hiring
managers will consider candidates from underrepresented backgrounds
when interviewing for an open position.
Second, we are working to reduce unconscious bias. Our publicly
available Managing Unconscious Bias class encourages our people to
challenge and correct bias as soon as they see it--in others, and in
themselves. We've also doubled down by adding two new internal
programs: Managing Inclusion, which trains managers to understand the
issues that affect marginalized communities, and Be The Ally, which
gives everyone the common language, tools, and space to practice
supporting others.
Third, we have created Facebook University. We want to increase
access and opportunity for students with an interest in software
engineering, business, and analytics. Facebook University (FBU) gives
underrepresented students extra training and mentorship earlier in
their college education. We started FBU in 2013 with 30 students and
expect to have 280 in 2018. More than 500 students have graduated from
this program, with many returning to Facebook for internships and full-
time jobs.
Finally, we have many partnerships to move the numbers nationally
such as Black Girls Code, All Star Code, Hack the Hood, The Hidden
Genius Project, Level Playing Field Institute, Yes We Code, Streetcode
Academy, Dev Color, Dev Bootcamp and Techbridge. And, we now recruit at
300 Universities--including historically black colleges and
universities (HBCUs) like Spelman, Morehouse, Howard, NCA&T, and Morgan
State (EIR) and the HBCU Faculty Summit.
We're committed to building a more diverse, inclusive Facebook.
Much like our approach to launching new products on our platform, we
are willing to experiment and listen to feedback.
Question 9. How is your company working to address issues of
discrimination in your own workforces?
Answer. Please see the response to question 8.
Question 10. Do you believe those efforts are sufficient?
Answer. Please see the response to question 8.
Question 11. We know that so-called talent pipelines are not the
only obstacle to achieving a diverse workforce, and that discrimination
and harassment go hand in hand, distorting the operation of workplace
meritocracies. This is a moment when many victims of sexual assault and
harassment are bravely coming forward about their experiences, allowing
us to get a better sense of the true scope and effects of this
behavior. Persistent harassment, and the workplace culture that
tolerates, ignores, or even encourages such harassment, pushes people
out of their workplaces, stalls or derails promising careers, and
discourages some from pursuing certain opportunities altogether.
What is your company doing to evaluate the impact of harassment in
your workforces?
Answer. Harassment, discrimination, and retaliation in the
workplace are unacceptable but have been tolerated for far too long.
At Facebook, we treat any allegations of such behavior with great
seriousness, and we have invested significant time and resources into
developing our policies and processes. We have made our policies and
processes available publicly--not because we think we have all the
answers, but because we believe that the more companies are open about
their policies, the more we can all learn from one another. These are
complicated issues, and while we don't believe any company's
enforcement or policies are perfect, we think that sharing best
practices can help us all improve, especially smaller companies that
may not have the resources to develop their own policies. Every company
should aspire to doing the hard and continual work necessary to build a
safe and respectful workplace, and we should all join together to make
this happen.
Our internal policies on sexual harassment and bullying are
available on our Facebook People Practices website
(peoplepractices.fb.com), along with details of our investigation
process and tips and resources we have found helpful in preparing our
Respectful Workplace internal trainings. Our philosophy on harassment,
discrimination, and bullying is to go above and beyond what is required
by law. Our policies prohibit intimidating, offensive, and sexual
conduct even when that conduct might not meet the legal standard of
harassment. Even if it's legally acceptable, it's not the kind of
behavior we want in our workplace.
In developing our policies, we were guided by six basic principles:
First, develop training that sets the standard for respectful
behavior at work, so people understand what's expected of them right
from the start. In addition to prescribing mandatory harassment
training, we wrote our own unconscious bias training program at
Facebook, which is also available publicly on our People Practices
website
Second, treat all claims--and the people who voice them--with
seriousness, urgency, and respect. At Facebook, we make sure to have HR
business partners available to support everyone on the team, not just
senior leaders.
Third, create an investigation process that protects employees from
stigma or retaliation. Facebook has an investigations team made up of
experienced HR professionals and lawyers trained to handle sensitive
cases of sexual harassment and assault.
Fourth, follow a process that is consistently applied in every case
and is viewed by employees as providing fair procedures for both
victims and those accused.
Fifth, take swift and decisive action when it is determined that
wrongdoing has occurred. We have a zero-tolerance policy, and that
means that when we are able to determine that harassment has occurred,
those responsible are fired. Unfortunately, in some cases
investigations are inconclusive and come down to one person's word
against another's. When we don't feel we can make a termination
decision, we take other actions designed to help everyone feel safe,
including changing people's roles and reporting lines.
Sixth, make it clear that all employees are responsible for keeping
the workplace safe--and anyone who is silent or looks the other way is
complicit.
There's no question that it is complicated and challenging to get
this right. We are by no means perfect, and there will always be bad
actors. Unlike law enforcement agencies, companies don't have access to
forensic evidence and instead have to rely on reported conversations,
written evidence, and the best judgment of investigators and legal
experts. What we can do is be as transparent as possible, share best
practices, and learn from one another--recognizing that policies will
evolve as we gain experience. We don't have everything worked out at
Facebook on these issues, but we will never stop striving to make sure
we have a safe and respectful working environment for all our people.
Question 12. How are you working to create a culture where
harassment is no longer tolerated?
Answer. Please see the response to question 11.
Question 13. What more could you be doing to be a positive example
for other companies and industries?
Answer. Please see the response to question 11.
Question 14. Last October, Facebook announced that it would be
improving transparency for all ads run on its platform, including by
requiring political advertisers to include a disclaimer telling viewers
who paid for an ad, and allowing viewers to see all the ads a page is
running, even those that aren't targeting them. Twitter also announced
similar measures. Although these policies were announced in response to
Russia using social media to interfere in our elections, it seems these
transparency measures could help shine a spotlight on other forms of
influence campaigns by extremists or terrorists.
Can you provide an update on the status of these measures?
Answer. We recently announced steps to make advertising on Facebook
more transparent, increase requirements for authenticity, and
strengthen our enforcement against ads that violate our policies. We'll
require more thorough documentation from advertisers who want to run
election-related ads. We are starting with the 2018 Federal elections
in the United States, and will progress from there to additional
contests and elections in other countries and jurisdictions. As part of
the documentation process, advertisers may be required to identify that
they are running election-related advertising and verify both their
entity and location. Once verified, these advertisers will have to
include a disclosure in their election-related ads, which reads: ``Paid
for by.'' When users click on the disclosure, they will be able to see
details about the advertiser, and we will maintain a searchable archive
of information. Like other ads on Facebook, they will also be able to
see an explanation of why they saw that particular ad. For political
advertisers that do not proactively disclose themselves, we are
building machine learning tools that will help us find them and require
them to verify their identity.
Further, for all ads on Facebook--not just political ads--we have
begun testing a program where people will be able to click ``View Ads''
on a Page and view advertisements a Page is running on Facebook,
Instagram, and Messenger--whether or not the person viewing it is in
the intended target audience for the ad. All Pages will be part of this
effort, and we will require that all ads be associated with a Page as
part of the ad creation process. We will roll this feature out to the
United States by this summer, ahead of the U.S. midterm elections in
November, as well as broadly to all other countries around the same
time.
Question 15. When can we expect to see them fully implemented?
Answer. Please see the response to question 14.
Question 16. How are you defining what constitutes a political ad
subject to these heightened transparency requirements?
Answer. Our commitment to ad transparency is not limited to
political ads. While our most recent announcements have focused on
election-related ads--although not necessarily only ads that mention
candidates by name--we are bringing greater transparency to all ads by
making sure that people can see all of the ads run by any Page,
regardless of whether those ads are targeted to them.
Question 17. On January 29, the Director of the Central
Intelligence Agency said he expects the Russian government to attempt
to influence the 2018 elections in this country.
What efforts is Facebook undertaking in the lead up to the 2018
elections to identify and close the platform's remaining
vulnerabilities to foreign exploitation?
Answer. The foreign interference we saw in the 2016 election is
reprehensible and outrageous and opened a new battleground for our
company, our industry, and our society. We're determined to do our part
to prevent it from happening again. We are more than doubling the
number of people who work on safety and security at Facebook and have
already hired thousands more content reviewers. They will be engaged in
processes that we are continuously refining, but this significant
investment of resources will help us to perform those processes more
accurately, quickly, and thoroughly. One improvement that we believe
will help to address more subtle kinds of abuse is that our ad review
team will do more to assess not just the content, but also the overall
context of an ad, including the buyer and intended audience. We will
also significantly expand the number of people who work specifically on
election integrity before the 2018 U.S. Federal elections, including
people who investigate this specific kind of abuse by foreign actors.
Additionally, we have begun testing a program where people will be
able to click ``View Ads'' on a Page and view advertisements a Page is
running on Facebook, Instagram, and Messenger--whether or not the
person viewing it is in the intended target audience for the ad. All
Pages will be part of this effort, and we will require that all ads be
associated with a Page as part of the ad creation process. We are also
taking steps to make political ads on Facebook more transparent,
increase requirements for authenticity, and strengthen our enforcement
of ads that violate our policies. And, we continue to make improvements
to our efforts to more effectively detect and deactivate fake accounts
to help reduce the spread of spam, false news, and misinformation. We
continually update our technical systems to identify, checkpoint, and
remove inauthentic accounts, and we block millions of attempts to
register fake accounts every day. These systems examine thousands of
detailed account attributes and prioritize signals that are more
difficult for bad actors to disguise. As with all security threats, we
have been incorporating new insights into our models for detecting fake
accounts, including information specific to election issues.
We are determined to do everything that we can to protect our
platform. The investments that we are making to address these issues
and other security issues will be so significant that we have informed
investors that we expect that the amount that we will spend will impact
our profitability.
Question 18. What assistance can Federal, state and local
government entities provide in that effort?
Answer. We have a long history of working successfully with the
DOJ, the FBI, and other law enforcement to address a wide variety of
threats to our platform, including threats emanating from Russia. We
deeply respect and value the seriousness, diligence, and support of
those organizations, and we would welcome their partnership as we work
to address this specific threat. We are particularly encouraged by the
FBI's creation of a task force dedicated to addressing election
interference and we are actively working with that newly-formed body.
This is a new kind of threat, and we believe that we will need to work
together--across industry and between industry and government--to be
successful.
Question 19. In November 2016, National Public Radio reported that
Facebook had a few hundred subcontractors across several countries
responsible for reviewing malicious content. That amounted to
contractors reviewing one tag that's been flagged approximately every
10 seconds. In your testimony you said you now have 7,500 employees
reviewing ``terror content and other potential violations.'' You also
said you have 180 people focused specifically on countering terrorism.
Can you please describe the content categories you have employees
reviewing, how each is defined and how material is categorized?
Answer. Our teams review a variety of content for compliance with
our policies. For example, our Business Integrity team focuses on
reviewing and removing ads that do not comply with our advertising
policies, which prohibit discriminatory practices, ads for certain
types of products, misleading or false content, and other activity and
content (see facebook.com/policies/ads). Our Community Operations team
reviews content that our users and automated tools flag as
inappropriate, dangerous, abusive, or otherwise violating our Community
Standards. Our Community Standards prohibit things like serious threats
of harm to public and personal safety, organizations engaged in
terrorist or other violent activity, hate speech, bullying and
harassment, criminal activity, sexual exploitation, the sale of certain
regulated goods, fraud and spam, and other content and activity (see
facebook.com/commu
nitystandards).
Question 20. Please provide the average number of posts or content
flagged per day, the number of employees devoted to each category, and
the amount of time each employee has to review the content. If this is
not a fair metric of employee time allocation to content review, please
provide the metrics requested, as well as an explanation of the content
review systems that allow employees to review a vast amount of content
quickly.
Answer. All content goes through some degree of automated review,
and we use human reviewers to check some content that has been flagged
by that automated review or reported by people that use Facebook. Our
content reviewers respond quickly to millions of reports each week from
people all over the world. We also use human reviewers to perform
reviews of content that was not flagged or reported to check the
accuracy and efficiency of our automated review systems. The amount of
content a reviewer views per day may vary based on the type of content
reviewed and the reason for review.
Question 21. What differentiates employees reviewing ``terror
[content] and other potential violations'' from employees focused on
``countering terrorism''? What is the role of each?
Answer. Our Community Standards prohibit, among other things,
individuals and organizations that are engaged in terrorist activity
and also prohibit content that expresses support for terrorism. Our
content review teams include more than 7,500 people around the world
who work 24 hours a day and in more than 40 of languages to review
reports of terrorist content and other content that may violate our
Community Standards. Separately, we have more than 180 highly trained
people who are exclusively or primarily focused on preventing terrorist
content from ever appearing on our platform and quickly and identifying
and removing it if it does. This group includes former academics who
are experts on counterterrorism, former prosecutors and law enforcement
agents, investigators and analysts, and engineers. Among other things,
this team helps to build tools and leverages counterterrorism research
to detect terrorist activity and prevent the spread of propaganda.
Question 22. Have you found that having 180 employees focused on
countering terrorism is sufficient for the vast amount of content
posted daily on Facebook?
Answer. While 180 employees focus on countering terrorism as the
core part of their job at Facebook, many others in the company share
this responsibility as part of their job. This includes the 7,500
content reviewers who remove from our site content that violates our
policies, including terrorism policies. We work continuously to make
our platform safer and more secure, and our effort to do so is a
holistic one that involves not only hiring additional employees when
issues arise, but also a continual evaluation of our processes and
policies. In addition to our counterterrorism specialists, thousands of
reviewers review content that our users and automated tools flag is
inappropriate, dangerous, abusive, or otherwise violating our Community
Standards--including those prohibiting terrorism. More broadly, we are
working now to ensure that we will more than double the number of
people working on safety and security at Facebook, from 10,000 to
20,000, by the end of 2018.
______
Response to Written Questions Submitted by Hon. Jerry Moran to
Juniper Downs
Question 1. Your testimony covered the ``counter-narratives'' that
YouTube is currently utilizing to speak out against terrorism. Last
week, I had the pleasure of hearing about the important work that is
being done in this effort within Jigsaw's project called ``Redirect
Method.''
Answer. Yes, these strategies have been researched and developed
over many years, and we appreciated your acknowledgement of that work
in the hearing.
Question 2. Could you please describe how this project targets the
most susceptible audience to ``redirect'' them to videos debunking
recruitment materials?
Answer. The Redirect Method uses Adwords targeting tools and third
party curated YouTube videos to confront online radicalization. As you
mentioned, the targeting efforts focus on the slice of ISIS' audience
that is most susceptible to its messaging, and redirects them toward
third party created YouTube videos debunking ISIS recruiting themes.
This open methodology was developed in part from interviews with ISIS
defectors. Jigsaw initially tested the Redirect Method in an ISIS-
focused campaign in Arabic and English. Over the course of 8 weeks,
320,000 individuals watched over half a million minutes of the 116
videos we selected to refute ISIS's recruiting themes. The Redirect
Method has recently been deployed in the UK and France. The Redirect
Method is open for any institution to use in their work.
Question 3. Does ``Redirect Method'' or YouTube create the videos
that program redirects the audience to? If not, why is that the case?
Answer. Many previous efforts to push back on extremist propaganda
have focused on creating new content--writing, videos, etc.--to dispel
extremist narratives. Through our research, we found that content that
had been created for the sole purpose of dispelling extremist
narratives didn't tend to resonate as well as much of the organic
content that was already available online. For this reason, The
Redirect Method focuses on curation of pre-existing content to push
back against extremist propaganda while more effectively reaching the
target audience. We work with local scholars and experts to curate the
videos in Redirect playlists. In France, our interdisciplinary research
team of scholars is based at the Castex Chair of Geostrategy. In the
UK, the Institute for Strategic Dialogue (ISD) and Moonshot CVE
participate in our curation and research efforts.
______
Response to Written Questions Submitted by Hon. Ron Johnson to
Juniper Downs
Question 1. Social media companies are increasingly able to remove
terrorist recruitment, incitement, and training materials before it
posts to their platforms by relying on improved automated systems.
Other than content removal, what else can be done to limit the audience
or distribution of these dangerous materials?
Answer. YouTube's Community Guidelines set the rules of the road
for content that we allow on the platform. Our policies include
prohibitions on hate speech, gratuitous violence, incitement to
violence, terrorist recruitment videos, and violent propaganda. We also
have robust advertiser-friendly guidelines and demonetize videos that
don't comply with those policies, and can age-restrict or place a
warning interstitial in front of content that may be shocking.
If our review teams determine that a video does not contain a
direct call to violence or incitement to hate but could be inflammatory
we may disable some features. Identified borderline content will remain
on YouTube behind a warning interstitial, won't be recommended, won't
be monetized, and won't have key features including comments, suggested
videos, and likes. This new treatment has been positive, with
substantial reduction in watch time of those videos.
We disable access to our services for users who repeatedly violate
our policies--and, for egregious violations, for the first offense. We
also terminate the Google accounts of entities on the U.S. State
Department's Foreign Terrorist Organization (FTO) list, regardless of
the content they are posting.
In addition to ensuring our policies are effectively enforced, we
invest heavily in counterspeech. We see lots of examples of
counterspeech working, such as creators stepping up to refute content
related to violent extremism. In many cases, these creators are driving
even more engagement than the original objectionable content. Exposing
susceptible individuals to counterspeech content is universally viewed
as a critical component of counterterrorism and other counter
radicalization strategies. To that end, we've held over 20
counterspeech workshops around the world, pairing anti-radicalization
NGOs with YouTube creators who know how to best engage with and relate
to their audiences. In 2016, we launched YouTube #CreatorsforChange, a
global counterspeech initiative aimed at amplifying and multiplying the
voices of role models who are tackling difficult social issues such as
xenophobia, hate speech, and extremism.
Question 2. Terrorist how-to guides are protected by the First
Amendment in the United States, but violate the content policies of
many social media companies as well as the laws of some international
partner nations. What countries have laws that go beyond your company's
content policies and can you give examples of how you have worked with
those countries to de-conflict those differences?
Answer. Although we are a U.S.-based company, we respect the law in
countries where we operate. Sometimes those laws restrict speech more
than our Community Guidelines require. Holocaust denial, for example,
while protected by the First Amendment in the United States, is against
the law in many European countries. In countries where we conclude the
law so requires, we would remove such content from our results.
______
Response to Written Questions Submitted by Hon. Maria Cantwell to
Juniper Downs
Question 1. We have strong principles of freedom of speech, but at
the same time, we need to balance that freedom with the need to protect
against bad actors who would leverage that freedom to plan and promote
illegal acts. How can we use artificial intelligence to help us achieve
a balance between our American ideal of free speech and the need to
protect against extremist acts of terror?
Answer. YouTube has always used a mix of humans and technology to
enforce our policies efficiently and at scale. Humans are a critical
component in the enforcement process as very often understanding nuance
and context is necessary to determine if content is in violation of the
company's policies. That said, as the technology improves in its
precision, YouTube leverages these advancements to enhance the
enforcement operations.
For example, YouTube has invested heavily in cutting-edge
artificial intelligence and machine learning technology designed to
help us identify and remove violent extremist and terrorism-related
content in a scalable way. Last year, we began deploying these
classifiers that detect potential terrorist material and flag it for
review by people trained to enforce our policies. These efforts have
resulted in some positive progress:
Speed and efficiency: Our machine learning systems are faster
and more effective than ever before. Last June, only 40 percent
of the videos we removed for violent extremism were identified
by our algorithms. Today, that number is 98 percent. Our
advances in machine learning let us now take down nearly 70
percent of violent extremism content within 8 hours of upload
and nearly half of it in 2 hours.
Accuracy: The efficiency of our systems has improved
dramatically due to our machine learning technology. While
these tools aren't perfect, and aren't right for every setting,
in many cases our systems have proven more accurate than humans
at flagging videos that need to be removed.
Scale: With over 400 hours of content uploaded to YouTube every
minute, finding and taking action on violent extremist content
poses a significant challenge. But since June, our teams have
manually reviewed approximately two million videos to improve
our machine-learning flagging technology by providing large
volumes of training examples. Noteworthy, every subsequent
decision on content that has been flagged by this technology
serves as an additional input that continues to train and
improve the system. We are encouraged by these improvements,
and will continue to develop our technology in order to make
even more progress. We are also hiring more people to help
review and enforce our policies--reaching 10,000 people across
Google working to address content that might violate our
policies by the end of this year--and will continue to invest
in technical resources to keep pace with these issues and
address them responsibly. Our commitment to combat these issues
is sustained and unwavering.
Question 2. Outside of artificial intelligence, what other
technologies could be used to combat potential radicalization on social
media platforms? What does the implementation of those technologies
look like?
Answer. No single component can solve the problem of extremist
content in isolation. In addition to our work on artificial
intelligence and machine learning, YouTube uses a mix of technology and
humans to remove violative content quickly. Users can alert us to
content that they think may violate our policies through a flag found
below every YouTube video. We also have teams charged with reviewing
flagged content 24/7 in multiple languages and countries around the
world. We also work closely with members of our Trusted Flagger
program, which is comprised of NGOs and government agencies with
specific expertise who are provided a bulk-flagging tool to alert us to
content that may violate our policies. Given the higher likelihood that
flags from these organizations are actionable, flags from Trusted
Flaggers are prioritized for review.
We disable access to our services for users who repeatedly violate
our policies--and, for egregious violations, for the first offense. We
also terminate the Google accounts of entities on the U.S. State
Department's Foreign Terrorist Organization (FTO) list, regardless of
the content they are posting.
We also invest heavily in and promote counterspeech to present
counternarratives and elevate voices that counter-extremism. For
example, our Creators for Change program supports creators who are
tackling difficult social issues, including extremism and hate, by
building empathy among their influential audiences and acting as
positive role models online. Similarly, Google's Jigsaw group, an
incubator to tackle some of the toughest global security challenges,
has deployed the Redirect Method. This Method uses Adwords targeting
tools and third party curated YouTube videos uploaded to disrupt online
radicalization.
Question 3. It seems like every week there is a new and more
dangerous security breach. It was recently announced that YouTube would
only be employing people, rather than relying on the newest
technologies, like artificial intelligence, to combat terror-related
content. Do you feel like this decision has the potential to open your
companies content up to bad actors who do utilize next level
technologies?
Answer. Technology will continue to be a part of the how YouTube
enforces our policies and protects our services against bad actors.
From our use of machine learning classifiers to detect violative
content, to video-matching techniques that prevent known bad content
from surfacing on the platform, to tools such as the Redirect Method,
we will continue enhancing our systems to combat the evolving nature of
the threat. We also understand the importance of using methods other
than technology to combat terror-related content. We expanded our
Trusted Flagger Program to an additional 50 NGOs in 2017, including to
groups like Anti-Defamation League and several counter-terrorism
experts such as the Institute of Strategic Dialogue and International
Centre for the Study of Radicalization. Working with these
organizations helps us to better identify emerging trends and
understand how these issues manifest and evolve. In 2018, we will have
10,000 people across Google working to address content that might
violate our policies. This includes engineers and reviewers who work
around the world, 24/7, and speak many different languages.
We also collaborate across the industry. In 2016, we created a
hash-sharing database with Facebook, Twitter and Microsoft, where we
share hashes (or ``digital fingerprints'') of terrorist content to stop
its spread across platforms. Late last year we added Ask.fm,
Cloudinary, Instagram, Justpaste.it, LinkedIn, Oath, and Snap to the
consortium. Industry collaboration is necessary and effective given
counter-terrorism research that shows that many terrorist organizations
engage in cross-platform abuse, and they especially migrate towards
smaller and less-resourced platforms.
______
Response to Written Questions Submitted by Hon. Richard Blumenthal to
Juniper Downs
Question 1. On October 30, 2017, nineteen civil rights groups,
including Muslim Advocates, Leadership Conference on Civil and Human
Rights, NAACP, Southern Poverty Law Center, and many others, co-signed
a letter to Facebook to express concern about the hateful content on
the social media platform used to divide the country, and in
particular, to promote anti-Muslim, anti-Black, anti-immigrant, and
anti-LGBTQ animus.
Ms. Downs, how would YouTube respond to this letter if it had
received it?
Answer. At YouTube, we believe everyone deserves to have a voice,
and that the world is a better place when we listen, share, and build
community through our stories. Our values are based on four essential
freedoms that define who we are:
Freedom of Expression: We believe people should be able to
speak freely, share opinions, foster open dialogue, and that
creative freedom leads to new voices, formats and
possibilities.
Freedom of Information: We believe everyone should have
easy, open access to information and that video is a powerful
force for education, building understanding, and documenting
world events, big and small.
Freedom of Opportunity: We believe everyone should have a
chance to be discovered, build a business and succeed..
Freedom to Belong: We believe everyone should be able to
find communities of support, break down barriers, transcend
borders and come together around shared interests and passions.
Using YouTube to promote violence, incite hate, or celebrate
violent extremism is not only strictly and specifically prohibited by
our terms of service, but is antithetical to our mission. To this end,
we believe that hate, extremism and violence are not confined to any
one community. We apply these policies to violent extremism of all
kinds, whether inciting violence on the basis of race or religion or as
part of an organized terrorist group.
We have worked to enhance not only our technology, but also to
expand and deepen our expertise and resources on these issues. In 2017,
we grew our Trusted Flagger Program by an additional 50 NGOs, experts
on various types of hate and extremism.
Removing content that violates our guidelines or the law is an
important part of the solution, but it's equally critical that we
foster a better ecosystem for positive content and narratives against
hate, extremism, and xenophobia. We are heavily invested in promoting
counterspeech on our platform. We have hosted dozens of workshops
around the world that teach the core skills needed to produce effective
counterspeech. We also launched our flagship counterspeech program,
Creators for Change, that is dedicated to amplifying and multiplying
the voices of content creators who are tackling important social issues
with their channels. YouTube just announced an additional $5M
investment into this program.
As a company, we are dedicated to being a part of the solution and
we will continue to invest in strategies that prevent the spread of
hatred in all its forms.
Question 2. It's now well known how Russian agents used social
media platforms to meddle in our elections--sow division and spread
disinformation in the United States.
Ms. Downs, I understand that on YouTube, you do not necessarily
need to be logged in to view content. However, many users--if not
most--are logged in when they are viewing content nonetheless. Will you
commit to proactively informing all of those identifiable users if they
were victims of Russia's disinformation campaign--as Twitter and
Facebook have already started to do? If you cannot, what percent of
views on YouTube are anonymous?
Answer. We appreciate your work to promote transparency of these
issues. The approximately 1,100 videos we identified as part of our
investigation were removed when we disabled the accounts of these
users. We have posted notice on the pages where those videos previously
appeared, explaining they were removed due to violation of our
Company's Terms of Service.
Question 3. Ms. Bickert, Mr. Monje, and Ms. Downs, please provide
copies (including images, text, dates and timestamps) of all content
identified by your platforms as generated by Russian agents or the
Internet Research Agency.
Answer. We have conducted an extensive review of this issue and we
provided both electronic and hardcopy versions of the ads associated
with accounts we identified as connected to this effort to the
Judiciary Committee. We identified limited activity on our platforms,
but did identify two Ads accounts with approximately $4,700 of spend.
In order to validate our findings, we broadly reviewed all political
ads from June 2015 until the election last November that had even the
loosest connection to Russia, which substantiated that we had
identified the ads connected to this effort.
Our investigation is ongoing, we continue to request and receive
leads from peers in our industry, and will be happy to continue
cooperating with Congressional investigations on this topic.
Question 4. At least one of your peers in the tech industry has
voluntarily initiated an outside assessment of the civil rights impacts
of its policies and programs. In response to concerns regarding
discrimination on the home-sharing platform, AirBNB hired former U.S.
attorney general Eric Holder to help craft an anti-discrimination
policy and has promised to pursue technological innovations to guard
against future discriminatory events.
Mr. Monje, Ms. Bickert, and Ms. Downs, can you each commit to
bringing in an independent entity to conduct a thorough and public
audit of the civil rights impact of your policies and programs,
including how your platform has been used by hate groups to stoke
religious resentment and violence?
Answer. We understand that Airbnb hired outside lawyers to conduct
a comprehensive review after Harvard University researchers published a
working paper that found that users with perceived to be African-
American were more likely to be rejected by Airbnb hosts relative to
guests perceived to be white. We applaud the seriousness that Airbnb
took in addressing this issue and the alleged violations of civil
rights and housing laws on its platform, including hiring Mr. Holder to
draft an anti-discrimination policy.
We are committed to preventing the use of our products for unlawful
activities, including the violation of civil rights laws. As a platform
that hosts content, we deal with difficult questions around many
issues, including hate speech, harassment, violence. But over the
years, and often in consultation with outside organizations, lawyers,
and experts, we have implemented and updated comprehensive policies to
deal with these issues of unwanted content on our platforms. We have
banned hate speech on YouTube and our other hosted platforms, and do
not allow these platforms to be used for harassment or cyberbullying.
And we recently announced a new YouTube policy that puts controversial
and inflammatory videos behind an interstitial warning, where they will
not be monetized, recommended or eligible for comments or user
endorsements. These videos will have less engagement and be harder to
find. We think this strikes the right balance between access to
information without promoting extremely offensive viewpoints.
We have and will continue to engage with outside groups, lawyers,
academics, non-governmental organizations and others to address and
improve the content on our platforms.
Question 5. A little over a year ago, Facebook, Twitter, Google,
and Microsoft announced a plan to create a joint industry database of
``content that promotes terrorism.'' Mr. Monje, Ms. Bickert, and Ms.
Downs, to what extent does this joint industry database focus on all
forms of terror, including the real terror threat presented by white
supremacists?
Answer. In December 2016, Facebook, Microsoft, Twitter and YouTube
announced a coalition to launch a shared database of hashes of
terrorist videos and images to prevent the spread of terrorist content
between services. We expanded this partnership in July 2017, with the
launch of the Global Internet Forum to Counter Terrorism (GIFCT).
GIFCT's first meeting was held in August 2017, where representatives
from the tech industry, government and non-governmental organizations
came together to focus on three key areas: technological approaches,
knowledge sharing, and research.
The GIFCT is committed to working on technological solutions to
help thwart terrorists' use of our services, and has built on the
groundwork laid by the EU Internet Forum, particularly through a shared
industry hash database, where companies can create ``digital
fingerprints'' for terrorist content and share it with participating
companies. The database now contains more than 50,000 hashes. It allows
member companies to use those hashes to identify and remove matching
content--videos and images--that violate our respective policies or, in
some cases, block terrorist content before it is even posted.
Each platform must make difficult decisions about how to balance
issues surrounding free speech. YouTube has been working for years to
combat extremist and hateful content on our platforms, and has long had
policies that prohibit terrorist content, including terrorist
recruitment, violent extremism, incitement to violence, and
instructional content that could be used to facilitate substantial
bodily injury or death.
On YouTube, we apply these policies to violent extremism of all
kinds, whether inciting violence on the basis of race or religion or as
part of an organized terrorist group.
______
Response to Written Questions Submitted by Hon. Brian Schatz to
Juniper Downs
Question 1. Please share a detailed update of YouTube's
investigation into Russia's interference of the 2016 election via
YouTube. Please include any other relevant investigations as they
relate to other U.S. Federal or state elections.
Answer. We conducted an extensive review of this issue, which
spanned across nearly twenty of our products, including YouTube.
As a result of this investigation, we found 18 YouTube channels
likely associated with this campaign that made videos publicly
available, in English, and with content that appeared to be political.
There were 1,108 such videos uploaded, representing 43 hours of content
and totaling 309,000 U.S. views from June 2015 to November 2016. These
videos generally had very low view counts; only around 3 percent had
more than 5,000 views. Upon confirmation with our internal systems
validating the identity of these accounts, we suspended these YouTube
channels at an account level.
Question 2. Google announced last year that it will identify the
names of advertisers running election-related campaigns on Search,
YouTube, and the Google Display Network within Election Ads. When will
this be implemented? How will the ad onboarding process change for
political advertisers on YouTube? How will political ads look to
YouTube's users? Please provide mock-ups for both the onboarding
process and users' view of the ad.
Answer. We've announced several measures to enhance transparency
within election advertising, which we'll roll out prior to the November
midterm elections:
Transparency Report. We'll release a transparency report for
election ads, where we'll share data about who is buying
election-related ads on our platforms and how much money is
being spent.
Creative Library. We'll also introduce a publicly accessible
library of election ads purchased on AdWords and YouTube (with
information about who bought each ad).
In-ad disclosures. We'll identify the names of advertisers
running election-related campaigns on our platforms, including
YouTube.
Verification program. We'll reinforce our existing
protections by requiring that advertisers proactively identify
who they are and where they are based during the onboarding
process before they can run any election-related ads. As they
do, we'll verify that they are permitted to run U.S. election
campaigns through our own checks.
We are in the process of implementing this change to our systems.
While we do not have mock-ups we are able to share at this time, we are
happy to provide information once we are closer to finalizing our work.
Question 3. Columbia University researcher Jonathan Albright has
identified several series of AI-generated videos that consist of a
slideshow and an auto-generated voice narrating an article. His
research has focused on fake news and misinformation content. In one
group of channels (``A Tease . . .''), a new video is created and
uploaded every three minutes. Some of the videos from ``A Tease . . .''
channels appear to be taken down. What specific YouTube policy
violation prompted this removal? Why did YouTube remove only a portion
of these videos but not all? Does YouTube plan to take down the rest of
them? How will YouTube prevent more of these from being uploaded in the
future?
Answer. We have strict policies that prohibit spam, including
posting post large amounts of untargeted, unwanted or repetitive
content. We've developed proprietary technologies to help us fight
spam, which are effective at capturing and preventing the widespread,
and often automated, dissemination of low-quality information. If we
detect large scale automated behavior around account creation, we can
terminate the accounts. We also have policies against videos with
misleading titles and metadata. Among other things, metadata added in
an attempt to mislead viewers or game search algorithms will lead to
the removal of videos and could lead to further action on the channel.
In 2017, we removed over 130,000 videos for violation of this specific
policy.
Question 4. Out of the total number of videos uploaded to YouTube
per day, how many are generated primarily by computers, with little or
no human input?
Answer. We have advanced systems to detect and terminate accounts
that upload high volumes of spam, fraud, and other low-quality content.
These systems are trained to analyze a variety of signals at account
creation and at content upload. As our systems adapt, so do the
behaviors of those who seek to abuse and game our systems. Given the
evolving nature of the threat, our enforcement methods must and do
evolve to respond to them. No matter what challenges emerge, our
commitment to combat them will be sustained and unwavering. We're
committed to getting this right and are increasing both human and
engineering resources to tackle this ever-evolving landscape.
Question 5. Can you list the number of videos that violate YouTube
Community Guidelines and the average amount of time YouTube takes to
address the violation? Please break out by all of the violation
categories currently listed today (1/30/2018).
Answer. We understand that people want a clearer view of how we're
tackling problematic content. Our Community Guidelines give users
notice about what we do not allow on our platforms and we want to share
more information about how these are enforced. Over the next few months
we will be creating a regular transparency report where we will provide
more aggregate data about the flags we receive and the actions we take.
When it comes to violent extremist content, we have removed over
160,000 videos and terminated approximately 30,000 accounts since June
2017. Machine learning is helping our human reviewers remove nearly
five times as many videos than they were previously. Today, 98 percent
of the videos we remove for violent extremism are detected by our
machine-learning algorithms. Our advances in machine learning let us
now take down nearly 70 percent of violent extremist content within
eight hours of upload and nearly half of it in two hours, and we
continue to accelerate that speed.
Question 6. For each of these categories, what percentage of videos
are primarily generated by computers with little or no human input?
Answer. As noted in the answer above, we have advanced systems to
detect and terminate accounts that upload high volumes of spam, fraud,
and other low-quality content. We are continuously investing in
improving these systems to counteract the ever-evolving nature of the
threat.
Question 7. Once a video is taken down, does YouTube delete the
data permanently, or does it just shield the content from public view?
What is YouTube's data retention policy?
Answer. Videos removed for policy violations are generally retained
for a period of time to allow for user appeal. We may also retain some
data about videos for a period of time to comply with applicable laws.
When a user deletes a video, we permanently remove the video and user-
identifiable metadata associated with the video after a recovery
window, which allows the user to recover the video in the case of
accidental deletion.
Question 8. It was announced at the hearing that Google will employ
10,000 people this year to address content that might violate its
policies. Will these employees be focused on any specific set of
policies? Will they be full-time, permanent employees of Google? How
many will be dedicated to YouTube? How many hours of content do you
expect a single person to be responsible for per day (with or without
the help of AI)?
Answer. These employees will primarily sit on the Trust & Safety
teams across YouTube and Google, which work with our in-house legal and
policy departments on escalations and also oversees vendors we hire to
help us scale our operations. The new hires will consist of engineers
and content reviewers, among others, who will work across Google to
address content that violates any of our policies. Many of these
reviewers will be dedicated solely to YouTube content moderation and
they will be made up of a mix of full-time employees and contractors.
Question 9. In terms of dollars and percentage of annual revenue,
how much is YouTube now spending on preventing foreign interference
with our elections? What was the figure in the election cycle leading
up to November 2016? What is the projected spend leading up to November
2018?
Answer. We've been tackling malicious actions directed at our users
or services, including those originating from government-backed actors,
since long before the 2016 elections. For more than a decade, we've
offered our Safe Browsing tool, which helps protect users from
phishing, malware, or other attacks; today it is used on more than
three billion devices worldwide. Additionally, when we detect that a
user's account has been targeted by a government-backed attacker, we
show a warning that includes proactive steps the user can take to
increase the security of his or her account.
Our existing advertising safeguards include policies that prohibit
foreign nationals from buying U.S. election ads. In 2016, we tightly
restricted which advertisers can serve ads to audiences based on their
political leanings. (We offered two categories--left-leaning and right-
leaning--starting in August 2016.) And we're continuing to invest in
enhancements to our safeguards, which will include a transparency
report and creative library that feature election ads bought using our
front-end advertising systems, including those appearing on YouTube.
Election-related advertising is not a large business for Google
relative to other advertising verticals; nevertheless, we understand
the importance of election advertising and are making investments in
launching transparency tools in 2018 that are industry-leading. Across
the world, we have a global team of thousands of policy experts,
reviewers, product managers, and data scientists focused on creating,
maintaining, and enforcing our policies, including those related to
these issues. It is difficult to estimate, however, the overall dollar
and revenue-percentage value of these investments as many of the
resources we're leveraging are widely shared among numerous technical
infrastructural and product development teams.
Question 10. Congress will judge success not by YouTube's efforts
but by its results. How will YouTube measure success? Will YouTube be
conducting an audit after November 2018? When will results be shared?
Answer. We found very limited activity connected to this effort on
our platform in our investigation into potential election interference
with respect to the 2016 Presidential election, however we are
committed to preventing the misuse of our platforms for the purpose of
interfering with democratic elections. At the end of last year, we
announced we would take the following steps:
Transparency Report. In 2018, we'll release a transparency
report for election ads, which will share data about who is
buying election-related ads on our platforms and how much money
is being spent.
Creative Library. We'll also introduce a publicly accessible
database of election ads purchased on AdWords and YouTube (with
information about who bought each ad). That means people will
not only be able to learn more about who's buying election-
related ads on our platforms; they'll be able to see the ads
themselves, regardless of to whom they were shown.
In-ad disclosures. Going forward, we'll identify the names
of advertisers running election-related campaigns on Search,
YouTube, and the Google Display Network via our ``Why This Ad''
icon.
Verification program. U.S. law restricts entities outside
the U.S. from running election-related ads. We'll reinforce our
existing protections by requiring that advertisers proactively
identify who they are and where they are based before running
any election-related ads.
We also have increased our longstanding support to non-profits and
journalists dedicated to ensuring the integrity of our election
systems. For example, we've recently contributed nearly $750,000 to the
bipartisan ``Defending Digital Democracy'' project, led by the Belfer
Center for Science and International Affairs at Harvard Kennedy School.
And we are deeply committed to helping people participate in the
election by providing users with timely and comprehensive information
they need to make their voice heard.
______
Response to Written Questions Submitted by Hon. Tammy Baldwin to
Juniper Downs
Question 1. In the context of extremist content, I would like to
learn more about each company's policy for proactively reporting users
to law enforcement. I understand your companies evaluate and respond to
law enforcement requests for information, but what framework do you use
to proactively report terrorist-related content to authorities,
including any identifying information of the user? For example, if you
use a standard of imminent harm, how do you define and apply it,
particularly in a threat environment where terrorist organizations
often call on recruits to attack carefully planned targets of
opportunity, rather than to launch an immediate, indiscriminate attack?
Answer. Google discloses information to government entities when
the threat of loss of life or serious physical injury is brought to our
attention by governmental entities, and when we learn of the threat
ourselves or from other sources. Emergency disclosures are handled
twenty-four hours a day, every day of the year and consistent with the
law.
Evaluating whether there is a credible threat presented in any
particular case requires evaluation of the facts as we know them at the
time. This can include looking to information on Google's platform as
well as from other sources. Public sources of information can also
assist in making the determination as well as having an understanding
of the threat environment more generally.
Question 2. I would like to hear from the companies whether they
support implementing Mr. Watts's recommendations to: first, fully
certify the authenticity of all users--in other words, ensure that each
user is a real person; and second, eliminate social bot networks to
reduce automated broadcasting of disinformation.
Answer. We have not seen the same degree of social media bots that
have been reported on other platforms. Our systems rely on a host of
inputs about historical use and pattern recognition across various
services in an effort to detect if an account creation or login is
likely to be abusive. The system operates to block ``bad'' account
creation or to close groups of such accounts. We prevent users from
creating a large number of Google Accounts in a short time period if
our systems detect that the user might be abusive. If we detect
suspicious conduct, we also require verification, aimed at detecting if
a bot is attempting to access or create an account. We have also
developed robust protections over the years to address attempts to
manipulate our systems by bots or other schemes, like link farms. (Our
webmaster guidelines provide more information about this: https://
support.google.com/webmasters/answer/35769.) We use both algorithmic
and manual methods, and we deploy these across our products including
Search and YouTube.
Question 3. What are the indicators that you use to identify a
Russian disinformation account, whether from the Kremlin's so-called
Internet Research Agency or an associated group of hackers or trolls,
and what thresholds must be met to disable an account?
Answer. We developed a list of actors we know or suspect were
involved in this effort from (1) our research of publicly available
information, (2) the work of our security team, and (3) leads we
received from others in the industry and Jigsaw. In addition, we
reviewed all ads from June 2015 until the election in November 2016
that were categorized as potentially political by our systems and had
even the loosest connection to Russia, such as a Russian I.P. address
or billing address or were paid for using Russian currency. When we
identified accounts that we believed were associated with this effort,
we removed them.
______
Response to Written Questions Submitted by Hon. Catherine Cortez Masto
to Juniper Downs
Question 1. During my time as the Attorney General for the State of
Nevada, I saw too many instances of sex trafficking cases involving
child victims that were dismissed because the conduct occurred online
or through social media. So that's why I'm a strong supporter of the
Stop Enabling Sex Traffickers Act of 2017 (SESTA), which clarifies the
Communications Decency Act (CDA) to allow state Attorneys General to
retain their jurisdiction to prosecute those who facilitate human
trafficking. We know that trafficking is happening online and on social
media, and SESTA is the only current legislative proposal that provides
sufficient deterrence to traffickers by providing the necessary tools
for successful prosecutions. As a former prosecutor, I know what it
will take to successfully prosecute those who engage in sex trafficking
through social media and other websites, and that's why I believe that
the House version of SESTA doesn't go far enough to give prosecutors
the tools they need to protect sex trafficking victims. I hope that
your organizations all agree that victims of sex trafficking deserve
meaningful protections and justice.
If so, I'd like to hear whether you will continue to support SESTA
over the weaker U.S. House version of the bill.
Answer. This is a very important issue and we are glad that
Congress has taken it up. There's no question that Backpage is a bad
actor, and we strongly agree that they and other bad actors should be
prosecuted for their crimes under the full extent of the law. We firmly
support both increased enforcement of existing laws against sex
trafficking, and strengthening of those laws where appropriate to
combat this heinous activity. As you know, we are members of the
Internet Association, which has endorsed both the House and Senate
legislation. We will continue to work with any lawmakers interested in
addressing this important problem, while ensuring that these efforts do
not inadvertently hinder the ability of websites to remove illegal
content and to fight sex trafficking.
Question 2. I was glad to hear that the Internet Association
supports SESTA, and I'd like to know what else your organization is
doing to address concerns about sex trafficking occurring on your
platforms and helping us pass this important legislation in the Senate?
Answer. Google has made it a priority to tackle the horrific crime
of sex trafficking. We have donated over $20 million to organizations
on the front lines of ending modern day slavery and human trafficking.
We have developed and built extensive technology to connect victims
with the resources they need. And we have helped pioneer the use of
technologies that identify trafficking networks to make it easier and
quicker for law enforcement to arrest these abusers.
Google has a zero tolerance policy for any advertising related to
sex trafficking and prostitution. We work tirelessly to ensure bad
actors are not able to exploit our products and use an industry-leading
combination of sophisticated technology and manual review to detect and
remove bad ads. We've invested millions of dollars in building these
systems to scale. We also work with law enforcement and NGOs that are
focused on stopping human trafficking.
Additionally, we have always had strict policies against child
endangerment, and we partner closely with regional authorities and
experts to help us enforce these policies and report to law enforcement
through the National Center for Missing and Exploited Children (NCMEC).
Key principles in Section 230 of the Communications Decency Act enable
us to develop new strategies around user flags, machine learning, and
automated enforcement to stop sex trafficking and other forms of child
endangerment on our services.
We will continue to work to prevent this type of activity from
occurring on our platform, and we welcome efforts by Congress to help
us do so.
Question 3. Over the past few months, our country has been
reckoning with some hard truths about the way that women and minorities
are treated in the workplace. And I think this is a moment for all
types of organizations, including tech giants like the ones represented
here, to take a clear-eyed accounting of their culture and practices,
to take responsibility for what hasn't worked, and to renew their
commitments to make meaningful improvements. The Equal Employment
Opportunity Commission's 2016 report on ``Diversity in High Tech''
found that women, African Americans, and Hispanics are all represented
at significantly lower levels in high tech than in private industry as
a whole. And while recent internal studies at Facebook and Google have
showed some progress in the hiring of women, there has not been equal
improvement in the representation of people of color and other
underrepresented groups.
What technically qualifies as diversity to your organization?
Answer. At Google we take a broad and intersectional view of
diversity--from race and gender, to sexual orientation, gender
identity, age, disability, socio-economic background and more. That
said, we are actively working to improve the representation of
underrepresented groups in our workforce.
We believe deeply in diversity and inclusion and it underpins
Google's business. We believe that if we tap the full range of human
experience, capability and contribution, we will move faster, increase
innovation and creativity, and can tackle more and more of the world's
problems. Increasing diversity not only makes good business sense, it's
also the right thing to do.
Google's core mission is to organize the world's information and
make it universally accessible and useful. Our goal for diversity is a
natural extension of this mission--to increase access to opportunity,
by breaking down barriers and empowering people through technology.
Products will only get better and more useful if we invite all segments
of society, and people from all over the world, to influence and create
technology.
Question 4. How is your company working to address issues of
discrimination in your own workforces?
Answer. We've worked hard over many years to create and foster a
fair and inclusive Google, and we absolutely do not tolerate
discrimination, or any actions that create a hostile work environment.
This work starts with creating and building a fair and inclusive
culture, that keeps discrimination from happening in the first place.
Once Googlers get here, we strive to ensure that our work
environment is fair and inclusive, so they can grow and flourish. And
we keep close tabs on our programs and process to ensure they yield
fair and equitable outcomes. When we learn that something is amiss, we
take action early and follow up to make sure we continue to create a
great place to work.
We have a culture that empowers Googlers to quickly raise up issues
of concern and where they feel they've been mistreated, or
discrimination has happened. If Googlers are concerned by any
inappropriate behavior they experience, or see, in the workplace, we
ask them to please report it.
There are many avenues to do this, including by anonymous means,
and we review each complaint. We do a lot to make Googlers aware of
those channels--ranging from reaching out directly to HR to a third-
party helpline if anyone wants to stay anonymous.
Question 5. Do you believe those efforts are sufficient?
Answer. Google's approach to our products and our business is that
we can always do better, and our approach to diversity and inclusion,
and how we oppose discrimination, is no different. We have always been
transparent about our commitment to diversity, inclusion, equity and
compliance in our workforce. More importantly, we're also transparent
about our challenges and key learnings in this arena. We believe the
best way to be a positive example is by making progress in our own
workforce and culture, and that's a key priority of ours.
Question 6. I've seen that Facebook works to make their labor
diversity information public, can you provide a status on your labor
figures, or commit to sharing those with the Committee and the public?
Answer. As you may know, Google was the first large tech company to
publish workforce diversity data in 2014 (all of this information is
available at google.com/diversity). We are committed to sharing our
numbers every year, and 2018 is no different. We plan on once again
releasing those numbers once again later this year.
As stated earlier, we believe it is important to be transparent
about our challenges and key learnings in this arena. Our original
decision to release our workforce (diversity) numbers led to other
companies following suit. Google stands firm in its commitment to
foster dialogue and to drive impact on this important issue.
Question 7. We know that so-called talent pipelines are not the
only obstacle to achieving a diverse workforce, and that discrimination
and harassment go hand in hand, distorting the operation of workplace
meritocracies. This is a moment when many victims of sexual assault and
harassment are bravely coming forward about their experiences, allowing
us to get a better sense of the true scope and effects of this
behavior. Persistent harassment, and the workplace culture that
tolerates, ignores, or even encourages such harassment, pushes people
out of their workplaces, stalls or derails promising careers, and
discourages some from pursuing certain opportunities altogether.
What is your company doing to evaluate the impact of harassment in
your workforces? How are you working to create a culture where
harassment is no longer tolerated?
Answer. Google has clear policies on appropriate behavior by
employees of the company. Harassment has never been tolerated. We
review all concerns, and take action when necessary. Here's a snapshot
of what some of this work has looked like over the past few years:
In 2015, we launched the Respect@ program as a way for
Googlers to raise concerns, share experiences and get support.
The program is a way for Googlers to learn about the standards
of behavior we expect, the different ways for reporting
unacceptable behavior, and the process we undertake to
investigate complaints.
Respect@ is supported by the senior-most leadership of
the company, and is championed by an Executive Oversight
Committee that currently consists of 18 VPs, across
functions and regions.
As part of Respect@, we also created go/saysomething,
an internal online resource to provide a discreet way to
report inappropriate behavior. If Googlers are concerned by
any inappropriate behavior they experience, or see, in the
workplace, we ask them to please report it. There are many
avenues to do this, including by anonymous means, and we
review and investigate each complaint.
Since launching Respect@, complaints/investigations
have increased significantly, as people became aware of the
avenues open to them
In 2015, we launched the first annual Internal Investigation
Report, so that Googlers can see the number and type of
complaints we receive, as well as the outcomes of these
complaints. We have published these reports every year since.
Question 8. What more could you be doing to be a positive example
for other companies and Industries?
Answer. Google has always been transparent about its commitment to
diversity, inclusion, equity and compliance in our workforce. More
importantly, we're also transparent about our challenges and key
learnings in this arena. We believe the best way to be a positive
example is by making progress in our own workforce and culture, and
that's a key priority of ours.
Question 9. Last October, Facebook announced that it would be
improving transparency for all ads run on its platform, including by
requiring political advertisers to include a disclaimer telling viewers
who paid for an ad, and allowing viewers to see all the ads a page is
running, even those that aren't targeting them. Twitter also announced
similar measures. Although these policies were announced in response to
Russia using social media to interfere in our elections, it seems these
transparency measures could help shine a spotlight on other forms of
influence campaigns by extremists or terrorists.
Can you provide an update on the status of any improvements YouTube
is making?
Answer. At the end of last year, we announced we would take the
following steps across Google's platforms, including on YouTube:
Transparency Report. In 2018, we'll release a transparency
report for election ads, which will share data about who is
buying election-related ads on our platforms and how much money
is being spent.
Creative Library. We'll also introduce a publicly accessible
database of election ads purchased on AdWords and YouTube (with
information about who bought each ad). That means people will
not only be able to learn more about who's buying election-
related ads on our platforms; they'll be able to see the ads
themselves, regardless of to whom they were shown.
In-ad disclosures. Going forward, we'll identify the names
of advertisers running election-related campaigns on Search,
YouTube, and the Google Display Network via our ``Why This Ad''
icon.
Verification program. U.S. law restricts entities outside
the U.S. from running election-related ads. We'll reinforce our
existing protections by requiring that advertisers proactively
identify who they are and where they are based before running
any election-related ads.
Question 10. If applicable, when can we expect to see them fully
implemented?
Answer. We are in the process of implementing these steps and plan
to have them completed in time to be helpful to users in understanding
how ads are purchased during the November elections.
Question 11. If applicable, how are you defining what constitutes a
political ad subject to these heightened transparency requirements?
Answer. We will apply the new requirements to political
advertisements that either constitute ``express advocacy'' or contain a
reference to a clearly identified candidate, as each of those terms is
defined by the Federal Election Commission.
Question 12. On January 29, the Director of the Central
Intelligence Agency said he expects the Russian government to attempt
to influence the 2018 elections in this country.
What efforts is YouTube undertaking in the lead up to the 2018
elections to identify and close the platform's remaining
vulnerabilities to foreign exploitation?
Answer. As mentioned above, at the end of last year, we announced
we would take the following steps across Google's platforms, including
on YouTube:
Transparency Report. In 2018, we'll release a transparency
report for election ads, which will share data about who is
buying election-related ads on our platforms and how much money
is being spent.
Creative Library. We'll also introduce a publicly accessible
database of election ads purchased on AdWords and YouTube (with
information about who bought each ad). That means people will
not only be able to learn more about who's buying election-
related ads on our platforms; they'll be able to see the ads
themselves, regardless of to whom they were shown.
In-ad disclosures. Going forward, we'll identify the names
of advertisers running election-related campaigns on Search,
YouTube, and the Google Display Network via our ``Why This Ad''
icon.
Verification program. U.S. law restricts entities outside
the U.S. from running election-related ads. We'll reinforce our
existing protections by requiring that advertisers proactively
identify who they are and where they are based before running
any election-related ads.
We also have increased our longstanding support to non-profits and
journalists dedicated to ensuring the integrity of our election
systems. For example, we've recently contributed nearly $750,000 to the
bipartisan ``Defending Digital Democracy'' project, led by the Belfer
Center for Science and International Affairs at Harvard Kennedy School.
And we are deeply committed to helping people participate in the
election by providing users with timely and comprehensive information
they need to make their voice heard.
Question 13. What assistance can Federal, state and local
government entities provide in that effort?
Answer. We are committed to working with Congress, law enforcement,
others in our industry, and the NGO community to strengthen protections
around elections, ensure the security of users, and help combat
disinformation, and we welcome tips and other information from
knowledgeable government agencies on these matters.
______
Response to Written Questions Submitted by Hon. Roger Wicker to
Carlos Monje, Jr.
Question 1. Has Twitter placed any restrictions on the U.S.
Government's use of publicly available information on your platform? If
yes, please describe what those restrictions are, why they have been
imposed and on which U.S. Government agencies?
Answer. Twitter is a public platform. When users choose to share
information by posting it to their public profile, the information is
available to anyone who visits those users' profiles. With respect to
Twitter's application programming interface (``API'') (public and
commercial), through which we provide developers and other third
parties access to subsets of public Twitter content, all users of our
developer products must comply with Twitter's developer terms and
policies. Those policies include long-standing provisions that
prohibit, among other things, the use of Twitter data for surveillance
purposes or for purposes in contravention of the Universal Declaration
of Human Rights. While Twitter works closely with its developer
community to address questions and investigate instances of potential
abuse, each developer is responsible for compliance with Twitter's
applicable policies.
Twitter maintains strong working relationships with law
enforcement. We publish guidelines for law enforcement personnel that
explain our policies and the process for submitting requests for
information. We regularly respond to law enforcement requests, have a
dedicated 24/7 response team for that purpose, and have developed a
user-friendly online submission form to streamline responses to law
enforcement agencies through properly scoped valid legal process. There
are also a number of news alert products that are available and used by
law enforcement, including the Federal Bureau of Investigation.
Question 2. Are U.S. Government agencies or intelligence
organizations permitted to search for or monitor--either directly with
Twitter or through third-party aggregators--counterterrorism
information or specific Twitter accounts that are likely affiliated
with terrorist organizations within the publicly available content
found on Twitter's platform? If not, why? Please explain.
Answer. The answer to Question 2 has been provided in response to
Question 1.
Question 3. Are companies (such as casinos) allowed to monitor--
either directly with Twitter or through third-party aggregators--
specific Twitter accounts that have made public threats against their
venues or staff?
Answer. The answer to Question 2 has been provided in response to
Question 1.
Question 4. Does Twitter have any policies that prohibit the use of
its data, by any public or private third-party, for counterterrorism
analyses focused on terrorist organizations? If so, can you please
explain the purpose of that policy and the parameters of it?
Answer. The answer to Question 2 has been provided in response to
Question 1.
Question 5. During the hearing, Mr. Monje testified that Twitter
works with law enforcement through the ``proper legal process''. Please
describe the legal process to which Mr. Monje was referring and how it
applies to law enforcement's use of Twitter's aggregate user data.
Answer. As we noted above in response to Question 1, Twitter
maintains strong working relationships with law enforcement. We publish
guidelines for law enforcement personnel that explain our policies and
the process for submitting requests for information. See https://
help.twitter.com/en/rules-and-policies/twitter-law-enforcement-support.
We regularly respond to law enforcement requests, have a dedicated 24/7
response team for that purpose, and have developed a user-friendly
online submission form to streamline responses to law enforcement
agencies through valid legal process. Before launching this system to
all U.S. law enforcement agencies, we conducted a pilot with the
Federal Bureau of Investigation. We have begun rolling out this tool
for global use.
In addition, we have offered and conducted training sessions to law
enforcement officials to familiarize them with our policies and
procedures. In 2017, we have attended and provided training at a
national conference for investigators of crimes against children,
training events for FBI legal attaches posted to U.S. embassies abroad,
and other conferences with the participation of federal, state and
local law enforcement. We continue to build upon and invest in our law
enforcement outreach and training. And we welcome feedback from law
enforcement experts and professionals about how we can improve our
systems.
We regularly and directly engage with law enforcement officials on
a wide range of issues, including extremist content online. We receive
and respond to ``Internet Referral Unit'' reports of extremist content.
Our recently published Transparency Report for the first half of 2017
details the statistics of those responses. See https://
blog.twitter.com/official/en_us/topics/company/2017/New-Data-Insights-
Twitters-Latest-Transparency-Report.html. In addition, we receive
briefings from government experts on terrorist use of online platforms,
which help inform our proactive efforts.
Law enforcement requests to Twitter must comply with applicable
laws in the jurisdiction where they are issued. For Federal law
enforcement, this includes the Electronic Communications Privacy Act 18
U.S.C. 2510 et seq. These requirements only apply to data sought from
Twitter. If law enforcement is able to access publicly available
information from the Twitter service they may do so subject to any
other legal or policy restrictions that may apply to their conduct
(e.g., Department of Justice guidance). If law enforcement seeks access
to Twitter data via our API directly or through a third party
developer, they must do so in a manner that complies with the
applicable Twitter terms and policies for our API.
Question 6. Does a U.S. Government agency have to obtain a warrant
(or go through a similar legal process as discussed in Question #5) to
search publicly available information found on Twitter? If yes, why? If
no, does Twitter allow U.S. Government agencies to gain access to
publicly available information on its platform through third-parties
that have purchased aggregate user data from Twitter?
Answer. The answer to Question 6 has been provided in response to
Question 1.
Question 7. Twitter's platform allows users to ``follow'' other
users. In your view, what is the difference between ``following''
someone and ``surveilling'' someone?
Answer. A user may follow another account holder via the Twitter
service. An account holder may view their list of followers at any
time. The account holder may take a range of actions in response to
receiving a ``follow'' from another user. They may decide to follow
that user in return. They may also choose to block that follower,
preventing the follower from viewing in their timeline content posted
that account holder or receiving notifications of posts by the account
holder Twitter users may also choose to make their account private so
as to restrict new followers to those that they have expressly allowed
as followers. These choices, including the ability to restrict
followers and the transparency inherent in our platform, are important
aspects of the Twitter service.
______
Response to Written Questions Submitted by Hon. Jerry Moran to
Carlos Monje, Jr.
Question 1. How is the battle against terrorist content different
from misinformation and abuse?
Answer. Twitter is committed to combating terrorist content,
abusive activity (including malicious automated activity) and the
spread of misinformation on our platform. Because bad actors often rely
on the same methods to propagate such content, to some extent, we
deploy similar tools to detect and stop all malicious activity on
Twitter, including the proliferation of terrorist content. At the same
time, we recognize that each of these areas presents unique challenges
to deploying our technology at scale. For example, with terrorist
content, we can more readily identify the signals of bad actors
intending to disseminate terrorist propaganda and can efficiently find
and suspend many of these accounts using machine learning. In contrast,
for abuse and misinformation, the context of conversations and the
content itself are often needed to determine whether something crosses
a policy line.
However, Twitter's approach to addressing the spread of malicious
automation and inauthentic accounts on our platform is to focus
wherever possible on identifying problematic behavior, and not on the
content itself. Those who are seeking to influence a wide audience
often find ways to try to artificially amplify their messages across
Twitter. As with spam, these behaviors frequently provide more precise
signals than focusing on content alone.
Accordingly, we monitor various behavioral signals related to the
frequency and timing of Tweets, Retweets, likes, and other such
activity, as well as to similarity in behavioral patterns across
accounts, in order to identify accounts that are likely to be
maliciously automated or acting in an automated and coordinated fashion
in ways that are unwelcome to our users. We monitor and review
unsolicited targeting of accounts, including accounts that mention or
follow other accounts with which they have had no prior engagement. For
example, if an account follows 1,000 users within the period of one
hour, or mentions 1,000 accounts within a short period of time, our
systems are capable of detecting that activity as aberrant and as
potentially originating from suspicious accounts.
Twitter is continuing its effort to detect and prevent malicious
automation by leveraging our technological capabilities and investing
in initiatives aimed at understanding and addressing behavioral
patterns associated with such accounts. For example, in early 2017, we
launched the Information Quality initiative, an effort aimed at
enhancing the strategies we use to detect and stop bad automation,
improve machine learning to spot spam, and increase the precision of
our tools designed to prevent such content from contaminating our
platform.
In 2017, we have made significant improvements to reduce external
attempts to manipulate content visibility. These improvements were
driven by investments in methods to detect malicious automation through
abuse of our API, limit the ability of malicious actors to create new
accounts in bulk, detect coordinated malicious activity across clusters
of accounts, and better enforce policies against abusive third-party
applications.
In addition, we have developed new techniques for identifying
patterns of activity inconsistent with legitimate use of our platform
(such as near-instantaneous replies to Tweets, nonrandom Tweet timing,
and coordinated engagement), and we are currently implementing these
detections across our platform. We have improved our phone verification
process and introduced new challenges, including reCAPTCHA (utilizing
an advanced risk-analysis engine developed by Google), to give us
additional tools to validate that a human is in control of an account.
We have enhanced our capabilities to link together accounts that were
formed by the same person or that are working in concert. And we are
improving how we detect when accounts may have been hacked or
compromised.
With our improved capabilities, we are now detecting and blocking
approximately 523,000 suspicious logins each day that we believe to be
generated through automation. In December 2017, our systems identified
and challenged more than 6.4 million suspicious accounts globally per
week--a 60 percent increase in our detection rate from October 2017.
Over three million of those accounts were challenged upon signup,
before their content or engagements could impact other users. Since
June 2017, we also suspended more than 220,000 malicious applications
for API abuse. These applications were collectively responsible for
more than 2.2 billion Tweets in 2017. We plan to continue building upon
our 2017 improvements, including through collaboration with our peers
and investments in machine-learning capabilities that help us detect
and mitigate the effect on users of fake, coordinated, and malicious
automated account activity.
We have also observed the expansion of malicious activity on our
platform from automated accounts to human-coordinated activity, which
poses additional challenges to making our platform safe. We are
determined to meet those challenges and have been successful in
addressing such abusive behavior in other contexts. We are committed to
leveraging our technological capabilities in order to do so again by
carefully refining and building tools that respond to signals in the
account behavior.
Those tools have also been successful at detecting and removing
terrorist content on our platform. For example, as of September 2017,
95 percent of account suspensions for promotion of terrorist activity
were accomplished using our existing proprietary detection tools--up
from 74 percent in 2016. These tools focus on indicia of violating
activity beyond the content of the Tweet. Although they have proved
successful, our efforts to address terrorist content on our platform do
not end with investments in our proprietary detection tools. We
recognize that the spread of terrorist and extremist content online is
not unique to Twitter, and we are committed to collaborating with our
industry peers to address this shared thread. Accordingly, in June
2017, we launched the Global Internet Forum to Counter Terrorism (the
``GIFCT''), a partnership among Twitter, YouTube, Facebook, and
Microsoft. The GIFCT facilitates, among other things, information
sharing; technical cooperation; and research collaboration, including
with academic institutions. In September 2017, the members of the GIFCT
announced a multimillion dollar commitment to support research on
terrorist abuse of the Internet, and how governments, tech companies,
and civil society can respond effectively. We are looking to establish
a network of experts that can develop these platform-agnostic research
questions and analysis that considers a range of geopolitical contexts.
The GIFCT opened a call for proposals in December 2017, and we look
forward to sharing further details of the initial projects this year.
The GIFCT has created a shared industry database of ``hashes''--
unique digital ``fingerprints''--for violent terrorist imagery or
terrorist recruitment videos or images that have been removed from our
individual services. The database allows a company that discovers
terrorist content on one of its sites to create a digital fingerprint
and share it with the other companies in the forum, who can then use
those hashes to identify such content on their services or platforms,
review against their respective policies and individual rules, and
remove matching content as appropriate, or even block extremist content
before it is posted in the first place. The database now contains more
than 40,000 hashes. Instagram, Justpaste.it, LinkedIn, Oath, and Snap
have also joined this initiative, and we are working to add several
additional companies in 2018. Twitter also participates in the
Technology Coalition, which shares images to counter child abuse.
As part of our work with the GIFCT, we have hosted more than 50
small companies at workshops through the Tech Against Terrorism
initiative, our partners under the UN CounterTerrorism Executive
Directorate. Twitter believes that this partnership will provide a
unique opportunity for us to share our knowledge and technical
expertise with smaller and emerging companies in the industry and for
all industry actors to harness the expertise that has been built up in
recent years.
We have also focused on NGO outreach and, since 2013, and have
participated in more than 100 Countering Violent Extremism training and
events around the world, including in Beirut, Bosnia, Belfast and
Brussels, and summits at the White House, at the United Nations,
London, and Sydney. Twitter has partnered with groups like the
Institute of Strategic Dialogue, the Anti-Defamation League and Imams
Online to bolster counterspeech that offers alternatives to
radicalization. As a result of that work, NGOs and activists around the
world are able to harness the power of our platform in order to offer
positive alternative narratives to those at risk and their wider
communities.
Finally, in addressing abuse directed at users on the platform,
context matters. A turn of phrase can be playful or offensive,
depending on the circumstance, topic, and author. This means we need
more nuanced and creative approaches to our machine learning models in
order to address abusive activity at scale. One example where we have
made progress is in our improving ability to action reports of abuse by
witnesses (instead of by victim directly). By looking at various
signals, including the relationship and activity between the reported
abuser and reported victim, we can better identify, escalate, and take
action against instances of abuse.
Question 2. Can you walk through your track record of removing
terrorist content?
Answer. As noted above, we have made considerable inroads against
the proliferation of terrorist content on our platform. For example, in
February 2016 when we first started sharing metrics for our enforcement
efforts, we announced that, since the middle of the preceding year, we
had suspended more than 125,000 accounts for threatening or promoting
terrorist acts. See https://blog.twitter.com/official/en_us/a/2016/
combating-violent-extremism.html. By August 2016, we announced that we
had suspended an additional 235,000 accounts for violating Twitter
policies related to the promotion of terrorism. https://
blog.twitter.com/official/en_us/a/2016/anupdate-on-our-efforts-to-
combat-violent-extremism.html. We also announced at that time that our
daily suspension records increased by more than 80 percent compared to
the previous year, and that our response time for suspending reported
accounts decreased dramatically.
We made additional improvements the following year. As we noted in
our September 2017 Transparency Report, for the reporting period
between January 1 and June 30, 2017, we suspended nearly 300,000
accounts for violations of Twitter policies prohibiting the promotion
of terrorism. Of those suspensions, 95 percent were accomplished using
our proprietary tools--up from 74 percent in 2016. Approximately 75
percent of those accounts were suspended before posting their first
Tweet. In total, between August 1, 2015 and June 30, 2017, we suspended
nearly 1 million accounts for violating Twitter rules and policies
prohibiting the promotion of violence or terrorist content.
Question 3. What is the next challenge on the Common
Vulnerabilities and Exposures (CVE) front? How do we empower smaller
platforms that the terrorists are moving to?
Answer. As noted above, we plan to continue building upon our 2017
improvements, including through collaboration with our peers and
investments in machine-learning capabilities that help us detect and
mitigate the effect on users of fake, coordinated, and malicious
automated account activity.
We have observed the expansion of malicious activity on our
platform from automated accounts to human-coordinated activity, which
poses additional challenges to making our platform safe. We are
determined to meet those challenges and have been successful in
addressing such abusive behavior in other contexts. We are committed to
leveraging our technological capabilities in order to do so again by
carefully refining and building tools that respond to signals in the
account behavior.
Those tools have also been successful at detecting and removing
terrorist content on our platform. For example, as of September 2017,
95 percent of account suspensions for promotion of terrorist activity
were accomplished using our existing proprietary detection tools--up
from 74 percent in 2016. These tools focus on indicia of violating
activity beyond the content of the Tweet.
Although they have proved successful, our efforts to address
terrorist content on our platform do not end with investments in our
proprietary detection tools. We recognize that the spread of terrorist
and extremist content online is not unique to Twitter, and we are
committed to collaborating with our industry peers to address this
shared thread. Accordingly, in June 2017, we launched the Global
Internet Forum to Counter Terrorism (the ``GIFCT''), a partnership
among Twitter, YouTube, Facebook, and Microsoft. The GIFCT facilitates,
among other things, information sharing; technical cooperation; and
research collaboration, including with academic institutions. In
September 2017, the members of the GIFCT announced a multimillion
dollar commitment to support research on terrorist abuse of the
Internet and how governments, tech companies, and civil society can
respond effectively. We are looking to establish a network of experts
that can develop these platform-agnostic research questions and
analysis that considers a range of geopolitical contexts. The GIFCT
opened a call for proposals in December 2017, and we look forward to
sharing further details of the initial projects this year.
The GIFCT has created a shared industry database of ``hashes''--
unique digital ``fingerprints''--for violent terrorist imagery or
terrorist recruitment videos or images that have been removed from our
individual services. The database allows a company that discovers
terrorist content on one of its sites to create a digital fingerprint
and share it with the other companies in the forum, who can then use
those hashes to identify such content on their services or platforms,
review against their respective policies and individual rules, and
remove matching content as appropriate, or even block extremist content
before it is posted in the first place. The database now contains more
than 40,000 hashes. Instagram, Justpaste.it, LinkedIn, Oath, and Snap
have also joined this initiative, and we are working to add several
additional companies in 2018.
As part of our work with the GIFCT, we have hosted more than 50
small companies at workshops through the Tech Against Terrorism
initiative, our partners under the UN CounterTerrorism Executive
Directorate. Twitter believes that this partnership will provide a
unique opportunity for us to share our knowledge and technical
expertise with smaller and emerging companies in the industry and for
all industry actors to harness the expertise that has been built up in
recent years.
______
Response to Written Questions Submitted by Hon. Ron Johnson to
Carlos Monje, Jr.
Question 1. Social media companies are increasingly able to remove
terrorist recruitment, incitement, and training materials before it
posts to their platforms by relying on improved automated systems.
Other than content removal, what else can be done to limit the audience
or distribution of these dangerous materials?
Answer. Twitter has been at the forefront of developing a
comprehensive response to the evolving challenge of preventing
terrorist exploitation of the Internet. We initially focused on scaling
up our own, in-house proprietary spam technology to detect and remove
accounts that promote terrorism. In early 2016, the technological tools
we had at our disposal detected approximately one-third of terrorism-
related accounts that we removed at that time. In 2017, 95 percent of
account suspensions for promotion of terrorist activity were
accomplished using our existing proprietary detection tools--up from 74
percent in 2016. Approximately 75 percent of those accounts were
suspended prior to sending their first Tweet. In total, since 2015, we
have suspended nearly a million accounts that we determined violated
our terms of service. In December 2016, for example, we took steps
toward a hash-sharing agreement with Facebook, Microsoft, and YouTube,
intended to further curb the spread of terrorist content online.
Pursuant to this agreement, the four companies created an industry
database of ``hashes''--unique digital ``fingerprints''--for violent
terrorist imagery or terrorist recruitment videos or images that we
have removed from our services. By sharing this information with each
other, we may use the shared hashes to help identify potential
terrorist content on our respective hosted consumer platforms.
In June 2017, we launched the Global Internet Forum to Counter
Terrorism (the ``GIFCT''), a partnership among Twitter, YouTube,
Facebook, and Microsoft. The GIFCT facilitates, among other things,
information sharing; technical cooperation; and research collaboration,
including with academic institutions.
In September 2017, the members of the GIFCT announced a
multimillion dollar commitment to support research on terrorist abuse
of the Internet and how governments, tech companies, and civil society
can respond effectively. We are looking to establish a network of
experts that can develop these platform-agnostic research questions and
analysis that considers a range of geopolitical contexts. The GIFCT
opened a call for proposals last month, and we look forward to sharing
further details of the initial projects early in 2018.
The GIFCT has created a shared industry database of ``hashes''--
unique digital ``fingerprints''--for violent terrorist imagery or
terrorist recruitment videos or images that have been removed from our
individual services. The database allows a company that discovers
terrorist content on one of its sites to create a digital fingerprint
and share it with the other companies in the forum, who can then use
those hashes to identify such content on their services or platforms,
review against their respective policies and individual rules, and
remove matching content as appropriate, or even block extremist content
before it is posted in the first place. The database now contains more
than 40,000 hashes. Instagram, Justpaste.it, LinkedIn, Oath, and Snap
have also joined this initiative, and we are working to add several
additional companies in 2018. Twitter also participates in the
Technology Coalition, which shares images to counter child abuse.
As part of our work with the GIFCT, we have hosted more than 50
small companies at workshops through the Tech Against Terrorism
initiative, our partners under the UN CounterTerrorism Executive
Directorate. Twitter believes that this partnership will provide a
unique opportunity for us to share our knowledge and technical
expertise with smaller and emerging companies in the industry and for
all industry actors to harness the expertise that has been built up in
recent years.
We have also focused on NGO outreach and, since 2013, and have
participated in more than 100 Countering Violent Extremism training and
events around the world, including in Beirut, Bosnia, Belfast and
Brussels and summits at the White House, at the United Nations, London,
and Sydney. Twitter has partnered with groups like the Institute of
Strategic Dialogue, the Anti-Defamation League and Imams Online to
bolster counterspeech that offers alternatives to radicalization. As a
result of that work, NGOs and activists around the world are able to
harness the power of our platform in order to offer positive
alternative narratives to those at risk and their wider communities.
Question 2. Terrorist how-to guides are protected by the First
Amendment in the United States, but violate the content policies of
many social media companies as well as the laws of some international
partner nations. What countries have laws that go beyond your company's
content policies and can you give examples of how you have worked with
those countries to de-conflict those differences?
Answer. The Twitter Rules prohibit violent threats and the
promotion or incitement of violence, including terrorism. Twitter is
committed to removing such content swiftly from the platform. In
addition, our Hateful Conduct policy is designed to protect users from
harassment on the basis of protected categories, such as race,
ethnicity, national origin, gender identity, age and religion. Examples
of hateful conduct that we do not tolerate include targeting users
with: (1) harassment; (2) wishes for the physical harm, death, or
disease of individuals or groups; (3) references to mass murder,
violent events, or specific means of violence in which or with which
such groups have been the primary victims; (4) behavior that incites
fear about a protected group; and (5) repeated and/or or non-consensual
slurs, epithets, racist and sexist tropes, or other content that
degrades someone. We also do not allow accounts whose primary purpose
is inciting harm towards others on the basis of these categories. We
have also updated our policies to clearly prohibit users who affiliate
with organizations that--whether by their own statements or activity
both on and off the platform--use or promote violence against civilians
to further their causes.
We have established channels for law enforcement agencies to
request removal of content that may be illegal under the requesting
jurisdiction's laws. Specifically, Twitter publishes global guidelines
for law enforcement personnel that explain our policies and the process
for submitting requests for content removal. See https://
help.twitter.com/en/rules-andpolicies/twitter-law-enforcement-support.
We accept requests from law enforcement agencies in countries in which
Twitter operates, and we evaluate each request and, if appropriate, we
will take action against the content at issue within the jurisdiction
from which the removal request originated. As part of our commitment to
transparency, since 2012, Twitter has published biannual Transparency
Reports, reflecting the number of requests that we have received for
user information and content removal on a per-country basis. See
https://transparency.twitter.com.
Those reports indicate the number of requests that we have received
and the number of requests with which we have complied.
______
Response to Written Questions Submitted by Hon. Maria Cantwell to
Carlos Monje, Jr.
Question 1. We have strong principles of freedom of speech, but at
the same time, we need to balance that freedom with the need to protect
against bad actors who would leverage that freedom to plan and promote
illegal acts. How can we use artificial intelligence to help us achieve
a balance between our American ideal of free speech and the need to
protect against extremist acts of terror?
Answer. Twitter recognizes that freedom of speech is a fundamental
human right and we are committed to providing a platform where users
feel safe to share their views and opinions. Twitter has a history of
facilitating civic engagement and political freedom, and we intend for
Twitter to remain a vital avenue for free expression here and abroad.
But we cannot foster free expression without ensuring trust in our
platform. We are determined to take the actions necessary to prevent
the manipulation of Twitter, and we can and must make sure Twitter is a
safe place.
Twitter's enforcement activity is designed with care to avoid, as
much as possible, having an inadvertent negative impact on free
expression. We do this by focusing our detection mechanisms primarily
on signals and behavior, rather than content. We also do this by
determining the proper use for automation, which is not always to take
direct action on accounts or content.
Twitter relies on automation, artificial intelligence, and machine
learning models to identify content for review by Twitter's teams and
to detect potentially malicious activity on the platform. For example,
when we learned that bad actors who shared terrorist propaganda on the
platform were attempting to avoid permanent suspension from the
platform by creating serial accounts, Twitter designed automated
techniques to improve our detection rates for accounts that engage in
such activity. Our efforts have been successful and our detection and
suspension rates have increased as a result.
Question 2. Outside of artificial intelligence, what other
technologies could be used to combat potential radicalization on social
media platforms? What does the implementation of those technologies
look like?
Answer. Keeping Twitter safe includes maintaining the quality of
information on our platform. Our users look to us for useful, timely,
and appropriate information. To preserve that experience, we are always
working to ensure that we surface for our users the highest quality and
most relevant content first. We are taking active steps to stop
malicious accounts, abusive conduct, and terrorist and extremist
content from spreading, and we are determined that our strategies will
keep ahead of the tactics of bad actors.
For example, Twitter is continuing its effort to detect and prevent
malicious automation by leveraging our technological capabilities and
investing in initiatives aimed at understanding and addressing
behavioral patterns associated with such accounts. In early 2017, we
launched the Information Quality initiative, an effort aimed at
enhancing the strategies we use to detect and stop bad automation,
improve machine learning to spot spam, and increase the precision of
our tools designed to prevent such content from contaminating our
platform.
We have also made significant improvements to reduce external
attempts to manipulate content visibility. These improvements were
driven by investments in methods to detect malicious automation through
abuse of our API, limit the ability of malicious actors to create new
accounts in bulk, detect coordinated malicious activity across clusters
of accounts, and better enforce policies against abusive third-party
applications.
We have also introduced changes to our Twitter Rules, including how
we correspond with those who violate them, and to our rules'
enforcement process. We recently unveiled clarifications and updates to
rules regarding hateful display names, hateful imagery, violent groups,
and content that glorifies violence, which we began enforcing in
December 2017. We made the various updates available prior to
enforcement in order to provide our users and the general Twitter
community with sufficient time to review and understand them. Twitter
is continually working to make the platform a safe place for our users.
For example, we are introducing changes to our Twitter Rules, including
how we correspond with those who violate them, and to our rules'
enforcement process.
As with most technology-based threats, the best approach is to
share information and ideas to increase our collective knowledge.
Working with the broader community, we will continue to test, to learn,
to share, and to improve, so that our product remains effective and
safe.
Another important tool against radicalization online is fostering
alternative narratives from credible voices within communities. Twitter
has invested in groups like WISE Muslim Women (NYC), Imams Online (UK)
and Hedayah (UAE).
______
Response to Written Questions Submitted by Hon. Richard Blumenthal to
Carlos Monje, Jr.
Question 1. On October 30, 2017, nineteen civil rights groups,
including Muslim Advocates, Leadership Conference on Civil and Human
Rights, NAACP, Southern Poverty Law Center, and many others, co-signed
a letter to Facebook to express concern about the hateful content on
the social media platform used to divide the country, and in
particular, to promote anti-Muslim, anti-Black, anti-immigrant, and
anti-LGBTQ animus.
Mr. Monje, how would Twitter respond to this letter if it had
received it?
Answer. Keeping Twitter safe includes maintaining the quality of
information on our platform. Our users look to us for useful, timely,
and appropriate information. To preserve that experience, we are always
working to ensure that we surface for our users the highest quality and
most relevant content first. We are taking active steps to stop
malicious accounts, abusive conduct, and terrorist and extremist
content from spreading, and we are determined that our strategies will
keep ahead of the tactics of bad actors.
We have introduced changes to our Twitter Rules, including how we
correspond with those who violate them, and to our rules' enforcement
process. We recently unveiled clarifications and updates to rules
regarding hateful display names, hateful imagery, violent groups, and
content that glorifies violence, which we began enforcing in December
2017. We made the various updates available prior to enforcement in
order to provide our users and the general Twitter community with
sufficient time to review and understand them.
Question 2. Ms. Bickert, Mr. Monje, and Ms. Downs, please provide
copies (including images, text, dates and timestamps) of all content
identified by your platforms as generated by Russian agents or the
Internet Research Agency.
Answer. We can provide the set of data on the Internet Research
Agency that we have previously provided to other congressional
committees through a secure data transfer.
Question 3. At least one of your peers in the tech industry has
voluntarily initiated an outside assessment of the civil rights impacts
of its policies and programs. In response to concerns regarding
discrimination on the home-sharing platform, AirBNB hired former U.S.
attorney general Eric Holder to help craft an anti-discrimination
policy and has promised to pursue technological innovations to guard
against future discriminatory events.
Mr. Monje, Ms. Bickert, and Ms. Downs, can you each commit to
bringing in an independent entity to conduct a thorough and public
audit of the civil rights impact of your policies and programs,
including how your platform has been used by hate groups to stoke
religious resentment and violence?
Answer. We agree that the decisions we make can have tremendous
implications for civil rights, and we take that responsibility very
seriously. We strive to create a platform that is conducive to robust
democratic debate. Twitter is built around the idea of giving voice to
people who may otherwise not be heard. From the #MeToo movement, to
#BlackLivesMatter, to countless other campaigns through Twitter's
history, we repeatedly see and are awed by the power of people who use
Twitter to drive social change, encourage diverse perspectives, and
share their stories. It is critical that our platform remains a
welcoming place for these voices.
We share your concern regarding hate groups stoking religious
resentment and violence, and we have taken significant steps over the
past year to address such activity on our platform. We introduced
changes to our Twitter Rules, including how we correspond with those
who violate them, and to our rules' enforcement process. We recently
unveiled clarifications and updates to rules prohibiting hateful
display names, hateful imagery, violent groups, and content that
glorifies violence, which we began enforcing in December 2017. In
developing these changes, we worked closely with external advisers from
around the world in the form of our Trust and Safety Council. The
Twitter Trust and Safety Council provides input on our safety products,
policies and programs. It includes safety advocates, academics,
researchers, grassroots advocacy organizations, and community groups. A
full list of our Council members is available here: https://
about.twitter.com/en_us/safety/safety-partners.html.
Question 4. A little over a year ago, Facebook, Twitter, Google,
and Microsoft announced a plan to create a joint industry database of
``content that promotes terrorism.''
Mr. Monje, Ms. Bickert, and Ms. Downs, to what extent does this
joint industry database focus on all forms of terror, including the
real terror threat presented by white supremacists?
Answer. In June 2017, we launched the Global Internet Forum to
Counter Terrorism (the ``GIFCT''), a partnership among Twitter,
YouTube, Facebook, and Microsoft. The GIFCT facilitates, among other
things, information sharing; technical cooperation; and research
collaboration, including with academic institutions. In September 2017,
the members of the GIFCT announced a multimillion dollar commitment to
support research on terrorist abuse of the Internet and how
governments, tech companies, and civil society can respond effectively.
We are looking to establish a network of experts that can develop these
platform-agnostic research questions and analysis that consider a range
of geopolitical contexts. The GIFCT opened a call for proposals last
month and we look forward to sharing further details of the initial
projects early in 2018.
The GIFCT has created a shared industry database of ``hashes''--
unique digital ``fingerprints''--for violent terrorist imagery or
terrorist recruitment videos or images that have been removed from our
individual services. The database allows a company that discovers
terrorist content on one of its sites to create a digital fingerprint
and share it with the other companies in the forum, who can then use
those hashes to identify such content on their services or platforms,
review against their respective policies and individual rules, and
remove matching content as appropriate, or even block extremist content
before it is posted in the first place. The database now contains more
than 40,000 hashes. Instagram, Justpaste.it, LinkedIn, Oath, and Snap
have also joined this initiative, and we are working to add several
additional companies in 2018. Twitter also participates in the
Technology Coalition, which shares images to counter child abuse.
We have introduced changes to our Twitter Rules, including how we
correspond with those who violate them, and to our rules' enforcement
process. We recently unveiled clarifications and updates to rules
regarding hateful display names, hateful imagery, violent groups, and
content that glorifies violence, which we began enforcing in December
2017. We made the various updates available prior to enforcement in
order to provide our users and the general Twitter community with
sufficient time to review and understand them.
For example, pursuant to our violent extremist groups policy, users
are prohibited from engaging with our platform to make specific threats
of violence or wish for the serious physical harm, death, or disease of
an individual or a group of individuals. The policy makes clear that
this prohibition includes, but is not limited to, threatening or
promoting terrorist acts. We also indicate that users may not use our
platform to affiliate with organizations that--whether by their own
statements or by their activity both on and off the Twitter platform--
use or promote violence against civilians to further their causes. We
consider extremist groups to (1) identify as such through their stated
purpose, publication, or actions; (2) have engaged in (or currently
engage in) violence and/or the promotion of violence as a means of
furthering their cause; and (3) target civilians in their acts and/or
promotion of violence.
______
Response to Written Questions Submitted by Hon. Brian Schatz to
Carlos Monje, Jr.
Question 1. Please quantify and explain Twitter's progress in
tackling the fake-user account issue. For the most recent full month
available and every month in the two years preceding provide:
number of fake accounts created
number of fake accounts removed
number of accounts hacked
number of hacked accounts restored
number of duplicate accounts created
number of duplicate accounts removed
number of inactive accounts existing
number of inactive accounts removed
number of monthly active users
average number of days a fake account remains on the
platform
Answer. In December 2017, Twitter challenged more than 6.4 million
accounts per week and prevented more than 523,000 suspicious logins per
day. These actions, in addition to protections built into our account
signup process, are part of a range of enforcement actions designed to
catch and prevent the creation and use of fake accounts. Given our
increasing focus on proactive detection (i.e., detections focused on
identifying fake or suspicious accounts at signup and preventing the
creation of those accounts in the first place), attempting to quantify
the precise number of accounts created or removed on a monthly basis
would under-represent both the scale of the challenges we face and the
scope of Twitter's activities to defend against them.
Twitter takes a wide range of actions to protect the security of
user accounts on Twitter. For example, we partner with vendors that
provide real-time information regarding data breaches on other websites
and services that may impact consumers who also have Twitter accounts
(since many consumers reuse passwords across services). In these cases,
we take proactive steps to require users to secure their accounts (for
instance, through e-mail reconfirmation and password changes)--even if
there is no evidence of any malicious activity on Twitter itself. We
also offer a range of security features, including two-factor
authentication that users can take advantage of to help ensure that
their accounts remain safe even if their passwords are compromised. We
also maintain a Vulnerability Rewards Program (also known as a ``bug
bounty'') that provides financial incentives for security researchers
to report vulnerabilities in our services so that we can quickly
identify and remediate security risks before they affect our users.
The Twitter Rules prohibit the creation of ``duplicate or
substantially similar accounts'' for spamming purposes, and we
aggressively enforce these rules as part of our overall anti-spam and
Information Quality initiatives. Accordingly, it is not possible to
disaggregate specifically duplicative accounts from the overall volume
of accounts challenged or suspended on the basis of spamming or
malicious activity on Twitter.
Twitter does not proactively remove inactive accounts from the
platform; however, we take a range of steps to ensure that inactive
accounts remain secure. Consistent with our overall approach to account
security, we may take steps to proactively lock down dormant or
inactive accounts if we receive information that suggests they are at
potential risk of compromise. We also employ real-time detections built
to detect malicious or abnormal activity from inactive accounts which
suddenly become active; an account which was previously inactive but
suddenly begins producing a high volume of automated content would
likely be challenged by our automated systems for detecting spamming or
malicious activity.
As noted above, Twitter's efforts to combat fake accounts generally
take two forms: (1) measures to challenge accounts at signup and
prevent the creation of fake accounts; and (2) measures to detect,
challenge, and prevent spamming or malicious activity in real time. We
continue to invest heavily in improving our signup process to prevent
the creation of new fake accounts. However, fake accounts which already
exist on the platform are challenged and remediated consistent with
their activity. A ``fake'' account which is completely inactive on
Twitter would likely not be caught by our detections until or unless it
begins to engage in spamming or malicious activity.
In terms of our monthly active users (``MAUs''), we have reported
the following numbers over the last two years: 330 million in the
fourth quarter of 2017; 330 million in third quarter of 2017; 326
million in second quarter of 2017; 327 million in the first quarter of
2017; 318 million in fourth quarter of 2016; 317 million in third
quarter of 2016; 313 million in second quarter of 2016; and 310 million
in the first quarter of 2016.
We report the number of active users on the platform quarterly as
part of our earnings report, which you can find here: https://
investor.twitterinc.com/results.cfm. As part of our commitment to
transparency, since 2012, Twitter has published bi-annual Transparency
Reports, reflecting the number of requests that we have received for
user information and content removal on a per-country basis. See
https://transparency.twitter.com/.
Question 2. How does a user find out if they are being impersonated
on Twitter? Does Twitter notify users proactively? Or are users
expected to monitor the platform and report to Twitter?
Answer. The Twitter Rules prohibit impersonation accounts. In
response to reports--from either the user who is being impersonated or
their authorized representatives--Twitter takes action against accounts
that deceptively impersonate another user or account. Users and non-
users alike can report impersonation accounts through a dedicated form
in our Help Center or directly from the impersonated account's profile
on the platform.
We are determined to expedite the suspension process for accounts
deemed to be impersonating other users. Once we receive a report of
potential user impersonation, we investigate the reported accounts to
determine if the accounts are in violation of the Twitter Rules, which
prohibit such profiles. Accounts determined to be in violation of our
impersonation policy, or those not in compliance with our parody,
commentary, and fan account policy, are either suspended or asked to
update their profile so they no longer violate our policies.
In addition, Twitter strictly prohibits the purchasing and selling
of account interactions on our platform. We advise our users that, by
purchasing followers, Retweets, and likes, they are often purchasing
bots, fake, or hacked accounts. Accounts found to have purchased, sold,
or promoted the selling of followers, Retweets, and likes are in
violation of the Twitter Rules and may be subject to suspension. See
https://help.twitter.com/en/rules-and-policies/twitter-rules. Twitter
has initiated a process of reviewing such activity and accounts,
including accounts that appear to impersonate an actual, existing user.
Where we determine that an account is fake (or that it impersonates an
existing user), we immediately suspend the account or require that it
complete a series of challenges before it can resume engaging with the
platform in order to verify that it belongs to an actual user.
Question 3. What is the average number of days or hours that
Twitter takes to investigate impersonation complaints before they are
resolved?
Answer. We address impersonation reports as we receive them.
Depending on the source of the report, we may request additional
information before we can take action. Because our response takes place
on a case-by-case basis and involves a fact-specific inquiry and manual
review by Twitter personnel, there is not a uniform frequency or
pattern to such enforcement actions or an average response time that we
are able to provide.
Question 4. Does Twitter have a separate, expedited process for
resolving impersonation of minors' accounts? Does Twitter know the age
of its users?
Answer. We take seriously all reported impersonations and any other
such violations of the Twitter Rules. We address all such reports with
the same expediency and strive to suspend as quickly as possible those
accounts that have been found to violate our policy against
impersonation. Children younger than 13 are not allowed to create
accounts or otherwise use our platform.
Question 5. Even a relatively small number of fake users can have
an outsized impact in misleading voters. During the 2016 general
election, there were 529 accounts that pushed incorrect, ``text-to-
vote'' tweets. A similar phenomenon happened during the Virginia
gubernatorial election in 2017. How many users saw these tweets? How
many users interacted with these tweets? Who were behind these two
voter suppression campaigns? If it is still unknown, is Twitter working
with law enforcement to identify the accounts' creators?
Answer. During the period leading up to the 2016 election, Twitter
labeled as ``restricted pending deletion'' a total of 918 such Tweets
from 529 Twitter accounts. Assigning that label to a Tweet requires the
user to delete the Tweet before the user is permitted to continue using
the account and engage with the platform. So long as the Tweet has that
label--and until the user deletes the tweet--the Tweet remains
inaccessible to and hidden from all Twitter users. The user is blocked
from Tweeting unless and until he or she deletes the labeled Tweet.
Twitter's review indicates that the 918 labeled Tweets (a) were
viewed 222,111 times (an average of 242 views per tweet (also known as
``impressions'')); (b) liked by 10 users; (c) Retweeted by 801 users;
and (d) received 318 replies.
In addition to labeling the individual Tweets, Twitter permanently
suspended 106 accounts that were collectively responsible for 734
``vote-by-text'' Tweets. Twitter's review of the suspended accounts'
history indicates that those 734 Tweets (a) were viewed 162,656 times;
(b) liked by 75 users; (c) Retweeted by 603 users; and (d) received 153
replies.
Twitter identified an additional 286 Tweets from 239 Twitter
accounts with the relevant voting-related content upon which Twitter
did not take any action. Twitter determined that those Tweets
propagated the content in order to refute the message and alert other
users that the information is false and misleading. Those Tweets
generated significantly greater engagement across the platform compared
to the 918 Tweets that Twitter labeled and the 106 accounts that
Twitter suspended. Specifically, the 286 refuting tweets (a) were
viewed 1,634,063 times (an average of 5,714 impressions per Tweet); (b)
liked by 358 users; (c) Retweeted by 11,620 users; and (d) Received 611
replies.
During the period leading up to the Virginia gubernatorial
election, Twitter received reports about an account that posted similar
Tweets. We suspended the user upon receiving those reports. Our action
against this account is consistent with the approach we took against
illegal voter suppression Tweets during the 2016 election. Here,
however, and well before our manual review of the account's activity
resulted in its permanent suspension, Twitter's automated spam
detection systems identified malicious behavior originating from this
account and took action to hide that user's Tweets from appearing in
searches and counting toward trends. Those automated systems, which we
continue to invest in as part of our Information Quality initiative,
help us address emerging malicious behavior even before a human
reviewer can assess the content.
Question 6. Have there been any other voter suppression campaigns
in elections following the 2016 general election--in the United States
or abroad?
Answer. Other than the voter suppression Tweets discussed in
Question 5, Twitter is not aware of similar voter suppression
campaigns.
Question 7. Twitter has found 3,814 accounts so far linked with the
Internet Research Agency (IRA) in a ``relevant time period'' to the
2016 election. Of the 3,814 IRA-linked accounts, how many of these were
automated bots? How many were trolls? How many were impersonations of
real American users' accounts? If so, do you plan to notify the
accounts' owners?
Answer. We were able to determine that 307 of the 3,814 accounts
that we have previously identified as linked to the IRA appear to be
automated accounts. As we reported in our January 19 and January 31,
2018, blog posts, we notified U.S.-based users with an active e-mail
address if they had directly engaged with or actively followed one of
the 3,814 IRA-linked accounts we had identified. See https://
blog.twitter.com/official/en_us//company/2018/2016-election-
update.html. In total, approximately 1.4 million Twitter users received
a notification from Twitter.
In addition, the Twitter Rules prohibit impersonation accounts. In
response to reports--from either the user who is being impersonated or
their authorized representatives--Twitter takes action against accounts
that deceptively impersonate another user or account. Users and non-
users alike can report impersonation accounts through a dedicated form
in our Help Center or directly from the impersonated account's profile
on the platform.
We are determined to expedite the suspension process for accounts
deemed to be impersonating other users. Once we receive a report of
potential user impersonation, we investigate the reported accounts to
determine if the accounts are in violation of the Twitter Rules, which
prohibit such profiles. Accounts determined to be in violation of our
impersonation policy, or those not in compliance with our parody,
commentary, and fan account policy, are either suspended or asked to
update their profile so they no longer violate our policies.
Question 8. What was the ``relevant time period'' under which
Twitter found the 3,814 IRA-linked accounts? What was the justification
for using this time period?
Answer. Our search for IRA-linked accounts was not limited to a
particular time period. We conducted a thorough review of data
available to us in order to identify any account we could reasonably
link to accounts we had previously determined to be associated with the
IRA either by our independent research or through information provided
to us by third parties (as discussed in greater detail below in
response to Question 9).
Question 9. The IRA has been targeting American users well before
the 2016 election--for instance during Russia's invasion of Crimea in
2014. When did Twitter first detect IRA activity on the platform? When
did Twitter first detect IRA activity specifically targeting American
users? From the first time Twitter detected IRA activity to the most
recent time period available, how many accounts has the IRA created on
Twitter?
Answer. Twitter first identified and suspended IRA-linked accounts
in June 2015 following a June 2, 2015, New York Times article about the
IRA and its well-known online Kremlin propaganda activities. The
Agency: From a Nondescript Office Building in St. Petersburg, Russia,
an Army of Well-Paid ``Trolls'' Has Tried to Wreak Havoc All Around the
Internet--and in Real-Life American Communities, NYTimes.com, available
at https://www.nytimes.com/2015/06/07/magazine/the-agency.html?_r=0.
Twitter suspended a total of 467 those accounts within days of the
article's publication. Those accounts were suspended for violating
Twitter's spam rules.
On August 8, 2017, a third-party security firm provided Twitter
with a report listing Russian-linked accounts that were suspected of
being associated with the IRA. Twitter commenced its review of that
list immediately. Based on that review, between August 8 and August 10,
2017, Twitter suspended 473 accounts listed in the report for engaging
in spam activity prohibited under the Twitter Rules.
Also on August 8, 2017--but separately from the security firm
report--Twitter's Information Security team received from Facebook a
list of e-mail addresses, which Facebook indicated were connected to
the IRA. Twitter reviewed that list and identified 22 accounts that
matched the e-mail addresses that Facebook provided. All 22 accounts
had already been suspended or were subsequently suspended. On August
10, 2017, Facebook shared with Twitter account-related information for
one additional Facebook account. Twitter identified 181 accounts that
were linked or related to the 23 accounts that Facebook shared with us,
bringing the total of Russian-related accounts under examination to
204. As of August 22, 2017, all but 14 of those accounts had already
been suspended or set in read-only mode pending phone number
verification. Following a manual review of all 204 accounts, three were
determined to be non-automated, legitimate users; those accounts remain
active on the platform.
Finally, in connection with our retrospective review of Russian
interference in the 2016 U.S. election through activity on our
platform, we identified additional accounts linked to the IRA, bringing
the total number of such accounts to 2,752. And, as we reported in our
January 19, 2018, update to Congress, through our continued analysis,
we identified an additional 1,062, for a total of 3,814 IRA-linked
accounts. All 3,814 IRA-linked accounts were suspended for Terms of
Service violations, and all but a few compromised accounts that have
subsequently been restored to their legitimate account owners remain
suspended.
Question 10. Between September 1 and November 15, 2016, 175,993
tweets from IRA-linked accounts received a total of 351,632,679
impressions within the first seven days after posting. And these tweets
were retweeted 4,509,781 times. Out of the total number of impressions,
how many were via organic reach vs. promoted or paid reach? How do
these metrics compare to Twitter's typical performance benchmarks?
Please give specific numbers, for instance, the number of impressions
that an average tweet got between September 1 and November 15, 2016,
for both organic reach vs. promoted or paid reach.
Answer. None of the Tweets in question were promoted.
The number of impressions on a Tweet can vary substantially
depending on the content in question, its proliferation on Twitter and
through other online channels, and the accounts responsible for posting
or sharing it. For example, a user who has many followers will
generally receive more impressions per Tweet than a user with a smaller
number of followers. Similarly, a news organization embedding a Tweet
in an article would likely generate an increase in the number of
impressions--compared to Tweets with similar characteristics which were
not embedded--given the additional impressions from off-platform views.
Due to the variability of accounts, content, and sharing patterns on
Twitter, we do not have a general or baseline measure of Tweet
performance across different contexts.
Question 11. How many users did the aforementioned 351,632,579
impressions reach exactly? What technological tools (e.g., bots, third-
party applications) were used to drive up the number of impressions?
Answer. Impression counts do not drive content ranking or content
display and are not publicly visible in the Twitter product. We are not
aware of specific attempts to manipulate the number of impressions on
content--and believe that such attempts, if made, would not have
appreciable impact on the Twitter platform.
Due to the variability of accounts, content, and sharing patterns
on Twitter, there is not a one-to-one correlation between impressions
and Twitter users. Further, impressions will include the number of
times a Tweet is viewed across a range of products (such as the Twitter
website, mobile apps, or Tweets embedded on an external website),
including by logged out users about whom Twitter has limited
information. We do not have a way of measuring impressions on Tweets
from any non-Twitter applications or tools, and impressions on content
via third-party applications would not contribute to the overall
impression count on a Tweet.
Question 12. Twitter removed 935,000 accounts for terrorism
promotion. How many of these were automated accounts?
Answer. We do not track suspensions for promotion of terrorism in a
manner that specifically flags automation. As we have previously noted
in our blog posts on this subject, however, our proprietary spam-
fighting tools offer significant assistance in the fight against the
dissemination of terrorist propaganda online. We are able to leverage
our spam-fighting tools to stop the spread of this content given the
prevalence of spam signals such as indicia of automation and attempts
at ban evasion.
Question 13. Please share more details about the medium of content
in tweets identified with active measures. What percentage of these
tweets are text only? Contain static image? Video? How many of these
tweets embed an external hyperlink?
Answer. Of the 175,993 Tweets posted by the 3,814 IRA-linked
accounts during the election time period: approximately 92,000 embedded
external hyperlinks; 114,000 were text only; 58,000 contained static
images; and 3,000 contained videos.
Question 14. Once an account is flagged by an algorithm for
removal, what is the average amount of time before the account is
removed from public view? After the account is removed from public
view, what is the average amount of time before the data is deleted?
Why is the data deleted rather than just removed from public view?
Answer. Twitter systems are designed to take different steps with
respect to different types of malicious activity on the platform. Our
systems cast a wide net to detect and label malicious accounts or
malicious activity, and we may take additional steps to confirm the
accuracy of those processes before removing the content from public
view. Those additional steps are designed to minimize false positives
and inadvertent action against users who we ultimately determine not to
be in violation of our policies.
In other circumstances, where our automated tools operate with high
precision, we may take more immediate action. Such circumstances
include, but are not limited to, instances where we detect child sexual
exploitation or malware on our platform. As a general matter, content
removal on Twitter reflects a careful balancing of platform protection
and individual user rights, and it requires constant evaluation and
assessment to refine and improve upon our existing methodologies. Such
frequent iterations and reexamination of our actions are critical to
enhancing and improving our detection tools.
Data from accounts that have been suspended from the Twitter
platform are retained in order to inform and improve existing detection
systems and for responding to requests from law enforcement.
Question 15. What is Twitter's data retention policy, as it relates
to the suspension of accounts for violating Twitter's terms of service?
Does this policy apply to all parties--such as independent researchers,
users, advertisers, and data brokers--in the same way?
Answer. Data from accounts that have been suspended from the
Twitter platform and are no longer publicly visible are retained in
Twitter's internal systems for safety and security purposes, including
to inform and improve existing detection systems.
Through our API, we give developers and other third parties access
to subsets of public Twitter content. Access to this publicly available
data through our API is conditioned on acceptance of our policies,
including the requirement that developers not use the API to undertake
activities with respect to content that users have removed from the
platform. Examples of situations this policy is designed to address
include a parent deciding to remove pictures of their children if they
have safety concerns or a college student removing Tweets as they
prepare to apply for jobs. This is a long-standing Twitter policy.
Question 16. Given the importance of collaborating with third-party
or independent researchers to prevent further interference by Russia,
will Twitter be updating its data retention policy?
Answer. Twitter is committed to addressing how information spreads
online, crosses between platforms and services, and raises the
attention of voters, elected officials, and the media. Consistent with
our commitment to transparency, we recognize that our efforts at
addressing this issue must be part of a broader discussion about how
important societal conversations take place online and how Russia has
leveraged digital services, including Twitter, to interfere with U.S.
elections. Indeed, cooperation to combat this challenge is essential.
We cannot defeat this novel, shared threat alone. As with most
technology-based threats, the best approach is to share information and
ideas to increase our collective knowledge. Working with the broader
community, we will continue to test, to learn, to share, and to
improve, so that our product remains effective and safe.
Twitter looks forward to continuing to work closely with third
party researchers consistent with its commitment to transparency and
improvement in these critical areas for democracy and elections. Last
year, Twitter offboarded Russia Today and Sputnik as advertisers on our
platform and dedicated the $1.9 million those accounts had spent on
advertising globally on the platform to research in these areas.
Question 17. Do hashes, as mentioned during the hearing, exist only
for static images? What about video content? If hashes for video
content are not yet fully deployed, please share the timeline to do so.
Answer. Hashes are deployed for static images, in close partnership
with industry groups, to fight terrorist content as well as child
sexual exploitation online. We continue to work with peer companies and
industry groups to expand on hash sharing partnerships, including
through the potential use of hashes for video content. It is critical
that these programs are deployed urgently, but also with careful cross-
industry collaboration and buy-in, to maximize the potential for shared
success in fighting these challenges.
Question 18. Last year, Twitter shared that 220,000 malicious
applications were suspended for abuse of the Twitter application
programming interface (API). How does Twitter define API abuse? Who
created, managed, or plugged in the malicious applications to the
Twitter API?
Answer. Please see the answer to question 19 below.
Question 19. How often does Twitter monitor the network of
applications that use the Twitter API? Is suspension done on a rolling
basis? Is it solely up to Twitter employees (rather than its users or
developer community) to identify malicious applications?
Answer. Any developer signing up for access to Twitter's API is
required to agree to the Developer Agreement and Policy prior to
obtaining access. See https://t.co/devpolicy. The Developer Agreement,
Developer Policy, Automation Rules (https://t.co/automation), Display
Requirements (https://developer.twitter.com/en/developer-terms/display-
requirements), and Geo Guidelines (https://developer.twitter.com
/en/developer-terms/geo-guidelines) collectively make up the body of
rules that govern developers' use of our platform. We also make
available to developers additional guidance regarding how to interpret
and implement these guidelines, either via our Developer website (e.g.
https://developer.twitter.com/en/developer-terms/more-on-restricted-
use-cases) or via the Twitter Community forums (https://twittercom
munity.com/c/dev-platform-feedback/rules-and-policies).
At this time, subject to limits on use as well as automated and
human review for policy compliance, any user with a valid Twitter
account can register for access to Twitter's API. As we noted in our
January 19, 2018, blog post, we are currently working on introducing an
improved developer onboarding process to better manage the use cases
for developers building on Twitter's platform. See https://blog
.twitter.com/developer/en_us/topics/tools/2017/introducing-twitter-
premium-apis
.html.
Twitter enforces the terms of the Twitter API through multiple
channels. Through Twitter's compliance team, we investigate and address
instances of potential violations of Twitter's policies with respect to
Twitter's commercial data products. We also work with third parties and
other stakeholders to investigate and address other reported abuses of
our APIs. As with any community, our enforcement mechanisms are best
served by a combination of affirmative steps and reactive
investigations that we take to address concerns raised by community
members in order to protect our users and customers.
Question 20. In terms of dollars and percentage of annual revenue,
how much is Twitter now spending on preventing foreign interference
with our elections? What was the figure in the election cycle leading
up to November 2016? What is the projected spend leading up to November
2018?
Answer. As we stated in our February 8, 2018, shareholder letter,
Twitter continues to invest considerable resources in our Information
Quality efforts. See http://files.shareholder.com/downloads/AMDA-
2F526X/5990710870x0x970892/F9B4F616-659A-454B-89C6-28480DA53CCA/
Q4_2017_Shareholder_Letter.pdf
Based on the understanding we have gained from our retrospective
review, we have also established an internal, cross-functional team
dedicated to addressing election-related instances of abuse on Twitter,
as we discussed with the Senate Commerce Committee in January.
The election team will address this challenge in a number of ways.
Among other things, to detect and promptly address impersonation
attempts, the team will verify major party candidates for all statewide
and Federal offices, as well as all major national party accounts. In
addition to monitoring and enforcing the Twitter Terms of Service and
Twitter Rules, the election team will cooperate and communicate with
Federal and state election officials to swiftly escalate and address in
real time attempts at election interference. And consistent with
Twitter's commitment to curtailing malicious automation, spam, and
false accounts on our platform, the election team will focus on
deploying our proprietary tools specifically to detect and stop
malicious election-related activity.
Question 21. Congress will judge success not by Twitter's efforts
but by its results. How will Twitter measure its success? Will Twitter
be conducting an audit after November 2018? When will the results be
shared?
Answer. Twitter will continue to work closely with Congress, our
industry peers, civil society, experts, and law enforcement agencies to
consider these challenges and novel threats for Twitter, the Internet,
and society as a whole. We are committed to addressing, mitigating, and
ultimately preventing any future attempts to interfere in elections and
the democratic process, and to doing so in the most transparent way
possible. We look forward to continuing to provide information to the
Committee about malicious activity we detect on our platforms and the
measures we take to address such activity.
______
Response to Written Questions Submitted by Hon. Tammy Baldwin to
Carlos Monje, Jr.
Question 1. Mr. Monje, you note in your testimony that Twitter's
effective anti-spam technology--which the company will be using to
address networks of malicious bots attempting to interfere with the
2018 U.S. midterms--is ``in-house proprietary.'' At the same time, you
tout Twitter's record of and support for industry collaboration.
Wouldn't that collaboration--as well as our shared fight against
Russian active measures--be improved by Twitter sharing its anti-spam
technology with other social media companies? Do you share that
technology with others?
Answer. We agree that collaboration with our industry peers and
civil society is critically important to addressing common threats and
that it has been successful in meeting shared challenges. In June 2017,
for example, we launched the Global Internet Forum to Counter Terrorism
(the ``GIFCT''), a partnership among Twitter, YouTube, Facebook, and
Microsoft. The GIFCT facilitates, among other things, information
sharing; technical cooperation; and research collaboration, including
with academic institutions. In September 2017, the members of the GIFCT
announced a multimillion dollar commitment to support research on
terrorist abuse of the Internet and how governments, tech companies,
and civil society can respond effectively. We are looking to establish
a network of experts that can develop these platform-agnostic research
questions and analysis that consider a range of geopolitical contexts.
The GIFCT opened a call for proposals in December 2017 and we look
forward to sharing further details of the initial projects this year.
The GIFCT has created a shared industry database of ``hashes''--
unique digital ``fingerprints''--for violent terrorist imagery or
terrorist recruitment videos or images that have been removed from our
individual services. The database allows a company that discovers
terrorist content on one of its sites to create a digital fingerprint
and share it with the other companies in the forum, who can then use
those hashes to identify such content on their services or platforms,
review against their respective policies and individual rules, and
remove matching content as appropriate, or even block extremist content
before it is posted in the first place. The database now contains more
than 40,000 hashes. Instagram, Justpaste.it, LinkedIn, Oath, and Snap
have also joined this initiative, and we are working to add several
additional companies in 2018. Twitter also participates in the
Technology Coalition, which shares images to counter child abuse.
Because each platform is unique, there are many elements of our
coordinated work that do not translate easily across platforms.
Although we share with other companies our approach to addressing
shared threats, including certain signals that we use to identify
malicious content, solutions applicable to the Twitter platform are not
always applicable to other companies. We describe our tools as ``in-
house and proprietary'' to distinguish them from tools that are
developed by and licensed from third-party vendors.
Question 2. Mr. Monje, ensuring Americans know the source of
political advertising on social media is one of the best ways to combat
interference in U.S. elections by foreign actors. Put simply, we should
apply to social media the same rules that apply to TV and print media.
Please tell us more about Twitter's Ads Transparency Center, including
the status of implementation and any hurdles you foresee.
Answer. Twitter's approach to greater transparency in political
advertising centers on two components: a new electioneering policy and
an industry-leading Transparency Center. We expect to roll out the new
policy in the U.S. during the first quarter of 2018. To make it clear
when a user is viewing or engaging with content considered to be an
electioneering ad, our policy will require that advertisers that meet
the definition of electioneering to identify their campaigns as such.
We will also change the interface of such ads and include a visual
political ad indicator (see, e.g., Fig. 1 below).
Twitter's definition of electioneering ads will be derived from the
FEC regulations' definition of that term, which includes any broadcast,
cable, or satellite communication that refers clearly to a candidate
for Federal office, is published 60 days before a general election or
30 days before a primary, convention, or caucus, and is targeted to the
relevant electorate (if the candidate is running for Congress).
The goal of the Transparency Center is to offer the public
increased visibility into all advertising on the platform and to
provide users with tools to share feedback with us. With respect to
electioneering ads and the Transparency Center, we intend to better
enable users and outside parties to conduct their own research or
evaluation regarding particular ads. Electioneering ads information
accessible through the Transparency Center will include, among other
things, the identity of the organization funding the campaign, all ads
that are currently running or have run on Twitter, campaign spend, and
targeting demographics for specific ads or campaigns. We plan to launch
the Transparency Center as soon as feasible after rolling out our
electioneering policy in the first quarter of 2018, and we are
continuing to refine the tools we will make available in conjunction
with launching the Transparency Center to ensure the best experience
for our users.
Question 3. In the context of extremist content, I would like to
learn more about each company's policy for proactively reporting users
to law enforcement. I understand your companies evaluate and respond to
law enforcement requests for information, but what framework do you use
to proactively report terrorist-related content to authorities,
including any identifying information of the user? For example, if you
use a standard of imminent harm, how do you define and apply it,
particularly in a threat environment where terrorist organizations
often call on recruits to attack carefully planned targets of
opportunity, rather than to launch an immediate, indiscriminate attack?
Answer. Twitter actively works on establishing and maintaining
close relationships with law enforcement by providing ongoing training
opportunities and through recurring meetings that allow for urgent,
proactive outreach in the event Twitter becomes aware of imminent harm
related to content on the platform. In these circumstances, it is of
paramount importance that relationships allow for immediate connection
at any time, day or night. This includes regularly working with the
F.B.I. on domestic issues and U.S. legal attaches across the globe to
assist in vetting complex international situations. The circumstances
leading to proactive reporting will depend on the nature of the issue.
Typically, when we become aware of content on one of Twitter's products
that contains a serious immediate threat or an actual depiction of live
physical violence, whether of harm to one's self or to other persons,
Twitter will proactively contact law enforcement or other appropriate
authorities so they can properly assess the nature of the situation. As
noted in Twitter's Law Enforcement Guidelines, our emergency request
protocols are covered 24/7, every day of the year. Twitter evaluates
emergency disclosure requests on a case-by-case basis in compliance
with relevant law (e.g., 18 U.S.C. Sec. 2702(b)(8)). If we receive
information that provides us with a good faith belief that there is an
exigent emergency involving the danger of death or serious physical
injury to a person, we may provide information necessary to prevent
that harm, if we have it. Where law enforcement believes identifying
information is warranted, we can immediately provide such information
through our emergency response protocols.
Question 4. I would like to hear from the companies whether they
support implementing Mr. Watts's recommendations to: first, fully
certify the authenticity of all users--in other words, ensure that each
user is a real person; and second, eliminate social bot networks to
reduce automated broadcasting of disinformation.
Answer. We have dramatically improved our ability to identify and
disrupt social bot networks, as explained elsewhere in these responses.
Using malicious automation is a violation of our terms of service and
we action accounts with increasing effectiveness.
Twitter is committed to defending the voices of our users,
including those who rely on anonymous or pseudonymous accounts to do
their work safely. Journalists, activists, political dissidents,
whistleblowers, and human rights practitioners have been imprisoned,
tortured and worse on the basis of their personally identifiable
information online. This decision to protect their identity was made in
consultation with leading NGOs working on the front lines of these
issues worldwide. We seek to protect them on Twitter.
Question 5. What are the indicators that you use to identify a
Russian disinformation account, whether from the Kremlin's so-called
Internet Research Agency or an associated group of hackers or trolls,
and what thresholds must be met to disable an account?
Answer. We relied on a number of different sources in order to
identify accounts linked to the IRA. Twitter first identified and
suspended IRA-linked accounts in June 2015 following a June 2, 2015,
New York Times article about the IRA and its well-known online Kremlin
propaganda activities. The Agency: From a Nondescript Office Building
in St. Petersburg, Russia, an Army of Well-Paid ``Trolls'' Has Tried to
Wreak Havoc All Around the Internet--and in Real-Life American
Communities, NYTimes.com, available at https://www.nytimes.com/2015/06/
07/magazine/the-agency.html?_r=0. Twitter suspended a total of 467
those accounts within days of the article's publication. Those accounts
were suspended for violating Twitter's anti-spam rules.
On August 8, 2017, a third-party security firm provided Twitter
with a report listing Russian-linked accounts that were suspected of
being associated with the IRA. Twitter commenced its review of that
list immediately. Based on that review, between August 8 and August 10,
2017, Twitter suspended 473 accounts listed in the report for engaging
in spam activity prohibited under the Twitter Rules.
Also on August 8, 2017--but separately from the security firm
report--Twitter's Information Security team received from Facebook a
list of e-mail addresses, which Facebook indicated were connected to
the IRA. Twitter reviewed that list and identified 22 accounts that
matched the e-mail addresses that Facebook provided. All 22 accounts
had already been suspended or were subsequently suspended. On August
10, 2017, Facebook shared with Twitter account-related information for
one additional Facebook account. Twitter identified 181 accounts that
were linked or related to the 23 accounts that Facebook shared with us,
bringing the total of Russian-related accounts under examination to
204. As of August 22, 2017, all but 14 of those accounts had already
been suspended or set in read-only mode pending phone number
verification. Following a manual review of all 204 accounts, three were
determined to be non-automated, legitimate users; those accounts remain
active on the platform.
Finally, in connection with our retrospective review of Russian
interference in the 2016 U.S. election through activity on our
platform, we identified additional accounts linked to the IRA, bringing
the total number of such accounts to 2,752. And, as we reported in our
January 19, 2018, update to Congress, through our continued analysis,
we identified an additional 1,062, for a total of 3,814 IRA-linked
accounts. All 3,814 IRA-linked accounts were suspended for Twitter
Terms of Service violations, and all but a few compromised accounts
that have subsequently been restored to their legitimate account owners
remain suspended.
______
Response to Written Questions Submitted by Hon. Catherine Cortez Masto
to Carlos Monje, Jr.
Question 1. During my time as the Attorney General for the State of
Nevada, I saw too many instances of sex trafficking cases involving
child victims that were dismissed because the conduct occurred online
or through social media. So that's why I'm a strong supporter of the
Stop Enabling Sex Traffickers Act of 2017 (SESTA), which clarifies the
Communications Decency Act (CDA) to allow state Attorneys General to
retain their jurisdiction to prosecute those who facilitate human
trafficking. We know that trafficking is happening online and on social
media, and SESTA is the only current legislative proposal that provides
sufficient deterrence to traffickers by providing the necessary tools
for successful prosecutions. As a former prosecutor, I know what it
will take to successfully prosecute those who engage in sex trafficking
through social media and other websites, and that's why I believe that
the House version of SESTA doesn't go far enough to give prosecutors
the tools they need to protect sex trafficking victims. I hope that
your organizations all agree that victims of sex trafficking deserve
meaningful protections and justice.
If so, I'd like to hear whether you will continue to support SESTA
over the weaker U.S. House version of the bill.
Answer. The answer to Question 1 has been provide in response to
Question 2 below.
Question 2. I was glad to hear that the Internet Association
supports SESTA, and I'd like to know what else your organization is
doing to address concerns about sex trafficking occurring on your
platforms and helping us pass this important legislation in the Senate.
Answer. Human trafficking or the facilitation of such activities
have no place on Twitter. Twitter is deeply committed to working
together with Congress, law enforcement, victims groups, and NGOs to
combat such heinous crimes. To that end, we take a multifaceted
approach to this issue.
We do not tolerate child sexual exploitation on Twitter. When we
are made aware of links to images of or content promoting child sexual
exploitation, they will be removed from the site without further notice
and reported to The National Center for Missing & Exploited Children
(or ``NCMEC''). We also permanently suspend accounts promoting or
containing links to child sexual exploitation or engaging in child sex
trafficking. Furthermore, we take measures to discourage repeat signups
from these users.
More broadly, sharing explicit sexual images or videos of someone
online without their consent is a violation of their privacy and one of
the most serious violations of the Twitter Rules. We will suspend any
account we identify as the original poster of intimate media that has
been produced or distributed without the subject's consent. We will
also suspend any account dedicated to posting this type of content.
Twitter works with NGOs and victims' groups which focus on counter-
child sex trafficking measures, including Love146, Thorn and NCMEC--all
of which serve on our Twitter Trust Council.
Twitter is also a member of and serves as the current chair of the
Technology Coalition, powered by leaders in the Internet services
sector. Formed in 2006, the Coalition's vision is to eradicate online
child sexual exploitation. The group's strategy is to sponsor the
development of technology solutions that disrupt the ability to use the
Internet to exploit children or distribute child pornography. The
Technology Coalition works with the NCMEC and its sister agency, the
International Centre for Missing & Exploited Children (the ``ICMEC''),
to identify and propagate technology solutions that create effective
disruption.
In addition, the Technology Coalition seeks and creates platforms
for collaboration with the private and public sectors for the creation
of standards, the sharing of best practices, and similar initiatives
that advance the fight against online sexual exploitation of children.
The Technology Coalition's efforts are structured with a view
toward balancing the privacy interests of Internet users with its
mission to eradicate online child sexual exploitation.
We also have close working relationships with law enforcement and
expeditiously review and action legal requests. Guidelines intended for
law enforcement authorities seeking information about Twitter accounts
are posted on our website. Information concerning requests to withhold
content on Twitter is available. More general information is available
in our Privacy Policy, Terms of Service, and Twitter Rules.
Twitter has been an active participant in Congress' recent efforts
to address human trafficking. Our team has held dozens of meetings with
lawmakers to engage in meaningful dialogue about how we can work
together to meet this challenge.
Question 3. Over the past few months, our country has been
reckoning with some hard truths about the way that women and minorities
are treated in the workplace. And I think this is a moment for all
types of organizations, including tech giants like the ones represented
here, to take a clear-eyed accounting of their culture and practices,
to take responsibility for what hasn't worked, and to renew their
commitments to make meaningful improvements. The Equal Employment
Opportunity Commission's 2016 report on ``Diversity in High Tech''
found that women, African Americans, and Hispanics are all represented
at significantly lower levels in high tech than in private industry as
a whole. And while recent internal studies at Facebook and Google have
showed some progress in the hiring of women, there has not been equal
improvement in the representation of people of color and other
underrepresented groups.
What technically qualifies as diversity to your organization?
Answer. Twitter takes a holistic approach to the way we define
diversity. At Twitter, diversity includes the hiring, retention and
advancement of individuals with protected characteristics (e.g., age,
gender, race, sexual orientation, etc.). However, these are not the
only characteristics that impact our diversity profile. We value the
uniqueness of our people. This means that we challenge ourselves to
hire talented people who have different ideas, perspectives, and
approaches to the work. In our experience, diversity allows employees
to feel comfortable sharing the intersectionality of who they are as
individuals, which leads to the exchange and development of cutting-
edge ideas.
Question 4. How is your company working to address issues of
discrimination in your own workforces?
Answer. At Twitter we value inclusion and diversity. For that
reason, we have a multi-tiered approach to addressing discrimination in
the workplace. Some of the methods we use include, but are not limited
to, clear and accessible policies against discrimination; consistent
messaging to employees that discrimination is not tolerated at Twitter;
an active Inclusion & Diversity team; annual training for our people
managers to ensure they are equipped to promptly respond to any issue
that may be perceived as discriminatory; a dedicated hotline for
employees to report concerns of discrimination; and an Employee
Relations (``ER'') Team that is responsible for prompt, thorough and
neutral investigation of discrimination complaints.
We also sponsor Business Resource Groups (``BRGs''), which are an
excellent support system within Twitter to foster awareness, respect,
and inclusion within the workplace. These groups include Blackbirds
(African American), Twitter Alas (Hispanic/LatinX), Twitter Women, and
Twitter Open (LGBTQ). BRGs serve as a sounding board around strategic
diversity objectives within the organization to help create a more
inclusive work environment.
In addition, Twitter's Diversity Advisory Council offers
suggestions and advice on strategies. The Council provides a forum for
sharing best practices across the tech industry and in the field of
diversity and inclusion. We are continuously assessing the work
environment at Twitter, conducting periodic pulse surveys of employees
and hosting lunches for women and people of color.
Question 5. Do you believe those efforts are sufficient?
Answer. Twitter's current anti-discrimination practices are
targeted to foster a fair, inclusive, and healthy environment, and they
exceed the minimum legal standards in every jurisdiction in which we
operate. To that end, while we believe that our efforts are sufficient,
we will continue to work to exceed minimal requirements and strive to
be an industry leader in this area.
Question 6. I've seen that Facebook works to make their labor
diversity information public, can you provide a status on your labor
figures, or commit to sharing those with the Committee and the public?
Answer. We commit to sharing labor diversity information with the
Committee and the public. The latest report can be found here: https://
blog.twitter.com/en_us/topics/company/2017/building-a-more-inclusive-
twitter-in-2016.html.
Question 7. We know that so-called talent pipelines are not the
only obstacle to achieving a diverse workforce, and that discrimination
and harassment go hand in hand, distorting the operation of workplace
meritocracies. This is a moment when many victims of sexual assault and
harassment are bravely coming forward about their experiences, allowing
us to get a better sense of the true scope and effects of this
behavior. Persistent harassment, and the workplace culture that
tolerates, ignores, or even encourages such harassment, pushes people
out of their workplaces, stalls or derails promising careers, and
discourages some from pursuing certain opportunities altogether.
What is your company doing to evaluate the impact of harassment in
your workforces?
Answer. Twitter conducts periodic and annual reviews and analysis
of Employee Relations data.
Question 8. How are you working to create a culture where
harassment is no longer tolerated?
Answer. Twitter maintains and enforces a clear and accessible
policy against sexual harassment. We foster an environment that
encourages employees to report concerns and maintain proper complaint
procedures for reporting concerns. Employee Relations promptly
investigates all allegations of sexual harassment. We provide annual
sexual harassment training for all managers and provide team and
individualized training on an as-needed basis.
Question 9. What more could you be doing to be a positive example
for other companies and industries?
Answer. Building and maintaining a diverse workforce and fostering
a culture of inclusion is a top priority for our company. Recognizing
that our platform serves as a powerful tool to educate and facilitate
conversations around these topics, we recently launched the Twitter
handle, @TwitterTogether, which allows Twitter users, industry peers,
and the public to learn about internal and external activities
happening at Twitter Inc.
Question 10. Last October, Facebook announced that it would be
improving transparency for all ads run on its platform, including by
requiring political advertisers to include a disclaimer telling viewers
who paid for an ad, and allowing viewers to see all the ads a page is
running, even those that aren't targeting them. Twitter also announced
similar measures. Although these policies were announced in response to
Russia using social media to interfere in our elections, it seems these
transparency measures could help shine a spotlight on other forms of
influence campaigns by extremists or terrorists.
Answer. Can you provide an update on the status of these measures?
Twitter's approach to greater transparency in political advertising
centers on two components: a new electioneering policy and an industry-
leading Transparency Center. We expect to roll out the new policy in
the U.S. during the first quarter of 2018.
To make it clear when a user is viewing or engaging with content
considered to be an electioneering ad, our policy will require that
advertisers that meet the definition of electioneering to identify
their campaigns as such. We will also change the interface of such ads
and include a visual political ad indicator (see, e.g., Fig. 1 below).
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
The goal of the Transparency Center is to offer the public
increased visibility into all advertising on the platform, and to
provide users with tools to share feedback with us. With respect to
electioneering ads and the Transparency Center, we intend to better
enable users and outside parties to conduct their own research or
evaluation regarding particular ads. Electioneering ads information
accessible through the Transparency Center will include, among other
things, the identity of the organization funding the campaign, all ads
that are currently running or have run on Twitter, campaign spend, and
targeting demographics for specific ads or campaigns.
Question 11. When can we expect to see them fully implemented?
Answer. We plan to launch the Transparency Center as soon as
feasible after rolling out our electioneering policy in the first
quarter of 2018, and we are continuing to refine the tools we will make
available in conjunction with launching the Transparency Center to
ensure the best experience for our users.
Question 12. How are you defining what constitutes a political ad
subject to these heightened transparency requirements?
Answer. Twitter's definition of electioneering ads will be derived
from the FEC regulations' definition of that term, which includes any
broadcast, cable, or satellite communication that refers clearly to a
candidate for Federal office, is published 60 days before a general
election or 30 days before a primary, convention, or caucus, and is
targeted to the relevant electorate (if the candidate is running for
Congress).
Question 13. On January 29, the Director of the Central
Intelligence Agency said he expects the Russian government to attempt
to influence the 2018 elections in this country.
What efforts is Twitter undertaking in the lead up to the 2018
elections to identify and close the platform's remaining
vulnerabilities to foreign exploitation?
Answer. Our efforts to detect and stop malicious activity on our
platform continue, particularly in the context of elections. Based on
the understanding we have gained from our retrospective review of
activity on our platform during the period leading up to the 2016
election, we have established an internal, cross-functional team
dedicated to addressing election-related instances of abuse on Twitter,
as we discussed with the Committee during the January 17, 2018,
hearing.
The election team will address this challenge in a number of ways.
Among other things, to detect and promptly address impersonation
attempts, the team will verify major party candidates for all statewide
and Federal offices, as well as all major national party accounts. In
addition to monitoring and enforcing the Twitter Terms of Service and
Twitter Rules, the election team will cooperate and communicate with
Federal and state election officials to swiftly escalate and address in
real time attempts at election interference. And consistent with
Twitter's commitment to curtailing malicious automation, spam, and
false accounts on our platform, the election team will focus on
deploying our proprietary tools specifically to detect and stop
malicious election-related activity.
Question 14. What assistance can Federal, state and local
government entities provide in that effort?
Answer. Twitter was pleased to learn that the Federal Bureau of
Investigation (``FBI'') has launched a Task Force to assist companies
and the public in identifying foreign manipulation efforts through
social media platforms. We believe that Federal law enforcement
agencies are uniquely positioned to access, synthesize, and comprehend
disparate sources of intelligence, and to alert the public, Congress,
and social media companies of their findings in a way that provides
broader picture of the activity.
______
Response to Written Question Submitted by Hon. Jerry Moran to
Clint Watts
Question. What is the benefit to social media companies employing
human verification systems? How prevalent are the use of ``social
bots'' by extremists?
Answer. Social bots allow the replication of anonymous accounts to
promote and distribute falsehoods or manipulated truths at such high
volumes that they alter the perceptions of reality and blur the line
between fact and fiction. Human verification systems will prevent the
computer generation of bots by ensuring real people, not computer
programs that replicate accounts, are behind the communications on
their platforms. This will limit the creation of social bots and
increase real human communication on social media platforms. Social
media companies should also want to implement these controls as it
improves the integrity of their systems and ensures the authenticity of
accounts on the platform.
As for extremist use of bots, I am less familiar. The Islamic State
did try to construct its own applications, but I'm not aware of any
terrorist group employing social bots on a large scale. I'd recommend
contacting J.M. Berger, co-author of ISIS: The State of Terror who may
have greater insight into terrorists' use of social bots.
______
Response to Written Questions Submitted by Hon. Tammy Baldwin to
Clint Watts
Question 1. Mr. Watts, what lessons are other authoritarians and
adversarial nations learning from Russia's malign activities on social
media? Do you see others copying the Russian playbook?
Answer. Authoritarian regimes have begun using the Kremlin playbook
largely to suppress internal political dissent. The two most recent and
poignant examples of this phenomenon are the violent oppression and
uprooting of the Rohingya population in Myanmar. Media reports suggest
a large portion of the content on Facebook contains false information
and smears against this minority Muslim population in Myanmar, which
has created a refugee crisis in Bangladesh. The Myanmar government and
many of its security services members now seek to rewrite history by
denying the oppression of the Rohingya or that the group even
exists.\1\
---------------------------------------------------------------------------
\1\ See Hannah Beech, ``No Such Thing as Rohingya': Myanmar Erases
a History.'' New York Times, 2 December 2017 available at: https://
www.nytimes.com/2017/12/02/world/asia/myanmar-
rohingya_denial_history.html.
---------------------------------------------------------------------------
The Philippines leader Rodrigo Duterte also uses Facebook to
suppress internal domestic challenges to his rule and promote his
regime's authoritarian actions.\2\ The most startling development has
been U.S. political campaigns and associated political action groups,
which have adopted similar tactics and in some cases have hired public
relations companies to replicate the same methods employed by the
Kremlin. In short, absent regulation and political leadership,
everyone, in America and abroad, will be using these tactics on their
political opponents.
---------------------------------------------------------------------------
\2\ See Lauren Etter, ``What Happens When the Government Uses
Facebook As A Weapon?'', Bloomberg, 7 December 2017 available at
https://www.bloomberg.com/news/features/2017-12-07/how-rodrigo-duterte-
turned-facebook-into-a-weapon-wtih-a-little-health-from-facebook.
Question 2. If so, how does that impact U.S. interests and how
would you assess the current administration's efforts to respond?
Answer. The U.S. in nearly all theaters and countries around the
world has seen its public image under attack and tarnished. As an
asymmetric approach, nefarious social media influence against the U.S.
is the most effective and least expensive method for undermining
America's power around the world, affecting our military, economic and
diplomatic strength and harming American companies. With regards to the
current administration's response, I've briefed nearly every element of
the U.S. Federal Government on the threat of Russian influence
operations in social media, and I've not been able to glean what their
strategy might be. I'm uncertain why there has not been concerted
action to counter the Kremlin despite so much of U.S. public discussion
and debate on the topic. I cannot imagine how an administration could
do less to respond on such a critical national security issue.
[all]
This page intentionally left blank.
This page intentionally left blank.
This page intentionally left blank.