[House Hearing, 119 Congress]
[From the U.S. Government Publishing Office]
THE FEDERAL GOVERNMENT
IN THE AGE OF ARTIFICIAL INTELLIGENCE
=======================================================================
HEARING
BEFORE THE
COMMITTEE ON OVERSIGHT
AND GOVERNMENT REFORM
U.S. HOUSE OF REPRESENTATIVES
ONE HUNDRED NINETEENTH CONGRESS
FIRST SESSION
__________
JUNE 5, 2025
__________
Serial No. 119-31
__________
Printed for the use of the Committee on Oversight and Government Reform
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
Available on: govinfo.gov,
oversight.house.gov or
docs.house.gov
__________
U.S. GOVERNMENT PUBLISHING OFFICE
60-818 PDF WASHINGTON : 2025
-----------------------------------------------------------------------------------
COMMITTEE ON OVERSIGHT AND GOVERNMENT REFORM
JAMES COMER, Kentucky, Chairman
Jim Jordan, Ohio Vacant, Ranking Minority Member
Mike Turner, Ohio Eleanor Holmes Norton, District of
Paul Gosar, Arizona Columbia
Virginia Foxx, North Carolina Stephen F. Lynch, Massachusetts
Glenn Grothman, Wisconsin Raja Krishnamoorthi, Illinois
Michael Cloud, Texas Ro Khanna, California
Gary Palmer, Alabama Kweisi Mfume, Maryland
Clay Higgins, Louisiana Shontel Brown, Ohio
Pete Sessions, Texas Melanie Stansbury, New Mexico
Andy Biggs, Arizona Robert Garcia, California
Nancy Mace, South Carolina Maxwell Frost, Florida
Pat Fallon, Texas Summer Lee, Pennsylvania
Byron Donalds, Florida Greg Casar, Texas
Scott Perry, Pennsylvania Jasmine Crockett, Texas
William Timmons, South Carolina Emily Randall, Washington
Tim Burchett, Tennessee Suhas Subramanyam, Virginia
Marjorie Taylor Greene, Georgia Yassamin Ansari, Arizona
Lauren Boebert, Colorado Wesley Bell, Missouri
Anna Paulina Luna, Florida Lateefah Simon, California
Nick Langworthy, New York Dave Min, California
Eric Burlison, Missouri Ayanna Pressley, Massachusetts
Eli Crane, Arizona Rashida Tlaib, Michigan
Brian Jack, Georgia
John McGuire, Virginia
Brandon Gill, Texas
------
Mark Marin, Staff Director
James Rust, Chief Counsel for Oversight
Mitch Benzine, General Counsel
Duncan Wright, Senior Professional Staff Member
Josh Mathis, Senior Advisor
Lisa Mortier, Senior Advisor
Frederick Hill, Senior Advisor
Mallory Cogar, Deputy Director of Operations and Chief Clerk
Contact Number: 202-225-5074
Jamie Smith, Minority Staff Director
Contact Number: 202-225-5051
------
C O N T E N T S
----------
Page
Hearing held on June 5, 2025..................................... 1
WITNESSES
----------
Mr. Yll Bajraktari, President, Special Competitive Studies
Project
Oral Statement............................................... 12
Mr. Bhavin Shah, Founder and Chief Executive Officer, Moveworks
Oral Statement............................................... 13
Ms. Linda Miller, Founder and Chief Growth Officer, TrackLight
Oral Statement............................................... 15
Mr. Adam Thierer, Senior Research Fellow, R Street Institute
Oral Statement............................................... 17
Mr. Bruce Schneier (Minority Witness), Fellow and Lecturer,
Harvard Kennedy School
Oral Statement............................................... 19
Written opening statements and bios are available on the U.S.
House of Representatives Document Repository at:
docs.house.gov.
INDEX OF DOCUMENTS
----------
* Statement for the Record, Cyera; submitted by Rep. Mace.
* Statement for the Record, Flashpoint; submitted by Rep. Mace.
* Article, Arizona Technology Council, ``Here's How AI is
Flourishing in Arizona Tech Sector''; submitted by Rep. Biggs.
* Legislation, AZ HB2678; submitted by Rep. Biggs.
* Article, ABC News, ``Trump wants to send U.S. citizens to
foreign prisons, legal experts say''; submitted by Rep.
Crockett.
* Attendance Sheet, Full Committee Attendance Sheet for Hearing
Suspended; submitted by Rep. Crockett.
* Image, COGR Attendance at Musk Subpoena Vote Suspension;
submitted by Rep. Crockett.
* Statement for the Record, Chairman Comer; submitted by Rep.
Mace.
* Report, ``Frontier Models are Capable of In-Context Scheming
- Apollo Research''; submitted by Rep. Perry.
* Letter, re: Federal AI Moratorium - AITIC; submitted by Rep.
Pressley.
* Report, ``Black Tech Agenda Advancing Equity and Reimagining
Technology - Color of Change''; submitted by Rep. Pressley.
* E&C Vote Sheet, CRPT-119-IF00-Vote014-20250513; submitted by
Rep. Trahan.
The documents listed are available at: docs.house.gov.
ADDITIONAL DOCUMENTS
----------
* Questions for the Record: to Mr. Bajraktari; submitted by
Rep. Gosar.
* Questions for the Record: to Mr. Bajraktari; submitted by
Rep. Longworthy.
* Questions for the Record: to Ms. Miller; submitted by Rep.
Gosar.
* Questions for the Record: to Mr. Schneier; submitted by Rep.
Ansari.
* Questions for the Record: to Mr. Shah; submitted by Rep.
Gosar.
* Questions for the Record: to Mr. Thierer; submitted by Rep.
Gosar.
* Questions for the Record: to Mr. Thierer; submitted by Rep.
Langworthy.
These documents were submitted after the hearing, and may be
available upon request.
THE FEDERAL GOVERNMENT
IN THE AGE OF ARTIFICIAL INTELLIGENCE
----------
Thursday, June 5, 2025
U.S. House of Representatives
Committee on Oversight and Government Reform
Washington, D.C.
The Committee met, pursuant to notice, at 10:03 a.m., in
HVC-210, U.S. Capitol Visitor Center, Hon. Nancy Mace
presiding.
Present: Representatives Mace, Jordan, Turner, Gosar, Foxx,
Grothman, Cloud, Palmer, Higgins, Sessions, Biggs, Fallon,
Donalds, Perry, Timmons, Burchett, Greene, Boebert, Luna,
Burlison, Crane, McGuire, Gill, Norton, Lynch, Krishnamoorthi,
Khanna, Mfume, Brown, Stansbury, Garcia, Frost, Lee, Casar,
Crockett, Randall, Subramanyam, Ansari, Bell, Simon, Min, and
Pressley.
Also present: Representatives Trahan and Moskowitz.
Ms. Mace. Good morning. The Committee on Oversight and
Government Reform will now come to order, and we welcome
everyone this morning.
Without objection, the Chair may declare a recess at any
time.
I recognize myself for the purpose of making an opening
statement.
Good morning and thank you for joining us for this
discussion on the Federal Government and artificial
intelligence today. AI is no longer some futuristic idea. It is
here and it is already reshaping everything from health care
and national defense to finance and fraud prevention. The
Federal Government has a responsibility to harness this
technology to make government work faster and more efficiently
for the American people. Today, we will highlight the critical
uses of AI in the Federal Government, which create
efficiencies, improve services, and save taxpayers a whole lot
of money.
The Department of Defense is using AI to improve decision-
making and protect our men and women in uniform. Federal
agencies are using AI to detect fraud before it happens by
using the technology to identify patterns of fraudulent
behavior and working proactively to prevent improper payments.
AI agents are beginning to help the administrative tasks,
freeing up Federal employees to focus on their mission instead
of paperwork. However, barriers remain and challenges must be
addressed so the government is to fully realize the benefits of
this transformative technology. For example, many outdated
legacy IT systems do not integrate with new technologies. A
cumbersome procurement process limits agency access to some of
the best new technologies, including those utilizing AI, and
poor data management leads to information silos and duplicative
work, which cost taxpayers money and prevents agencies from
realizing the full potential of AI.
This Committee has continued to highlight these challenges
and will play a major role in addressing them, including by
continuing bipartisan legislative efforts from last Congress.
This includes the bipartisan Federal AI Governance and
Transparency Act led by Chairman Comer and then Ranking Member
Raskin, which laid out a vision for transparent and trustworthy
AI use in government. It also includes the AI Training
Extension Act, which I sponsored last Congress and reintroduced
today. This bill will equip Federal employees with the AI
literacy and skills necessary to leverage AI across the Federal
Government.
I ask unanimous consent to enter into the record a
statement from the Chairman Comer highlighting the pivotal role
this Committee has to ensure agencies deploy AI responsibly and
the bipartisan efforts of the Committee on these issues.
Without objection.
President Trump has rightly identified the importance of
U.S. dominance in AI and has taken action to remove unnecessary
barriers and unleash innovation. This is a refreshing reversal
from the heavy-handed regulation first approach to AI taken by
the Biden Administration. Last Congress, my Subcommittee on
Cybersecurity, Information Technology, and Government
Innovation held multiple hearings on the disastrous impacts of
overreaching policies, and I was relieved to see the former
President Biden's AI executive order rescinded on day one. This
was a critical first step, and since then, the White House has
directed agencies to prioritize the deployment of AI and
improve procurement processes to eliminate barriers to the
Federal Government's use of AI. I believe it is important for
this Committee to watch these developments closely and work to
ensure the executive branch has the tools and authorities to
deploy AI at scale and realize the full potential of this
incredible technology.
I want to thank each of our witnesses this morning for
sharing their time with us today and look forward to hearing
from each of you. I also want to thank my colleagues from the
other side of the aisle who showed up in great number this
morning, too. I appreciate your interest in this topic. This
has been a bipartisan topic for this Committee. And with that,
I would like to recognize Ranking Member Lynch for his opening
statement.
Mr. Lynch. Thank you very much. Good morning, Madam Chair,
and again, thank you for calling this hearing on the use of
artificial intelligence in our government.
Optimizing the Federal Government's use of technology has
long been a bipartisan priority of this Committee. Indeed, as
you know, Madam Chair, our late colleague and friend, Gerry
Connolly, was a steadfast champion for using modern technology
to improve the delivery of government services, strengthen
oversight, and enhance public trust. We continue to fight his
fight on this and so many other critical challenges facing our
government. We cannot sit here, however, and have the
traditional bipartisan conversation about Federal IT
modernization without acknowledging the fact that the Trump
Administration, Elon Musk, and DOGE are leading technology
initiatives that threaten the privacy and security of all
Americans and undermine our government and the vital services
it provides to red states and blue states. Musk may say he has
stepped away from his role in the Federal Government, but his
recklessness will continue to have devastating consequences for
America for years, possibly decades to come.
Today, we will work to better understand the disaster and
danger the Trump Administration has created by turning our
government over to his biggest campaign donor and people who
are more interested in self-enrichment than public service.
Since President Trump's inauguration, he has given Elon Musk
free rein to terrorize our civil servants and drive more than
275,000 Federal employees from their jobs serving the American
people. His DOGE team has done this so capriciously that they
have had to go back to people that they originally fired and
beg them to return to their jobs, serving our veterans,
maintaining our nuclear stockpile, and fighting bird flu. At
the Social Security Administration, whistleblowers have told us
that the Agency has lost so many critical information
technology and support staff, that very few of the remaining
workers even know how to handle errors or alerts in their
computer systems. This could mean missed checks and millions of
seniors and people with disabilities losing access to their
financial lifeline. While they purged the Federal Government of
expertise, Elon Musk and DOGE reportedly fed sensitive
government data, including Americans' private personal
information, into unvetted and unaccountable AI software. They
also reportedly deployed a Musk-owned generative AI chatbot
called Grok to consume government data and assist with
sensitive government decisions. Grok has even been deployed
onto systems at the Department of Homeland Security, despite
the fact that it has not been approved for use.
Musk has been operating without any oversight whatsoever,
while posing a very real risk of violating security and privacy
laws, accessing confidential information about his competitors,
and using valuable Federal data bases to turbocharge his own AI
system, which would mean more money and more power for Mr.
Musk. The American people deserve answers. Musk and DOGE have
put the most personal data of every American at risk.
Cybersecurity experts like Mr. Schneier, who is here with us
today, are sounding the alarm that the holes DOGE has created
in our national cybersecurity system are gaping and pervasive.
And now, in this horrifying, insecure environment, the Trump
Administration wants to centralize all of Americans' data
into one massive data base, which would allow full-spectrum
surveillance of American citizens, making it even easier for
bad actors to hack, and, of course, they have contracted with a
big, private tech company owned by one of President Trump's
billionaire supporters to do just that.
Four months ago, Democrats moved to subpoena Elon Musk to
provide public testimony to this Committee. At that time, our
Republican colleagues complained that their constituents were
calling them, their offices, and showing up to town halls to
yell at them about Musk. Well, Americans' concerns have only
deepened since then. From his erratic purge of the Federal
workforce to his exploitation of sensitive taxpayer data, to
the cybersecurity nightmare he has created, and the horrifying
surveillance state we fear, Musk has put the American people
into a position where they demand answers from Elon Musk. I
believe, too, that my Republican colleagues should have had
some questions of their own for Mr. Musk.
Recent allegations have left Americans wondering if he was
potentially under the influence of drugs while at the very
heart of the Trump Administration's most consequential and
sensitive decisions. It is worth investigating whether one of
President Trump's most influential advisors was under the
influence of heavy drugs while upending hundreds of thousands
of American lives, breaking government services beyond repair,
and handing out death sentences for hundreds of thousands of
innocent people around the world who relied on American foreign
aid and medical care for survival, most of whom were children.
If this does not move my Republican colleagues, maybe they
just want to hear from Musk on why he called their bill to cut
Medicaid and give tax breaks to billionaires--this is Elon
Musk's words--``a disgusting abomination,'' and ``massive,
outrageous, pork filled congressional spending bill,'' or why
he spent the last 24 hours leveraging his social media to
``Kill the bill.'' Or maybe they want to know why Musk has said
that his next round of firings might be targeted at them,
Republican politicians who have not toed the line on Musk's
priorities.
Regardless of your motivations, Democrats and Republicans
alike simply cannot allow Elon Musk to escape accountability
for his actions. He has dismantled our government, endangered
Americans, and weaponized public service for his own financial
gain. I ask my Republican colleagues, let us work together on
this. Let us support oversight that Democrats today in moving
that this Committee subpoena Elon Musk to answer for his
actions. All of us deserve answers on behalf of the American
people.
So, Madam Chair, therefore, pursuant to Clause 2(k)(6) of
House Rule XI, I move that we subpoena, this Committee,
Republican and Democrat, subpoena Elon Musk to testify before
this Committee.
Mr. Mfume. I second the motion.
Ms. Mace. We will suspend for a moment. Thank you.
Mr. Casar. Chairwoman, point of order. When are we going to
vote on subpoenaing Musk?
Ms. Mace. We are doing a point of order right now with the
parliamentarian.
Mr. Higgins. Madam Chair?
Ms. Mace. Thank you. We are suspended.
Mr. Krishnamoorthi. Madam Chair, may I be recognized?
Ms. Mace. Yes, sir.
Mr. Higgins. Madam Chair, I move to table the gentleman's
motion.
Ms. Mace. OK. The motion is not debatable.
As many as are in favor of tabling, signify by saying aye.
[Chorus of ayes.]
Ms. Mace. All those opposed, signify by saying no.
[Chorus of noes.]
Mr. Lynch. I ask for a recorded vote.
Ms. Mace. In the opinion of the Chair, the ayes have it,
and the motion to table is agreed to.
The Committee will now resume consideration----
Mr. Lynch. I ask for a recorded vote.
Mr. Casar. Interesting opinion.
Mr. Krishnamoorthi. Madam Chair?
Mr. Mfume. A recorded vote is requested.
Mr. Krishnamoorthi. Parliamentary inquiry.
Ms. Mace. OK. A recorded vote is ordered, and the clerk
will call the roll.
OK. We are going to suspend to allow the clerk to get
ready.
Mr. Krishnamoorthi. Parliamentary inquiry.
Ms. Mace. Yes, sir.
Mr. Krishnamoorthi. Madam Chair, thank you. Is it the
custom of the Committee to shut down debate on such an
important motion?
Ms. Mace. This is not an inquiry. This is not an inquiry.
We are suspended.
Mr. Krishnamoorthi. The gentleman from Massachusetts
mentioned that Mr. Musk was potentially under the influence of
ketamine----
Ms. Mace. This is not debatable right now. I love you. It
is not debatable right now, and this is not a point of inquiry.
Mr. Krishnamoorthi [continuing]. And other----
Ms. Mace. But you look good in front of the photos. You
look great right now. You look good.
Mr. Krishnamoorthi. So, is this the----
Ms. Mace. You are not recognized right now. There is not a
point of inquiry. Thank you. We are suspended until the clerk
is ready to call the roll.
Mr. Frost. Move to call the roll.
Ms. Mace. We are suspended again. For the third time, we
are suspended until the clerk can call the roll. Thank you.
[Pause.]
Ms. Crockett. Madam Chair, point of inquiry.
Ms. Mace. OK. Yes, ma'am.
Ms. Crockett. Is the clerk here?
Ms. Mace. The clerk is here and preparing to call the roll.
Ms. Crockett. OK.
Ms. Mace. Thank you.
Ms. Crockett. Can you explain exactly what preparing
requires?
Ms. Mace. She has to get paperwork. She has to get seated.
She has to do all that. Thank you. She sure does, yes.
Ms. Stansbury. Or you guys have to find your Members who
did not attend the hearing. That is probably more likely.
Mr. Krishnamoorthi. Madam Chair, inquiry, please.
Ms. Mace. Yes, sir. Do you want more video with Elon behind
you? You already got a good clip.
Mr. Krishnamoorthi. Could we have a unanimous consent
motion to structure debate for 10 minutes, each side gets 5
minutes to debate----
Ms. Mace. We are getting ready to do a recorded vote.
Mr. Krishnamoorthi [continuing]. Whether he was under the
influence of drugs?
Ms. Mace. We cannot do that. We are not able to do that
right now. Let us do the recorded vote, then we can talk about
it after. Thank you.
Ms. Crockett. Parliamentary inquiry.
Ms. Mace. Yes, ma'am.
Ms. Crockett. Do we have an assistant clerk that can move
maybe a little bit faster? We have a room full of people.
Ms. Mace. This is not an inquiry. Thank you.
Ms. Crockett. It was a question. Do we have an assistant
clerk?
Ms. Mace. It is not a parliamentarian inquiry, so, no, we
are not having this debate.
Mr. Mfume. But it is a point of order.
Ms. Mace. It is not. A motion to table is not debatable.
Thank you.
Ms. Crockett. We are not debating that. We are asking about
the clerk.
Mr. Mfume. We were not asking for a point of order. Will
the parliamentarian rule on the point of order?
Mr. Frost. Is the parliamentarian here?
Ms. Mace. The gentleman must state his point of order.
Mr. Mfume. It was the gentlelady's point of order that was
overruled because it was determined that it was not a point of
order when in fact it is. I think that she should be allowed
to----
Ms. Mace. She stated it was an inquiry, not a point of
order, from my recollection, but if she would like to restate
it, that is fine.
Mr. Mfume. Oh.
Ms. Crockett. Madam Chair, point of order.
Ms. Mace. Yes, ma'am.
Ms. Crockett. Do we not have an assistant clerk who can
make sure that we can continue to do the business in an
efficient way in Oversight because it has historically never
taken this long for the clerk to call the roll.
Ms. Mace. We are going through the process right now, and
we will have a recorded vote as soon as we are able to.
Mr. Lynch. Just to restate the gentlelady's point,
Democrats are here. We think this is an important issue. We are
ready to vote.
Ms. Mace. The gentleman must state a point of order.
Mr. Lynch. And we welcome the opportunity to debate this.
This is a serious issue.
Ms. Mace. Is there a point of order, Mr. Lynch?
Mr. Krishnamoorthi. The gentleman has the Floor. Let him
finish.
Ms. Mace. I am the Chairwoman. Thank you. Women are allowed
to speak on this Committee, and I ask for a point of order.
Ms. Crockett. Point of inquiry. Madam Chair, point of
inquiry. I should, hopefully, get to speak?
Ms. Mace. Yes, ma'am. Yes, you are. Yes, you are babe.
Ms. Crockett. Is it possible for us to just do a roll call
without the clerk to determine what Members are here so that we
can determine if we all really have the votes?
Ms. Mace. No, ma'am. Thank you.
Ms. Stansbury. For those of you watching at home, the
Republicans are not in the room, so they are trying to find
their Members so that they can----
Ms. Mace. You are not recognized, Ms. Stansbury. Thank you.
You can be quiet now.
Ms. Stansbury. Neither are any of your Members because they
are not in the room. Thank you.
Mr. Lynch. Madam Chair, on a point of order, Democrats have
a clerk that is available to record a vote.
Ms. Mace. OK. Mr. Lynch, the Chair will not entertain
points of order regarding the propriety or expediency of a
proposed course of action. This is my final ruling on such
points of order. Thank you.
Mr. Lynch. Madam Chair, a point of inquiry then.
Ms. Mace. Yes, sir.
Mr. Lynch. Would it not be permissible to have the
Democratic clerk record the vote and to begin it now?
Ms. Mace. No, sir, it is not. Thank you.
Ms. Crockett. Madam Chair?
Ms. Mace. Yes, ma'am.
Ms. Crockett. Is the only reason we cannot take a vote is
because you know you are about to lose it?
Ms. Mace. Is there a point of inquiry here, because I did
not hear one.
Ms. Crockett. That was my inquiry. Is the reason we----
Ms. Mace. That is not a valid inquiry. Thank you.
Ms. Crockett. OK.
Mr. Min. I have a point of inquiry, Madam Chair.
Ms. Mace. Yes, sir.
Mr. Min. I was just wondering, being new to this Committee,
what is the basis or rule on which you are not allowing the
Democratic clerk to start the vote?
Ms. Mace. We have our own clerk. That is why.
Mr. Min. Is that based on the rule or?
Mr. Crane. It is based on the fact that you guys lost the
election because of your crazy policies. And you guys do not
control the chamber.
Mr. Min. Is that in the rules? Madam Chair, I just wanted
to follow-up, would appreciate an answer to the question. We
have witnesses who have flown in from around the country.
Ms. Mace. This is Committee practice. Thank you for your
inquiry, and we will start the vote momentarily. Thank you.
Mr. Lynch. Madam Chair, just a point of inquiry.
Ms. Mace. Yes, sir.
Mr. Lynch. It has now been 10 minutes, and this motion is
simply to bring Mr. Musk forward so that we can ask him
questions on behalf of the American people. I think the
Democratic position has been very reasonable, very deliberate.
You have got a lot----
Ms. Mace. The motion is to table, and there is no debate.
Thank you, Mr. Lynch.
Mr. Lynch. Again, Madam Chair, Democrats are all here, and
we are ready to vote. It is now 11 minutes.
Ms. Mace. You are not recognized, but you do look good for
your clips later, for social media, with Elon and all the words
behind you. It looks good.
Mr. Lynch. Appreciate it.
Ms. Mace. The blue is good with the red color.
Mr. Lynch. I appreciate that. Thank you. I really
appreciate that. Thank you.
Mr. Fallon. Madam Chair, parliamentary inquiry. In a
minute, will it be another minute?
Mr. Burchett. Probably 12, right?
Ms. Mace. Could be.
Mr. Fallon. It is debatable, depending on the math, I
suppose.
Ms. Crockett. Madam Chair. Madam Chair, I have a unanimous
consent.
Ms. Mace. We are in the middle of getting ready to vote, so
we cannot do that right now. We can do that after.
Ms. Crockett. Is that the rule because the clerk has not
been ready for the last 15 minutes, almost.
Ms. Mace. You are not recognized. Thank you.
Ms. Crockett. So, we are not doing unanimous consent.
Ms. Mace. Not in the middle of a motion to table and a
recorded vote we are preparing to take.
Ms. Crockett. Is the clerk seated?
Ms. Mace. You are not recognized right now. Thank you.
Ms. Crockett. OK. Since we are sitting here in silence for
the----
Ms. Mace. You are not recognized, and there is no point of
inquiry. You are not recognized.
Ms. Crockett. OK. I will do a point of inquiry.
Ms. Mace. OK.
Ms. Crockett. I am seeking an inquiry to determine whether
or not it is OK to put it into the record that at the time that
this motion was called, only Mr. Gosar, Mr. Grothman, Mr.
Higgins, Mr. Fallon, and Mr. Crane, along with yourself, were
the only Republicans that were actually in the room.
Ms. Mace. This is not a real point of inquiry. You are not
recognized on it either. Thank you.
Ms. Pressley. Madam Chair, I have a legitimate point of
inquiry. Madam Chair.
Ms. Mace. Is it for real, this time?
Ms. Pressley. Oh, no, it is very legitimate, very for real.
How much taxpayer money are Republicans wasting while delaying
this vote to subpoena Elon Musk?
Ms. Mace. That is not a legitimate point of inquiry.
Ms. Pressley. It is very legitimate to the American people.
Ms. Mace. It is not. Thank you. It is not. You are not
recognized.
Ms. Pressley. Very legitimate to the American people.
Ms. Mace. It is not, and you are not recognized on it.
Thank you.
Mr. Fallon. Madam Chair, parliamentary inquiry.
Ms. Mace. Yes, sir.
Mr. Fallon. How much taxpayer money was wasted when some of
the Democrats did not show up to work for almost 2 years during
COVID?
Ms. Mace. Also, not a point of inquiry. I love you, Mr.
Fallon.
Mr. Fallon. Particularly one from Boston.
Ms. Mace. Also, not a point of inquiry. Thank you.
Ms. Pressley. Who are you talking about? I am sorry, are
you engaging in personalities right now? Who are you talking
about? Who are you talking about?
Mr. Fallon. I did not say anybody's name.
Ms. Pressley. Who are you talking about?
Ms. Mace. This is not time for debate right now. Thank you.
Ms. Pressley. Well, please.
Ms. Mace. I am being respectful of everybody on both sides
of the aisle. Let us be respectful of one another, please.
Mr. Fallon. I am from Pittsville, and I was here.
Ms. Mace. This is not debatable. We are not having a debate
on this. We are getting ready to take a recorded vote.
Everybody can sit down and turn their microphones off unless
you have a point of inquiry, an appropriate point of inquiry.
Thank you.
Ms. Crockett. Madam Chair, point of inquiry.
Ms. Mace. Yes, ma'am.
Ms. Crockett. You said that we were about to take the vote.
Oh, the clerk.
Ms. Mace. Yes, we are about to take the vote.
Ms. Crockett. Never mind.
Ms. Mace. Thank you.
Ms. Crockett. Because I was about to say, your Members
finally showed up. I mean, let us go.
Ms. Mace. Thank you, ma'am. All right. A recorded vote is
ordered and the clerk will call the roll.
The Clerk. Mr. Jordan?
Mr. Jordan. Yes.
The Clerk. Mr. Jordan votes yes.
Mr. Turner?
[No response.]
The Clerk. Mr. Gosar?
[No response.]
The Clerk. Ms. Foxx?
Ms. Foxx. Yes.
The Clerk. Ms. Foxx votes yes.
Mr. Grothman?
Mr. Grothman. Yes.
The Clerk. Mr. Grothman votes yes.
Mr. Cloud?
Mr. Cloud. Yes.
The Clerk. Mr. Cloud votes yes.
Mr. Palmer?
Mr. Palmer. Yes.
The Clerk. Mr. Palmer votes yes.
Mr. Higgins?
Mr. Higgins. Yes.
The Clerk. Mr. Higgins votes yes.
Mr. Sessions?
Mr. Sessions. Aye.
The Clerk. Mr. Sessions votes aye.
Mr. Biggs?
[No response.]
The Clerk. Mr. Fallon?
Mr. Fallon. Aye.
The Clerk. Mr. Fallon votes aye.
Mr. Donalds?
Mr. Donalds. Yes.
The Clerk. Mr. Donalds votes yes.
Mr. Perry?
Mr. Perry. Aye.
The Clerk. Mr. Perry votes aye.
Mr. Timmons?
Mr. Timmons. Yes.
The Clerk. Mr. Timmons votes yes.
Mr. Burchett?
Mr. Burchett. Yes.
The Clerk. Mr. Burchett votes yes.
Ms. Greene?
Ms. Greene. Yes.
The Clerk. Ms. Greene votes yes.
Ms. Boebert?
[No response.]
The Clerk. Mrs. Luna?
Mrs. Luna. Aye.
The Clerk. Mrs. Luna votes aye.
Mr. Langworthy?
[No response.]
The Clerk. Mr. Burlison?
Mr. Burlison. Yes.
The Clerk. Mr. Burlison votes yes.
Mr. Crane?
Mr. Crane. Yes.
The Clerk. Mr. Crane votes yes.
Mr. Jack?
[No response.]
The Clerk. Mr. McGuire?
Mr. McGuire. Yes.
The Clerk. Mr. McGuire votes yes.
Mr. Gill?
Mr. Gill. Yes.
The Clerk. Mr. Gill votes yes.
Mr. Lynch?
Mr. Lynch. No.
The Clerk. Mr. Lynch votes no.
Ms. Norton?
Ms. Norton. No.
The Clerk. Ms. Norton votes no.
Mr. Krishnamoorthi?
Mr. Krishnamoorthi. No.
The Clerk. Mr. Krishnamoorthi votes no.
Mr. Khanna?
Mr. Khanna. No.
The Clerk. Mr. Khanna votes no.
Mr. Mfume?
Mr. Mfume. No.
The Clerk. Mr. Mfume votes no.
Ms. Brown?
Ms. Brown. No.
The Clerk. Ms. Brown votes no.
Ms. Stansbury?
Ms. Stansbury. No.
The Clerk. Ms. Stansbury votes no.
Mr. Garcia?
Mr. Garcia. No.
The Clerk. Mr. Garcia votes no.
Mr. Frost?
Mr. Frost. No.
The Clerk. Mr. Frost votes no.
Ms. Lee?
Ms. Lee. No.
The Clerk. Ms. Lee votes no.
Mr. Casar?
Mr. Casar. No.
The Clerk. Mr. Casar votes no.
Ms. Crockett?
Ms. Crockett. No.
The Clerk. Ms. Crockett votes no.
Ms. Randall?
Ms. Randall. No.
The Clerk. Ms. Randall votes no.
Mr. Subramanyam?
Mr. Subramanyam. No.
The Clerk. Mr. Subramanyam votes no.
Ms. Ansari?
Ms. Ansari. No.
The Clerk. Ms. Ansari votes no.
Mr. Bell?
Mr. Bell. No.
The Clerk. Mr. Bell votes no.
Ms. Simon?
Ms. Simon. No.
The Clerk. Ms. Simon votes no.
Mr. Min?
Mr. Min. No.
The Clerk. Mr. Min votes no.
Ms. Pressley?
Ms. Pressley. No. No.
The Clerk. Ms. Pressley votes no.
Ms. Tlaib?
[No response.]
The Clerk. Madam Chairwoman?
Ms. Mace. Aye.
The Clerk. Madam Chairwoman votes aye.
Ms. Mace. And how is Mr. Biggs recorded?
The Clerk. Mr. Biggs is not recorded.
Mr. Biggs. Aye.
The Clerk. Mr. Biggs votes aye.
Ms. Mace. And how is Ms. Boebert recorded?
The Clerk. Ms. Boebert is not recorded.
Ms. Boebert. Aye.
The Clerk. Ms. Boebert votes aye.
Mr. Frost. Madam Chair, parliamentary inquiry.
Ms. Mace. The clerk will report the tally.
The Clerk. Madam Chairwoman, on this vote, the ayes are 21.
The nays are 19.
Ms. Mace. The ayes have it, and the motion to table is
agreed to.
The Committee will now resume.
OK. I now request unanimous consent that Representative
Moskowitz of Florida and Trahan of Massachusetts will be waived
on to today's hearing for the purpose of asking questions, and
without objection, so ordered.
I am pleased to introduce our witnesses for today's
hearing. Our first witness today is Mr. Yll Bajraktari, the
President of the Special Competitive Studies Project. Our
second witness is Mr. Bhavin Shah, founder and chief executive
of the AI company, Moveworks. Our third witness is Ms. Linda
Miller, former Deputy Executive Director of the Pandemic
Response Accountability Committee and founder of TrackLight.
Our fourth witness is Mr. Adam Thierer, Senior Research Fellow
at R Street Institute, and our fifth witness today is Mr. Bruce
Schneier from the Harvard Kennedy School. We welcome everyone,
and we are pleased to have you here this morning.
So, pursuant to Committee Rule 9(g), the witnesses will
please stand and raise your right hands.
Do you solemnly swear or affirm the testimony you are about
to give is the truth, the whole truth, and nothing but the
truth, so help you God?
[A chorus of ayes.]
Ms. Mace. Let the record show the witnesses all answered in
the affirmative.
We appreciate all of you being here today and look forward
to your testimony. Let me remind the witnesses that we have
read your written statements, and they will appear in full on
the hearing record. Please limit your oral statements to 5
minutes. As a reminder, please press the button on the
microphone in front of you so that it is turned on and the
Members up here can hear you. When you begin to speak, the
light in front of you will turn green. After 4 minutes, the
light will turn yellow. When the red light comes on, your 5
minutes has expired, and we would ask that you please wrap it
up.
So, I now recognize Mr. Bajraktari to please begin your
opening statement.
STATEMENT OF YLL BAJRAKTARI
PRESIDENT
SPECIAL COMPETITIVE STUDIES PROJECT
Mr. Bajraktari. Chairwoman Mace, Ranking Member Lynch, and
Members of the Committee, thank you for the opportunity to
testify here today. I am here to make one clear point. While
developing advanced AI is a significant moment, the true
measure of a Nation's future advantage will not be who has the
best models, but who has built the essential digital
infrastructure and fosters the deepest adoption of AI across
all of the society. To understand the stakes, imagine if at the
beginning of the last century, our Nation had failed to embrace
electricity and yielded that transformative advantage to our
competitors. The consequences would have been devastating.
Today, we face a similar moment. The strategic competition
in AI with People's Republic of China is intensifying. Since
2017, Beijing has launched an ambitious campaign to achieve
technological supremacy as the key to its global dominance.
China is executing this strategy by dedicating vast resources
to integrate AI across its entire economy, society, and
military. For example, they recently announced $47.5 billion
semiconductor fund, which is designed to eliminate reliance on
Western technology. They are on track to surpass the U.S. in
raw R&D spending, having already tripled their investments
since Xi Jinping came to power in 2012 to $500 billion in 2024.
This is all built on dominant infrastructure, including
more than 60 percent of the world's 5G base stations. This
entire effort is backed by state-sponsored cyber espionage and
intellectual property theft designed to acquire and weaponize
advanced technology. While the United States still leads AI in
research, our primary disadvantage is the slow adoption of
these breakthroughs. We are hampered by bureaucratic inertia,
outdated IT infrastructure, and a lack of AI literacy in the
workforce. Overcoming these barriers is the key to winning the
global tech competition. To secure our leadership, United
States must take decisive action in the following areas.
First, we need to establish a high-level technology
competitiveness council at the White House modeled on the
approach we took for the space race to coordinate a unified
national strategy and drive rapid AI adoption across the
government. Second, we need to significantly increase our
investment in AI infrastructure by doubling non-defense AI R&D
spending to $32 billion over time and codifying the national AI
research and resource. Third, we need to launch a comprehensive
national talent strategy to make the U.S. the magnet for AI
experts. This means eliminating green card caps for the top
STEM talent and promoting AI literacy throughout our
educational system. Fourth, we must overhaul the Federal
procurement process to rapidly integrate cutting-edge AI from
our most innovative companies. Last, we must strengthen our
global alliances in AI, 5G, and cybersecurity to ensure the
resilience and security for us and our allies.
The bottom line is this: this is not just a competition of
invention, but of adoption. It is about who can set the rules
for the future. We have faced moments like this before, from
the space race to the digital revolution. Our Nation's greatest
strength has always been our ability to unite our vision with
decisive action. By embracing this challenge together across
government, industry, and academia, we will not only secure our
leadership, but build a more prosperous and secure future for
all Americans. Thank you, and I look forward to your questions.
Ms. Mace. Thank you. I will now recognize Mr. Shah for his
opening statement.
STATEMENT OF BHAVIN SHAH
FOUNDER AND CHIEF EXECUTIVE OFFICER
MOVEWORKS
Mr. Shah. Chairwoman Mace, Ranking Member Lynch, and
distinguished Members of the Committee, thank you for the
opportunity to testify on government use and procurement of AI.
I am Bhavin Shah, CEO and founder of Moveworks. Moveworks is an
enterprise AI company that has spent 9 years successfully
helping some of the world's largest organizations to use AI to
transform their operations. Today, we face a moment of national
importance. The question before us is not whether the Federal
Government should adopt AI. It is whether we will lead or
follow. Before we talk about the procurement hurdles that we
have faced, let me explain to you what Moveworks does.
Moveworks transforms how organizations support their
employees. We do this through an enterprise AI assistant, and
employees interact conversationally with the assistant in a
chat-based interface. They can get IT support, ask H.R.
questions, and find company information. But the assistant is
not just for looking up information. It can also be used to
take actions, such as scheduling time off or obtaining software
licenses. Behind the scenes, our platform integrates with all
of an organization's technology systems.
Because the AI assistant is a single unified tool,
employees no longer have to log in and out of multiple systems
to get work done. Organizations can extend the system to
simplify and accelerate their most complex business processes,
all done safely and securely while respecting permissions. We
do this work by combining top-tier language models, like
OpenAI's GPT-4, Meta Llama 3, as well as other models from
Google's DeepMind, Microsoft, NVIDIA, just to name a few. We
put these together with our own enterprise-specific models, all
governed by enterprise-grade guardrails, including fact-
checking, citations, and security filters. The result is AI
that understands work context. It handles complex business
processes safely and effectively and follows industry-leading
data protection and privacy approaches.
Today, Moveworks serves 5 million employees across 350
organizations, including major companies like Broadcom,
Marriott, and CVS Health. We are also used by a number of local
governments, such as Scottsdale, Ventura, and Durham County. In
addition, government contractors, like Leidos and Northrop
Grumman, use Moveworks to make their businesses more efficient.
We have seen firsthand how AI drives measurable results.
Honeywell reduced IT help desk requests by 80 percent, Broadcom
scaled from 10,000 to 50,000 employees without increasing
service staff, and the city of Glendale achieved a 514 percent
return on its AI investment.
Our experience positions us uniquely to understand the
Federal Government's complex requirements. The Federal
Government has an unprecedent opportunity. You can leverage AI
for efficiency, you can achieve cost savings, you can improve
employee experience, and, most importantly, you can therefore
improve citizen services. We have engaged with agencies across
the government and consistently see enthusiasm for AI adoption,
but that enthusiasm has not translated into government
procuring AI products like Moveworks.
We have encountered three primary barriers. The first is
the FedRAMP process. Moveworks invested 3-and-a-half years and
over $8-and-a-half million to achieve FedRAMP-ready status.
This is a prohibited barrier for smaller AI innovators, the
companies that often develop the most cutting-edge solutions.
Second, complex reseller networks. These force innovative
companies to work through intermediaries rather than directly
with agencies, adding cost without adding value. Third, multi-
year contracting vehicles. These favor large incumbents over
agile AI companies. These create delays measured in years.
Meanwhile, AI technology advances every month.
American AI leadership requires action, not just intention.
The private sector has demonstrated that AI adoption can both
be rapid and secure when procurement processes match the pace
of innovation. I recommend Congress consider immediate steps to
improve government use and procurement of AI. Embed AI
education and adoption goals in agency leadership performance
plans. Create government developer programs for startups to
learn Federal systems. Scale AI proof-of-concept programs for
rapid pilots to all agencies, similar to the Investment
Horizon's pilot program developed by the Office of the CIO
within the Navy. And finally, establish agency innovator
programs that allow smaller agencies to serve as test beds for
AI implementations. Longer term, we must modernize acquisition
policies. We must accelerate procurement timelines. The Federal
Government employs the most important workforce in the world,
the workforce that operates our democracy and defends our
Nation. These public servants deserve the same AI-powered tools
that are transforming the private sector's productivity and
effectiveness.
We have a choice. We can modernize our procurement process
to embrace AI innovation or we can watch other Nations take the
lead in government AI adoption while our agencies struggle with
outdated processes and missed opportunities. I urge this
Committee to champion procurement reform. We need to maintain
our competitive edge while serving citizens more effectively.
Thank you, and I welcome your questions and discussions on
these critical topics.
Ms. Mace. Thank you. I will now recognize Ms. Miller for
her opening statement.
STATEMENT OF LINDA MILLER
FOUNDER AND CHIEF GROWTH OFFICER
TRACKLIGHT
Mr. Miller. Thank you, Chairwoman, Ranking Member, and
Members of the Committee. My name is Linda Miller. I bring a
unique perspective today because I spent 10 years at one of the
world's most admired agencies, the GAO, as well as I am also
the co-founder of an AI technology startup. So, I am coming
here with both the perspective of a long history in the
oversight community, as well as some of the challenges that
have already been raised about trying to bring innovative
technology to the U.S. Government.
As DOGE is seeing firsthand, it is hard, if not impossible,
to try and impose private sector innovation on the Federal
Government. The data challenges alone are formidable. Agencies
collect data poorly if they collect it at all. Programs collect
their own data in silos. They do not share it even amongst
themselves, much less with other agencies. In many cases,
agencies do not even know who they are paying. They send money
to states in the form of grants, and from there, the money
trail is opaque, making payment integrity almost impossible. At
TrackLight, to help illuminate who one Federal agency was
paying, we spoke to a company who FOIAs data on invoices to
state agencies and sells it. For our Federal client, we were
going to have to buy the data to tell them who they were
paying. The same issue exists at the state level. GAO reported
on a grant program that it administered by one state in which a
company applied for the same grant 13 times in neighboring
counties and got it because the states did not share that data.
The size and complexity of the administrative state and the
volume of regulations agencies must comply with make agility
and innovation next to impossible. My fellow panelists have
made some very smart suggestions about turbocharging the rip
and replace efforts of legacy IT systems and removing
procurement barriers, but we have to be realistic about how
capable government is to enact these rapid changes no matter
how badly they are needed. I could spend my time this morning
telling you about how we urgently need to adopt AI across
government to remain relevant, achieve cost savings, and
improve service delivery as well as prevent fraud, waste, and
abuse, but I think you already know that. So, I would like to
talk about what proactively can be done to move the needle.
GAO has reported that there are almost 25 existing AI
requirements in Federal law and guidance today. GSA has
established an AI center of excellence which was required by
law. Agencies have inventoried AI use cases. They have put in
place responsible AI officials, but the government continues to
show anemic AI adoption overall. My fear is we have created yet
another bureaucratic morass, hamstringing agencies with
compliance work and only added the layers of sludge, further
slowing down AI adoption. There has been progress. Recent EOs
have sought to improve data sharing, which will improve how
agencies better make use of data to use AI, and GSA's FedRAMP
overhaul is a step in the right direction. I completely agree
with Mr. Shah. He spent $8.5 million trying to get FedRAMP
certified and successfully getting FedRAMP certified. My small
technology company is going to have to put out millions of
dollars to do the same.
But AI adoption is the perfect candidate for what is called
a regulatory sandbox, which Congress can mandate to speed AI
projects to implementation in controlled settings, suspending
these layers of bureaucracy. Creative procurement solutions
must be explored and carefully monitored, allowing lessons to
be learned and shared across agencies. And outdated privacy
laws, privacy laws that were designed for a world that existed
long before the internet, much less artificial intelligence,
has to be revised. I cannot state this more clearly: the bad
guys have all our data. We cannot protect Americans' privacy in
2025. Today, privacy protection is going to have to involve
advanced technology, like anonymization of sensitive data, and
the use of other types of identifiers like behaviors and
geolocation.
I would be remiss not to mention the risks, which are real.
Lazy, careless use of AI does not just imperil the goals of the
AI project, it undermines the trust in the technology itself.
Hallucinations are real. Subject matter expertise is vital.
Responsible AI is not just a buzzword. Humans in the loop means
humans who are trained, knowledgeable, and paying close
attention. This is not a set it and forget it tool. Now is the
time to be bold. We cannot change how government operates.
Wholesale changes to legacy IT systems and the Federal
acquisition system will take years, but we can create
innovative laboratories where AI projects can operate in proof-
of-concept, regulatory sandboxes in carefully controlled
environments to start to show the art of the possible.
I welcome your questions. Thank you.
Ms. Mace. Thank you. I will now recognize Mr. Thierer for
his opening statement.
STATEMENT OF ADAM THIERER
SENIOR RESEARCH FELLOW
R STREET INSTITUTE
Mr. Thierer. Chairwoman Mace, Ranking Member Lynch, and
Members of the Committee, thank you for your invitation to
participate in this important hearing. My name is Adam Thierer.
I am a senior fellow at the R Street Institute where I cover
emerging technology policy. My message here today focuses on
three key points. First, there are meaningful benefits to
governmental use of artificial intelligence technologies.
Second, Congress and the Trump Administration need to take
steps to unlock those benefits by accelerating the
modernization of government systems and policies. Third, we
must appreciate the connection between broader AI regulation
and the benefits the government itself can accrue from these
systems.
While the Federal Government is already integrating AI into
systems and processes, progress has been slow, and there needs
to be more urgency to tap into the many benefits that AI can
offer. The bipartisan House Task Force on Artificial
Intelligence report from last December noted that, ``Each year,
the Federal Government spends over $100 billion on information
technology and cybersecurity. Approximately 80 percent of this
spending goes to operating existing legacy systems that are
typically outdated and underpinned by archaic software and
hardware components.'' AI can help address this persistent
problem.
Last year, the U.S. Chamber of Commerce released a study on
how improved IT processes in eight already helped to
``modernize infrastructure, enhance citizen services, improve
national security, and foster innovation, and unlock
significant long-term cost savings and efficiencies along the
way.'' Doing more with AI could generate even broader benefits.
Accenture estimates that by tapping AI systems, Federal
agencies could unleash ``a productivity windfall for the U.S.
Government worth up to $532 billion annually by 2028.''
Just earlier this week, the U.K. Government reported that a
new trial of AI tools revealed how civil servants who used them
freed up 2 weeks a year in working time, and 82 percent of them
said they wanted to continue to use AI tools to boost their
productivity. AI tools can be particularly helpful in
simplifying paperwork and procedural hassles. AI has already
been used to improve public records management at many
different agencies, including document declassification and
Freedom of Information Act requests. AI can also help
streamline specific regulations and make them more cost
effective to ensure greater value to taxpayers. AI is already
helping governments better administer various public services,
including infrastructure management, public safety efforts,
healthcare delivery, environmental monitoring, emergency
responses, workforce development, and various other citizen
services.
As Congress and the executive branch look to continue to
clear a path to efficient government AI uptake, the focus
should continue to be on addressing barriers to implementation
through five main priorities. First, modernize Federal
acquisition policies to boost AI adoption, especially by
encouraging easier acquisition of sophisticated but cost-
effective commercial systems. Second, streamline paperwork and
use requirements associated with Federal AI contracts and
deployments, especially for small businesses and open-source
providers. Third, ensure government data sets are AI ready and
interoperable such that the public, researchers, and other
organizations can more easily benefit from them. Fourth,
enhance in-house government AI talent and improve technical
literacy at all levels. And fifth, boost trust and security in
government technology systems.
Congress has already passed several laws to facilitate some
of these goals, and bipartisan efforts by both the Biden and
Trump Administrations have helped expand those efforts,
including through chief AI officers in Federal agencies. The
OMB recently released very important guidance requiring
agencies to ``adopt a forward-leaning and pro-innovation
approach that takes advantage of this technology to help shape
the future of government operations,'' and ``harness solutions
that bring the very best value to taxpayers.'' This Committee
and others should assist these efforts by helping the
Administration ensure that agencies are AI ready for the
future. That means closer oversight of agency modernization and
digitization efforts plus the necessary funding to get the job
done.
Finally, Congress should recognize there is an important
connection between broader AI policy and the benefits
governments can gain from AI systems. Proposals to regulate AI
systems are proliferating rapidly with over 1,000 AI-related
bills already introduced just 5 months into 2025. If these
mandates expand, they will significantly raise the cost of
deploying advanced AI systems because complicated, confusing
compliance regimes would hamstring developers, especially
smaller ones. A confusing patchwork of state and local red tape
will not only undermine the market for commercial systems, but
it will also undermine the government's ability to choose from
a diverse array of competitive, cost-effective options.
One recent study from CCIA reported the Federal preemption
of state-level AI regulation would benefit the Federal balance
sheet by up to $269 billion over the next decade by lowering
procurement costs for the Federal Government. Congress needs to
protect the interstate marketplace and ensure a robustly
innovative and competitive AI ecosystem can develop. Some
degree of preemption is needed to achieve this goal and ensure
a diverse array of cutting-edge AI solutions are available to
government agencies and to the public as consumers of both
private and public services.
Thank you so much for inviting me here today. I look
forward to your questions.
Ms. Mace. Thank you, and, Mr. Schneier, you are now
recognized for your opening statement.
STATEMENT OF BRUCE SCHNEIER
FELLOW AND LECTURER
HARVARD KENNEDY SCHOOL
Mr. Schneier. Thank you, and thanks for inviting me. So, I
am a technologist. I have been working in cryptography and
cybersecurity since the early 1990s. I write, I teach, I
consult, I start companies. I thought a lot about the role of
AI in democratic government. I actually wrote a book on this
topic that will be published in October. The previous four
speakers have talked about the promises of this technology. I
want to talk about the national security implications of the
way our country is consolidating data and feeding it to AI
models.
Chairman Comer said in his statement about this hearing
that he wanted to unleash AI responsibly while protecting
interests and rights of all Americans. DOGE's actions do
exactly the opposite. Over the last few months, DOGE's
affiliates have spread across government. They are still
exfiltrating massive U.S. data bases, processing them with AI,
and offering them to private companies such as Palantir. These
actions are causing irreparable harm to the security of our
country and the safety of everyone, including everyone in this
room, regardless of political affiliation.
This unprecedented security risk is a result of two things:
DOGE's sloppy cybersecurity regarding our data and regarding
the networks they have accessed, and also the over trusting of
AI, current technology, to take over human tasks. You all need
to assume that our adversaries have copies of all the data DOGE
has exfiltrated and have established access into all the
networks that DOGE has removed security controls from, and your
data can be used against you. Consider the coercive power of
your financial records or your medical records or the
government clearance form you filled out. Whoever you think of
as your enemy or an enemy of the United States, they can use
this data to coerce you or a family member or a member of your
staff, and this is true for any elected official or CEO or
police officer or judge or nuclear power plant operator. Like,
really, it is true for everyone. The more powerful you are, or
the more critical your role, the more likely it is that someone
will seek to use your data against you.
There is also a military concept called preparing the
battlefield, which are the things that you do to the enemy in
peacetime to make wartime easier. So, China hacking backdoors
into our power grid is an example of that, and preparing to use
our data against us is another. So, you should assume that any
significant military action against United States will start
with every general's bank account being zeroed out. And maybe
your bank account is being zeroed out because there is nothing
more distracting than your distraught families.
Data is power. Any entity, government or corporate, that
holds individual data has the ability to understand, predict,
even manipulate behavior. This is Facebook's business model.
Our government collects incredibly intimate data about
Americans, but its power is limited by how that data is
organized. There was security in our data being spread among
different government agencies with different rules governing
use and sharing. Using AI to reason across disconnected data
stores represents a massive increase in government power and,
therefore, a security threat. DOGE's sloppy security practices
also means we cannot trust the data they are using. Did they
make copying mistakes? Did some adversaries slip in bad data.
Remember that nonsense about 150-year-olds collecting Social
Security? That was someone misunderstanding how the data was
encoded.
Our adversaries are certainly capable of penetrating the
security of any of the companies that DOGE gave our data to and
just as capable at poisoning our AI systems. Data integrity is
vital, and good AI fed with bad data makes bad decisions and is
untrustworthy. Irreparable damage has already been done, but
the damage is ongoing. 2015, China hacked OPM, stole data about
all Federal workers. That was a major security breach.
Continuous real-time access to data is much more dangerous. The
longer these vulnerabilities persist, the more adversaries will
be able to manipulate data and install backdoors, the more they
will be able to manipulate the AIs we are creating, and the
more they will be able to secretly influence our policy
outcomes.
Sacrificing cybersecurity in an effort to create an AI
future not only risks the country, it risks the AIs, which in
turn risks our justice system, our legislative system, our
banking system, national defense, and all of us, all of you
personally. So, thank you, and I welcome questions and
conversation.
Ms. Mace. Thank you. I will now recognize Mr. Gosar for 5
minutes.
Mr. Gosar. Thanks, Chairwoman. Ms. Miller, you described
the COVID process as a catastrophic event with a high risk of
fraud. I agree. I am the only non-U.S. resident in the United
States to have ever killed a national emergency. I killed
COVID, but there is not a single receipt, though, and since
Bill Clinton through Joe Biden, we have spent over $13
trillion--let me say that again--$13 trillion without a
receipt. Could we use AI to examine the funds spent under
COVID?
Ms. Miller. It depends on where that data is available.
Certainly, AI can be used to mine any data sets that exist. The
question I think you are really asking is whether or not that
data exists, and I do not have the answer to that question.
Mr. Gosar. Well, I hope that we would have some data base
because we have a record of where it went, supposedly.
Ms. Miller. Well, yes. So, we have a record of where a lot
of the spending went, and of course the Pandemic Response
Accountability Committee, the organization that I helped lead,
has a lot of that data on their website about where money was
spent during the pandemic. A lot of that money was stolen,
something I talk about a lot. We do not know where that money
went. A lot of the money that was stolen during the pandemic
went to overseas foreign adversarial nation-states.
Mr. Gosar. So, they also could have--how should I say this?
My understanding, there was over 2,000 cases in Arizona alone,
that entities or individuals were actually sued by the
government to get it back, resulting in compensation of over
$300 million. We could use that as well, if it disappeared?
Once again, this is based on the data, right?
Ms. Miller. Yes, it is based on the data. I do not think I
am qualified to answer that. Thanks.
Mr. Gosar. OK. Mr. Shah, you talked about our procurement
system. Boy, it is broken. So, how would you use AI during that
whole process, because our overruns in our military are
outrageous. How can we keep them on track? How does that work?
Mr. Shah. Yes. I think there are a variety of things to
talk about on that topic. One, you can use AI to help with the
procurement process itself in terms of analyzing vendors,
summarizing submissions, generating use cases and requirements
for the different agencies that are looking for new technology.
But I think before that, we need to sort of create an
environment where these new technologies that can be applied to
procurement or anything else in the U.S. Government can make
their way into these agencies. And I think that is a human
process problem that we have an opportunity here in this
Committee to really kind of provide some oversight and
correction so that we can go about delivering these great new
technologies into these agencies quickly, right?
We have spent years working on this ourselves to try and
get into agencies, but if you look at a lot of these
procurement, sort of processes, they are multiyear. They are
outdated from the 1990s where, essentially, you are thinking
about installing software on some server rack that is going to
last 5 years, and it might take 2 years to actually go through
the procurement process. This is untenable for young companies
who have limited cash, limited capital, who want to be able to
monetize their investments and especially help the U.S.
Government move quickly.
Mr. Gosar. So, could you use something like a blockchain?
See, I am not a techie. I got questions. Could you use a
blockchain to kind of utilize some type of--I see you. I will
come back to you in a second, but I want Mr. Shah. Is there
some kind of technology that will help preserve some of this
data or save us?
Mr. Shah. So, we do not deploy blockchain technology for
our solution. We focus squarely on what employees can benefit
from when it comes to using AI to be more productive, so I am
unfortunately not able to answer how blockchain could be used
here. But, generally speaking, technology has been shown in the
private sector to have tremendous impacts on their efficiencies
and their ability to stay ahead and to lead in the industry.
Mr. Gosar. Yes, Ms. Miller, I got one more question for
you. Could you use AI to audit Medicaid, Medicare? Any
comparisons with hospital versus hospital?
Ms. Miller. Yes, that is a very, very good use case for AI
because there are a lot of claims data, and there are a lot of
information on providers, and you can look at all that data
using AI very quickly and identify patterns that look
suspicious. It is an area that I think really is ripe for a lot
of innovation in the space because CMS has the opportunity to
really mine a lot of that Medicare and Medicaid data using
artificial intelligence to identify taxpayer dollars that are
being stolen.
Mr. Gosar. Gotcha. Well, I am out of time. I yield back.
Ms. Mace. Thank you. I will now recognize Mr. Lynch for 5
minutes.
Mr. Lynch. Thank you, Madam Chair. Mr. Bajraktari, you
mentioned in your testimony the value of educating AI literate
workers, right? And Mr. Shah, I think you alluded to the same
thing. About 25 years ago, I co-founded a charter school in
Boston, a public charter school, and what we did was, we saw
that most of the jobs that were being created were in math and
science and STEM-related industries. So, we created this school
that doubled or tripled the amount of educational time that
students would be exposed to math and science. It has worked
very, very well. Those students score very, very high on math
and science and in the MCAS.
Unfortunately, on Monday, the Trump Administration
rescinded $325 million from the STEM education grants that we
provide to feed that type of activity, that sustain a move
toward stronger STEM disciplines within our student bodies, and
that was money that was already authorized, so it was not up
for debate. This is the stuff we did last year. Cutting $325
million in STEM grants, is that going to help our effort here
on AI?
Mr. Bajraktari. Ranking Member Lynch, I think we have to be
strategic about how we approach AI adoption at all levels of
our educational system. I think if you look at what China is
doing, they are introducing AI at every level of their
educational system. I think for our kids going forward, having
a capability in their pocket to have access to an Einstein AI
agentic application that can help them toward tailored
education would be an enormous benefit for the next generation
of Americans.
Mr. Lynch. All right. Reclaiming my time. Learning math and
science would probably help as well. I am going to just reclaim
my time. Thank you though. Thank you. I think you are on the
right track there.
Mr. Schneier, can you talk about--so, China has adopted an
AI system that is total surveillance of their entire population
there. It is non-democratic. It is quite oppressive. Their goal
is to be able to surveil every single member of their society,
know what they are doing, know what they are purchasing, and
for oppressive reasons so that no one gets out of line. Can you
talk about the need to, when we adopt, when we approve AI and
choose the architecture for our AI, how important it is to make
sure that transparency, that respecting privacy, independence,
how important it is to have those elements of democracy itself
ingrained and embodied in our AI models?
Mr. Schneier. It is vital, and this is why you will see a
lot of movements toward public AI, the notion of an AI model
that is not built and owned by a for-profit corporation, right?
You can imagine other entities creating an AI, and that is
really a matter of trust and integrity. Can we trust Grok or
even the model from Google or Amazon or any of the other
models, or do we as a country need a model that is built on
American democratic values and not necessarily corporate
values? It is critical to think about this. Do not think of it
as one AI. There will be many of them.
Mr. Lynch. Right.
Mr. Schneier. I mean, already there are over a million
models on Hugging Face that are free, and it is becoming
cheaper to create these models.
Mr. Lynch. In your testimony, you talked about the dangers
that Elon Musk created by the way he was diving into Social
Security data and IRS data, and the danger that that presented
to the United States from a national security standpoint and to
the privacy of U.S. Citizens. Can you talk about that, please?
Mr. Schneier. I mean, it is really a matter of process. The
goal might be a good idea, but the reason we have controls
around this data, who can access under what rules in what
configurations, are because it is so sensitive. You think about
the different data enclaves, whether it is census or military
data or OPM data about government employees or health data.
This is very critical, and we, as citizens, give this data to
the government for good reasons. And it is kept separate for
good reasons, and it is kept secure. So, when we read about
laptops of data being removed, data being given to Palantir to
train models that they are going to use to sell back to us, the
security is a real concern.
Mr. Lynch. Madam Chair, I yield back. Thank you.
Ms. Mace. Thank you. I will now recognize Mr. Higgins for 5
minutes.
Mr. Higgins. Thank you, Madam Chair. Mr. Thierer, I am
going to be talking with you mostly for the next 5 minutes
because I would like to dig into the controversy regarding the
moratorium section of the BBB legislation currently being
considered by Congress. It is across the country, as part of
the narrative of the discussion of AI, and I believe we have an
opportunity today to clarify the intent and the specifics of
that section of legislative language in BBB. So, in that topic,
and please help me to understand, and, in doing so, help the
Nation to understand, good sir.
My read of the actual bill states, to that section in
particular going to the moratorium under the section of
artificial intelligence and information technology
modernization, it says in general, except as provided in
paragraph 2, which is essentially an exception to the
moratorium that it allows for sovereign states to enact and
enforce laws that remove restrictions on use of AI. What the
paragraph says in general, except as provided in Paragraph 2,
no state or political subdivision thereof ``may enforce''
during the 10-year period, beginning on the date of enactment
of this bill, any law or regulation of that state.
So, my read is that the state is not restricted from
debating and considering, and voting on and passing, and using
their legislative bicameral body in each sovereign state to
debate and vote on and pass laws regarding AI. And for that
Governor to sign that law, the Big Beautiful Bill in this
section states that the Federal Government is asking for a
pause on enforcement. Am I right or wrong, and please
embellish.
Mr. Thierer. Yes. Thank you, Congressman for the question.
So, I testified about this in front of the Energy and Commerce
Committee just 2 weeks ago where there was a debate about the
AI moratorium. And it is important to keep in mind that the
reconciliation bill includes specific funding to modernize
Department of Commerce technology systems, and there is a
concern among Members about exactly how that can happen if
there is a host of conflicting laws at the state level.
Mr. Higgins. Right, and the so-called moratorium is under
that section.
Mr. Thierer. That is correct.
Mr. Higgins. Please continue.
Mr. Thierer. And so, therefore, that would prohibit certain
AI-specific types of regulations at the state and local level
again----
Mr. Higgins. It would prohibit the enforcement of state-
authored regulations.
Mr. Thierer. That are AI-specific regulations. It very
specifically says that laws of general applicability are not to
be covered by this, also criminal activity not covered. And so,
there are some important exceptions to this, and states are
going to be free to continue to act on a lot of these different
things.
Mr. Higgins. In the interest of time, just to clarify for
the country, sir, this moratorium does not restrict a sovereign
state's right to debate, vote on, and enact new AI legislation
within that state. The moratorium is asking the states to
recognize that AI, by nature, is interstate commerce, and under
this section for modernization of the Commerce Department of
the reconciliation bill, we are stating that the states should
refrain from enforcement of new AI laws. Is that correct?
Mr. Thierer. Yes, Congressman, and specifically, what you
said about protection of the interstate, the national
marketplace, is really crucial, and it reflects the broader
consensus of Congress going back over the last 30 years about
the internet and technology more broadly, that there does need
to be somewhat of a national framework and make sure that we do
not have a parochial patchwork that will interfere with the
free flow of algorithmic commerce and speech.
Ms. Mace. Six seconds.
Mr. Higgins. OK. That was a squared-away explanation. I
thank the gentleman. Madam Chair, I yield.
Ms. Mace. Thank you, Mr. Higgins. I will now recognize Ms.
Norton for 5 minutes.
Ms. Norton. Thank you, Madam Chair. Government must
leverage the best technology possible to meet the ever-changing
needs of the people it serves. From the creation of the
national Postal Service, to the discoveries that led to the
internet, the Federal Government has a long history of
embracing new technology to better deliver for our
constituents. But our North Star must always be service to the
American people, not tax breaks for billionaires or back-room
deals to benefit the richest man in the world at public
expense. As we discuss the use of artificial intelligence in
government, we must have people at the center of the
conversation, including the people we represent and Federal
workers who serve our country. AI is a tool that can empower
our public servants, but it is not a substitute for passionate
and committed Federal workers.
Ms. Miller, can AI alone replace Federal workers?
Ms. Miller. Absolutely not, no. The implementation of AI,
it is very important when we think about building AI tools,
that we are building them with the people that we will use
them. AI is talked about a lot as something that is going to
replace people. I am a strong advocate in using AI to automate
routine processes and repeatable processes and free up people
to do more high-level, high-skilled work. And so, I absolutely
do not believe that AI should be doing the work of the public
servants. I believe that AI should augment that work and that
we should work with public servants as we put in place AI
tools.
Ms. Norton. Thank you. Elon Musk and the Department of
Government Efficiency has gutted so much of the Federal
workforce, firing tens of thousands of people under the guise
of greater efficiency, but Americans know better. They see the
negative impacts every day. As a direct result of the
interference with Social Security Administration, senior
citizens now face 98-minute wait times to speak with customer
service, crowded lobbies at field offices, and being turned
away from scheduled appointments because the Agency simply does
not have enough employees for its mission. Poor customer
service is not efficiency. It is failing to uphold the
government's responsibility to our constituents. Now the Social
Security Administration wants to cut an additional 7,000
employees and somehow use AI to handle calls from seniors. That
does not work because I know that when my constituents have a
problem with Social Security, they would rather talk to a
person than a computer program.
The Department of Government Efficiency cuts are wreaking
havoc on our communities, and it will take years to clean up
the mess Elon Musk has left behind. Thank you, and I yield
back.
Ms. Mace. Thank you. I will now recognize Ms. Foxx for 5
minutes.
Ms. Foxx. Thank you very much, Madam Chairman, and thanks
to our witnesses for being here. Mr. Thierer, the Trump
Administration has expressed a goal for the United States to
maintain its dominance in AI. Why is it important for the
United States to maintain its dominance position in AI?
Mr. Thierer. Thank you for the question, Congresswoman. As
you have heard here today from several witnesses, we are in a
serious race with China and the Chinese Communist Party for AI
supremacy globally. And the broad-based ramifications of who
wins that race are significant for not just our geopolitical
security, but also our values broadly as a Nation and for the
world. And so, there is a real danger that if we do not get AI
policy right as a Nation, that China will take the lead. Its
products will diffuse throughout the world faster, and we are
already seeing this. I was testifying earlier this year at a
hearing on the so-called DeepSeek moment, and it surprised a
lot of us, including me, how advanced these Chinese systems are
today and how effective they have been at filling gaps globally
that America is not there to fill at the moment.
Ms. Foxx. Thanks. I do think the answer is pretty obvious,
but it was important to get something on the record.
Mr. Thierer. Sure.
Ms. Foxx. How has the Trump's Administration approach to AI
differed from the previous Biden Administration's, and why is
the departure from the Biden Administration's AI policy
necessary, Mr. Thierer?
Mr. Thierer. Well, the Biden Administration approach was
very voluminous in terms of rulemaking, obviously, executive
order, longest in American history on AI, over 110 pages. And a
lot of it was very fear based, unfortunately, and did not
embrace the technology as much as the new Trump Administration
executive order and subsequent orders have done. And so, I
think there has been a real sea change of approach here in
terms of understanding the benefits associated with AI
technologies broadly.
Ms. Foxx. Thank you. Mr. Shah, your company has been
successful at streamlining certain administrative tasks for
businesses in the private sector and helping boost
productivity. How do you see AI technologies like those
developed by your company being effectively used in the Federal
Government, and what kinds of productivity gains and cost
savings could we expect to see?
Mr. Shah. Thank you for the question. You know, to put it
in maybe legislative terms, imagine you had an assistant that
could summarize the status of a bill. Perhaps maybe look up a
constituent problem that you have been tackling before a
meeting or be able to search all of your notes and your memos
about various Committee hearings and tasks that have been
assigned. All these things are things that AI gives us today,
and the private sector is using this quite aggressively to make
their own teams more efficient and effective.
To your question about sort of what that looks like, you
know, there are sort of three categories I like to think about.
Scale. Broadcom, for example, was able to go from 10,000
employees to 50,000 while keeping the same support staff. So,
that is another benefit, which is you can take the same team
and allow them to support a lot more employees in this case,
but in our case here today, talking about constituents. I think
there are also employee time savings. The city of Glendale I
mentioned did a 500-percent plus ROI on their investment, but
that was 3,500 hours of savings from the employees themselves
that they could apply and use toward other tasks and functions
for their constituents. And then, of course, there are monetary
savings, which is that if we can do some of this more
efficiently, that goes back to the U.S. taxpayers. It goes back
into the economy to get spent on further innovation.
Ms. Foxx. Thank you very much. Ms. Miller, according to the
Office of Management and Budget, each year, taxpayers provide
more than $1.2 trillion, or nearly 5 percent of GDP, in funding
for thousands of programs across the entire government through
grants and other financial assistance. With so much taxpayer
money spent, it should come as no surprise that, besides a very
high-level accounting, the Federal Government largely cannot
tell whether a particular grant award was used wisely or if it
accomplished the stated goals of the grant program. This
Committee has worked on legislation such as my bill from 2019,
the GREAT Act, to address the problem, but as we have
discovered, there is not enough data available in the proper
formats to evaluate whether taxpayer funds given out as grants
are being used wisely.
How could AI help us solve this problem and make sure that
every taxpayer dollar given as a grant is used effectively and
for its stated purpose?
Ms. Miller. Thanks for the question. Actually, we did
something pretty innovative with my technology company,
TrackLight. We were trying to figure out who one of the
agencies that we were working with paid, and so we got access
to minutes of board meetings, and then we used AI to scrape
those minutes and identify vendors. And then we were able to
collect hundreds of vendors that this agency was paying that
they did not know they were paying. Now, we can all talk about
how insane it is that the agency does not know who they are
paying, which is a serious problem. But there are ways that AI
can help solve this problem and start to, sort of, piece
together who these grants are going to, and it is incredibly
important for taxpayers to know that information.
Ms. Foxx. It would be nice if we could just search things
on fireside. Thank you.
Ms. Miller. Yes.
Ms. Mace. Yes. All right. Thank you, and I will now
recognize Mr. Mfume for 5 minutes.
Mr. Mfume. Madam Chairwoman, thank you very much. My thanks
to the guests who have come today. One of the things that
strikes me about this hearing is that we do not have enough
time, and we are all trying to get a lot in here. So, I will
certainly use the right that Members have to submit questions
for the record for 5 business days because there are a number
of you. Mr. Shah, in particular, I want to ask some questions
of you through that process, and, Mr. Schneier, your testimony
was chilling to say the very least. So, hopefully, we can talk
about that, but, Ms. Miller, I would like to start with you.
Much of your work focuses on fraud prevention, and before
we get to that, I would like to go back to something that I
heard that was just as chilling, and that was your description
of these outdated privacy laws, which do not afford any
protection whatsoever to what is happening now in 2025 at the
rapid pace of this technology and its development. It is very
scary to believe and to hear, quite frankly, that we are behind
in that because the privacy of Americans is absolutely
important. It is just as important as what Ms. Norton brought
up earlier, and that is the fact that people who are looking
for Social Security or trying to get Medicaid or trying to
figure out SNAP benefits are running into this technology and
being run over by it in many instances. So, can we talk just a
minute about these privacy laws, which I assume they predate
all of this discussion, and that there is, as I understand, not
a real determined effort to upgrade them to be able to protect
industry, and to be able to protect government and individuals?
Ms. Miller. Yes. Thank you, Congressman. The Privacy Act
was passed in 1972, which was 52 years ago, almost 53. That is
still the law that governs how we protect privacy today, which
obviously is exceptionally outdated. In my opinion, there is
not a lot that can be done through the legal system to protect
Americans' personally identifiable information anymore. I think
we just live in an entirely new world now, and foreign
adversarial nation-state actors, criminal rings, all of our
data has been monetized, is being monetized today. Ninety-six
percent of information that is stolen through a data breach is
used to monetize by a threat actor.
And so, because we have such an asymmetric situation when
it comes to data, government is going to have to get much
better about using technology to protect privacy. And there are
technologies that are in place, I am sure, that some of the
panelists here today are using in the private sector. We can
anonymize data. And so, we can do things that can protect
people's privacies, and we are going to need to think about
this both from a legislative perspective, not hamstringing
agencies. The Privacy Act and the Computer Matching Act keep
agencies from being able to share data that they desperately
need to be able to find these actors, and, again, these actors
are foreign adversarial actors.
Mr. Mfume. Thank you very, very much. Mr. Schneier, let us
go back to your assessment of the bigger you are, the bigger
target you are, the more data is collected on you, and the more
harm that will come to you, your community, your agency, your
government in this instance. Walk me briefly down that street
again because you did not go into the fact that this is not
science fiction. This is happening every day, and it is
happening to the United States of America.
Mr. Schneier. I mentioned China, the OPM breach from the
last decade where China went into OPM and took the data, among
other things, all of the clearance forms of United States
citizens, and you can imagine why they might do that, and why
they might identify who spies are, might look for people they
can influence, and that is the sort of thing that countries are
doing today. I mean, I am assuming we are doing it to them, and
so this data is very valuable at the government level for a lot
of different things.
I mentioned two scenarios. I mentioned a potential war
scenario, potential peace scenario, and a lot of--our data is
being collected everywhere. And so, I mean, a lot of data is
had by social media companies that know a lot about what we
like, and who we are and who our friends are, and who is
important to us, but the data we give governments is a
different sort of data If you think about the data that is on
the tax return or the data that the Census Bureau collects, or
the military data, that is personal on a different level, and
countries want that. They are going to want that in aggregate
to figure out things about us, and they are going to want it
individually. And coercion means--lot of it depends on who you
are, what coercion means.
Mr. Mfume. Thank you. My time has expired. Madam Chair,
thank you very much.
Ms. Mace. Thank you. I will now recognize Mr. Sessions for
5 minutes.
Mr. Sessions. Madam Chair, thank you very much, and good
morning to each of the panel members. We appreciate you being
here.
Long time ago, I spent time at Bell Labs where we were
responsible for the network architecture but also the
development of the switch that would be utilized. We set
standards and did things to where we had unified way to look at
how switch operated. We both know we had Salt and Volt Typhoon
whereby our government people learned that the Chinese had been
into, through cyber, our data bases for a long time. Should
have found it. Did not find it. It cost us. My question really
is, and I am sure every one of you have an answer for this or
at least your vision of that. All it takes is one node to be
compromised. It is like a big fence that goes around. It just
takes one loose place.
Does the Federal Government, do we here in Congress need to
ask someone? Do we need to deal with other countries, that
there need to be a standards body that specifically, like when
I worked at the Labs, to set what those standards are? How do
we avoid, with all the things we are doing, getting something
that happens like that again? Sir, evidently you have got your
hand on the panel first. Any ideas of that?
Mr. Schneier. I did not know this was Jeopardy, that you
got to be quick.
Mr. Sessions. Well, OK. Then we would go to you last if you
want.
Mr. Schneier. No----
Mr. Sessions. But my point is, do you have any ideas on
these question?
Mr. Schneier. So, we have things. I mean, this is the job
of the NSA, right? We have security organizations in the United
States whose job is to protect American data, and I think we do
actually do a pretty good job. And we have problems that a lot
of data is in private hands, that, you know, right now, if you
think about just the Defense Department, there are thousands of
networks, but there are rules, and, you know, I think we are
doing OK. I do not think we need an international standard. I
mean, this is very much United States. I mean, NATO does things
together, so there are groups, but this is not like a corporate
security thing where you need an international standard.
Everyone is consistent. The U.S. has U.S. problems, U.S.
adversaries, and we do have U.S. solutions.
Mr. Sessions. We also have battleships and aircraft
carriers that are out in the middle of places. Anyone else?
Thank you, sir. Mr. Shah?
Mr. Shah. Yes. I think, you know, what we have seen,
especially in cybersecurity, the NIST has created standards
there that became adopted by a majority of the industry. And I
think when it comes to AI and protecting data, coming from the
lens of a startup, from a young company, we have looked at the
new standards that they have around AI and they are actually
pretty good, and they do provide for a lot of recommendations
that we actually follow. And then furthermore, you know, going
through the FedRAMP process, while I found it to be very
expensive, there was a lot of amazing sort of things that they
recommended in terms of infrastructure, software, process, and
procedures that allow us to really make sure that the data is
secured. And so, I think that if we can have more of these
types of frameworks, more of a lighter touch, it will allow us
to continue to innovate fast but still give us all a true north
of where we should take things in terms of our infrastructure
and our processes.
Mr. Sessions. Sir?
Mr. Bajraktari. Sir, I would say three things. No. 1, just
to follow on Mr. Shah's point, I think NIST is well positioned
for this. I think the re-missioning of the Safety Institute by
the Department of Commerce 2 days ago, that will focus on
really building the standards across our Federal Government and
how they use AI, is a step forward. Also, looking at, you know,
open-source models coming from countries of concern,
specifically China, and what kind of a risk they possess in our
ecosystem, I think that is the right step to do it. At the
allies and partners levels, as I mentioned in my opening
remarks, The Five Eye Alliance really provides us with a
stepping stone of, you know, coming with some common standards
among our five ally partners and then start bringing other
countries under the fold. And the third piece, as Mr. Schneier
mentioned here earlier, what can we do on the offensive side so
our adversary does not get a hold of our data, they do not use
it against us, and then actually we put them on the back foot
instead of letting them, you know, really attack us on a daily
and hourly basis.
Mr. Sessions. Good. Anyone else?
Mr. Thierer. Just very briefly. Sir, I just want to put in
a good word for what the Trump Administration has been doing
with its recent executive orders and OMB guidance on this
front, which has been to provide not only a framework for
expanding the use of AI throughout government, but also getting
more serious about security vulnerabilities and other types of
concerns, and recent OMB guidances like M-25-21 and 22,
basically try to address these concerns you are raising.
Mr. Sessions. Great. Thank you. I appreciate the panel very
much. Madam Chair, I yield back.
Ms. Mace. Thank you. I will now recognize Ms. Brown for 5
minutes.
Ms. Brown. Thank you, Madam Chair. I am glad this Committee
is finally turning its attention to the Federal Government's
use of artificial intelligence because under President Trump,
AI is not being used to serve the people. It is being abused to
serve the powerful. DOGE was a reckless, unregulated, and
unethical failure, and I would not be surprised if some of my
colleagues on the other side quietly agree now that Elon Musk
is out here trashing their one big ugly bill.
So, let us be clear about how we got here. Musk and his
cronies were handed the keys to our government's data, let
loose to sidestep security safeguards, strip away Federal jobs,
and exploit Americans' personal information unchecked and
unaccountable. They used AI as a blunt force tool to slash
thousands of Federal jobs and gut billions in government
spending with no transparency or regard for the human cost, but
that is not even the most dangerous part. DOGE has launched an
unprecedented power grab over Americans' personal data. Earlier
this year, reports surfaced that a DOGE staffer, with no
credentials and no clearance, reportedly accessed Americans'
personal information and shared private data with third
parties. I co-led a letter with Reps Trahan and DelBene
demanding an investigation, and it might not come as a surprise
that we are still waiting for answers.
And we now know this was not a one-off. Whistleblowers have
revealed that DOGE is actively building a master data base, a
centralized system combining sensitive data from across Federal
and state agencies that includes Americans' medical records,
their credit card numbers, their medical claims, their student
debts, and much, much more. To build it, the Administration has
turned to Palantir, the surveillance firm funded and founded by
Republican MAGA donor, Peter Thiel, and this massive trove of
personal information is being built with no oversight, no
security guarantees, and no guardrails.
Here is why every Americans should be alarmed. In the wrong
hands, this data base could be weaponized to target political
enemies, suppress dissent, or surveil the public, and even in
the best-case scenario, a breach could expose millions to
identity theft, blackmail or worse. What is most disturbing is
how much we still do not know, but what we do know raises
serious concerns about violations of the Privacy Act, FISMA,
and E-Government Act. This Committee must keep investigating
because we cannot allow AI to become a weapon for surveillance,
exploitation, and the dismantling of public institutions.
So, Mr. Schneier, how should Americans feel knowing that
DOGE is pulling together their most sensitive personal data to
train AI without clear safeguards, oversight, or consent, and
what protections are in place to keep that data from being
misused or exposed?
Mr. Schneier. Well, we do have a lot of laws in place. We
have laws protecting the data. We have laws prescribing how it
is used, how it is shared, and that is what is being bypassed.
So, it really is more a matter of enforcing the laws we have on
the books than writing new laws. I do not think we have a need
to do anything because of AI because these problems with data
sharing/security risks have been with us for decades. And we do
have laws, we do have rules, and they are being bypassed.
Ms. Brown. So, let me ask you this. Given DOGE's reported
practice of bypassing security protocols and exporting
sensitive data to less secure environments, how serious is the
national security risk, and are we making it easier for foreign
adversary hackers to access government systems and Americans'
personal information?
Mr. Schneier. I mean, we are. We knew from the first weeks
that some things DOGE was doing was bypassing security
protocols and removing audit trails. I mean, basically
eliminating the security mechanisms we have in place to protect
the data. We had one example of a Russian actor going in after
a DOGE employee using their credentials, and that is the one we
know about. These countries are not stupid. They know how to do
espionage. They have been doing it for decades. And so, I think
what I said in my remarks, we have to assume that all of this
data has been exfiltrated by our enemies. I would be very
concerned. Now, it depends on who you are, the more powerful
you are, the more concerned you are because more leverage can
be leveraged against you.
Ms. Brown. Thank you. Let me be clear. Artificial
intelligence and other emerging technologies can offer real
benefits but only when developed responsibly and with clear
guardrails that protect the American people. So, I stand ready
to work with my colleagues on the other side of the aisle who
are serious about securing our Nation's cybersecurity, safely
deploying AI within the Federal Government, and ensuring that
Americans' data remains safe. And with that, Madam Chairwoman,
I yield back.
Ms. Mace. Thank you. I will now recognize Mr. Biggs for 5
minutes.
Mr. Biggs. Thank you, Madam Chair. Thank the panelists for
being here today.
Mr. Shah, in your written testimony, you describe how
various communities, counties, cities, and two of them in
Arizona--Scottsdale and Glendale--are actually using AI,
perhaps becoming a hub of AI, anyway, using to become more
efficient. Can you expand on that for us? Tell us what they are
doing and the guidelines or guardrails that they have in place.
Mr. Shah. Yes, it is a great question. So, you know, in
terms of the cities and counties that have started to adopt
this technology, they are using it to make their employees more
effective. What does that mean? That means that they can self-
serve a lot more use cases than they would have otherwise. They
can find information. They can troubleshoot issues that they
might be having with IT or even an H.R. situation or use case.
They can submit performance reviews. They can do a variety of
tasks that were otherwise very manual and very slow, and
required a lot of coordination amongst folks.
And so, you know, as I was mentioning earlier, the city of
Glendale in Arizona is saving their employees 3,500 hours now
each month, being able to now do more with those same team
members to support the employees or the constituents of their
county. And so, I think that we see this across the private
sector. We have seen people use this technology to enable more
employees to do greater levels of work and greater levels of
output, while still sort of maintaining all the privacy and the
security that is required in any of those environments.
Mr. Biggs. Thanks for that. Now, Mr. Schneier was talking
about the laws that have been placed to protect privacy rights,
and they have been in place for a long time, largely ignored
from both parties, quite frankly, in my opinion. But what I
want to get to is something that Mr. Bajraktari said, when you
start talking about The Five Eyes community. And I just came
from another hearing where one of the issues that we are having
is that, pursuant to the CLOUD Act, the U.K. is trying to get
backdoor hacks into Apple, and they are claiming that they have
some kind of robust privacy protection there. But the reality
is they will not disclose what that robust privacy protection
is, and they are keeping the entire process closed, obscured.
And so, I am curious about how we can fully work with The Five
Eyes community in the development of AI and yet protect the
fundamental rights of American citizens, and so I am going to
ask you that question.
Mr. Bajraktari. Thank you, sir. Great question. We created
The Five Eye community during the cold war because one of the
ways we could share more and fast and be ahead of what Soviet
Union was doing was, you know, being able to share with our
allies and partners. I think there needs to be a starting point
in how we confront China today. I do not think we can confront
China alone without bringing, you know, our closest allies
under the fold. The speed and scale of how China adopts and
uses AI for national security purposes is really breathtaking.
So, my point is, we should start with a community of the most
trusted allies and partners that we have.
Mr. Biggs. So, my question is, though, when you have,
essentially, the abuse of rights that we are seeing through the
abuse of the CLOUD agreement between U.S. and U.K., can you
fully trust The Five Eyes to work together on this? And that
is----
Mr. Bajraktari. Yes, I understand. I understand your
concern, sir. I think we have to work really closely with them
to remove obstacles because I think the challenge we face ahead
of us is that, if we do not find a common language and build
common security protocols, I think we would have a different
world by the end of this decade, you know, a world in which CCP
will probably have access to all our data. The entire global
infrastructure will depend on the CCP and the companies
subsidized by Chinese, so we have to find ways to remove these
obstacles with our closest allies.
Mr. Biggs. I do not disagree with that. It is just, we need
to make changes right now, and we are not making changes fast
enough, and we are not implementing anything fast enough.
So, I am going to ask unanimous consent to enter into the
record the bill text from a new Arizona law passed earlier this
month to close dangerous loophole that allowed these predators
to use AI and digital editing tools to create child sexual
abuse material that appears real, but depicts no actual child.
It was HB-2678. It is sponsored by House Republican Majority
Whip, Julie Willoughby, and it updates state law to ensure AI-
generated and digitally altered images that are
indistinguishable from real minors carry the same criminal
penalties as those involving actual children. I would like to
submit for the record.
Ms. Mace. No objection.
Mr. Biggs. I also ask for unanimous consent to enter into
the record an article discussing how my home state of Arizona
is taking steps to allow AI to flourish, showcasing that states
can handle AI regulation responsibly and reap the economic
benefits.
Ms. Mace. No objection.
Mr. Biggs. Thank you.
Ms. Mace. All right. I will now recognize Ms. Stansbury for
5 minutes.
Ms. Stansbury. All right. Good morning, everyone. Thank you
so much for being here, and I want to say thank you to the
entire panel for your recommendations this morning. My
background is actually in the sciences. I have actually worked
on natural and earth sciences most of my career, including on
Big Data and modeling, so I do appreciate much of what the
recommendations are that have come from our experts.
But I am especially grateful that you are here, Mr.
Thierer, because when the huge abomination of a bill, I will
call it, came down the pike, my eye actually caught this
specific language and funding appropriation in the bill that
you referenced in your testimony. And I think many of us were
confused about where it came from, what its intent is, and this
is the first time I have had the opportunity to actually ask
some questions, so I am glad you are here. And as I understand
it from my look into your background, that you have actually
been directly involved in developing this language and request.
Is that correct?
Mr. Thierer. Not that language. I have written about
moratoria in the past, though, going back to the 1990s when I
worked with Congress on the Internet Tax Freedom Act.
Ms. Stansbury. Got it. OK. So, you mentioned that part of
why this deregulation language is in here that would preempt
state and local laws regarding the regulation of AI systems is,
in part, because of this $500 million request that is in the
bill. Can you help us understand what would that $500 million
be used to do specifically?
Mr. Thierer. Going back to the Bipartisan House AI Task
Force report from last December, there was a major section that
led off the report about the need for government modernization
of IT.
Ms. Stansbury. Just some specifics.
Mr. Thierer. So, specifically, at the Department of
Commerce, there is a need to actually improve how various
programs function there, including new ones that have just been
created.
Ms. Stansbury. Just some specifics, like, what kind of
systems would be built with this $500 million?
Mr. Thierer. Well, I do not know exactly. I do not know the
answer to that, Congresswoman. I am sorry, I do not know
specifically what Commerce will use that money for going
forward.
Ms. Stansbury. Do you know where this request came from?
Mr. Thierer. You would have to ask the House Energy and
Commerce Committee specifically who put that language forward
in the----
Ms. Stansbury. Do you know what companies would benefit
from this language?
Mr. Thierer. No, but I do know that there is a broad need
for many different types of vendors and services to be utilized
by the Federal Government, not just----
Ms. Stansbury. So, this would likely go to private
contracts, correct?
Mr. Thierer. It would go to many private contracts,
certainly.
Ms. Stansbury. Yes. OK.
Mr. Thierer. Yes.
Ms. Stansbury. And so, and some of the leading contractors
who do work in this area that already have government contracts
are companies like Palantir, correct?
Mr. Thierer. I believe that is one of many, but there are
many other vendors of IT services and AI.
Ms. Stansbury. So, correct, yes. And I heard you argue just
a few moments ago when Mr. Higgins was asking you about this
preemption language, that it was needed because there would be
integration of data from state and local entities. And it is my
understanding--I think others have kind of alluded to this--
that right now, the Trump Administration is working on creating
some sort of master data base that brings together data from
multiple different sources across state lines. And in fact,
there was a request that went out to the states asking for data
about SNAP beneficiaries because my understanding is Homeland
Security is interested in integrating data for immigration
enforcement purposes. So, they would like to integrate data
using the census data that they hacked in, exfiltrated, and
downloaded. They would like to integrate data with the
immigration data bases that they have, and now they are asking
states to send data that falls under state jurisdiction to
integrate for immigration enforcement.
And, you know, you can call it whatever you want. I have
heard folks say this is not a master data base, but it
absolutely is, and we know that the Trump Administration is
trying to take American data and hire private companies like
Palantir to create means to use American data for their own
policy ends, including endangering the lives of Americans. So,
I think that one of the reasons why folks are reacting so
strongly to this language that is in this bill is because it is
so extreme. And yesterday, I heard Marjorie Taylor Greene, I
was standing on the Floor while she gave a speech on this very
language that is in this bill that would literally say that
states and localities cannot enforce their own data rules or
AI. It preempts federalism. It preempts state law. It would
prevent states from protecting their own citizens.
And you know, it is rare these days that we have bipartisan
agreement, but I agree with literally everything that Marjorie
Taylor Greene said on the Floor yesterday. This is dangerous.
It could endanger American lives, and we cannot allow this
language to get through. And with that, I yield back.
Ms. Mace. Thank you, and I will recognize Mr. Perry for 5
minutes.
Mr. Perry. Thank you, Madam Chair, and, ladies and
gentlemen, thanks for making the trip and being here. I am not
sure what questions I have for you, but I do want to make
everybody aware of some things that I am aware of. I am going
to refer directly to an AI safety firm named Palisade Research,
and another one named Apollo Research and talk about things
like the fact that the Chinese Communist Party likes to pollute
the, I do not know, the available information in the world with
all kinds of things that support the Communist Party of China's
view of the world, and not the free world's.
And with that having been said, I suspect that all of you
are aware of the times that when various AI has been tested. I
will give you a couple names--Codex-mini, OpenAI model o3,
OpenAI model o4--and other ones, the AI has chosen, on
occasion, to further its goals, as opposed to the goals of the
people that wrote the code. And on occasion, how about the
instance where Anthropic's Claude 4 model attempted to
blackmail the people it believed were trying to shut it down,
and then it would ignore and other ones would ignore shutdown
commands, even when told in advance that shut down commands
were coming, and went as far as to rewrite itself and preserve
itself on a different server to avoid the commands of the
humans in charge of it. Now, that might seem like the view of a
Luddite, but, I mean, I have got pages and pages of AI
deliberately disabling various developer-installed oversight
mechanisms, shutdown commands being ignored, replicating
itself, preserving itself without the user even knowing.
Now, let me be clear. I think that the United States of
America has to win the race and be preeminent in AI, but I also
think, while I sit on the Oversight Committee, that knowing
these things are occurring during tests, that us encouraging,
as a committee or as a Congress, the use of these systems in
places like nuclear security or national security without
having these pretty obvious things worked out is dangerous,
like, at a minimum. I cannot even believe we are having the
conversation. Like, how in the hell are we in here telling
everybody that we have got to incorporate this into the Federal
Government, state government, all levels of government,
including, like, National Defense, nuclear security, when we
know these things are occurring?
And I do not know which one of you to ask. You all seem to
be all for AI, and I get it is your vocation, your passion, et
cetera, but we got a duty here on this Committee and in this
Congress to make sure that the right thing happens. And you
know, humans, Americans, are supposed to be in charge of the
government, not some machine that learned from another machine
that picked up data that was polluted with Chinese propaganda
and is going to, like, push the button for us. Ms. Miller, you
seem interested. I got a minute.
Ms. Miller. Yes.
Mr. Perry. What do you got?
Ms. Miller. Well, this is a real concern, and GAO and other
standard-setting entities have been really pushing for
responsible AI frameworks. And this is one of those reasons
why, when we talk about AI, you can get a bunch of people in a
room that are like, go, go, go, lean in, you know, do more
adoption faster, faster. But it is so important to think about
these safeguards and how important they are. Responsible AI is
incredibly important when we are dealing with national security
information and American citizens' information, things that
could threaten the actual, you know, safety and security of
American citizens. And so, it is incredibly important that we
are slow and deliberate enough, and that is one of the reasons
why GAO and others have put out these frameworks. And it is
really important, I agree with you, as this Oversight
Committee----
Mr. Perry. So, are you saying that the frameworks deal
adequately with these things, or what am I missing?
Ms. Miller. Well, I am saying that if they are considered
and carefully rolled out, then those concerns will be at least
taken into consideration. I do think there is a lot of risk in
AI, and we should, I agree with you, consider where we want to
apply this and where we want to be----
Mr. Perry. So, ma'am, I have a little bit of time left, and
I just want to say this. I think, for me, it is beyond
consideration. Like, before we hand this thing the nuclear
codes, I do not want it to be considered. I want to know the
damn stuff is right, not polluted with Chinese Communist facts
or their alleged version of the facts and that the thing is not
going to avoid the humans and avoid, you know, detection and
all that stuff.
Madam Chair, I would like to enter into the record or
request that be entered into the record, Apollo Research's
report titled, ``Frontier Models are Capable of In-Context
Scheming,'' published on January 16, 2025, and I ask that
actually----
Ms. Mace. No objection.
Mr. Perry. Thank you. I yield.
Ms. Mace. Thank you. I will recognize Ms. Randall for 5
minutes.
Ms. Randall. Thank you, Madam Chair, and thank you to our
witnesses today for being here, maybe longer than you even
planned.
You know, Mr. Perry mentioned some of the important reasons
that AI use in the Federal Government is responsible for the
lives and livelihoods of so many more individuals than, you
know, one of us asking ChatGPT a question, you know, for our
own purposes. And we have seen some examples of potential
Federal Government uses of AI that have gone wrong. Mr.
Schneier, you know, Elon Musk and DOGE have tried to throw AI
at problems throughout the government with no regard to
Americans' privacy, safety, and security, no regard to the
27,000 Federal employees who I represent and do not seem to be
able to parse the difference between gender research and
specifying that female flies must be used in NIH research
studies. Can you discuss why the need for some of those higher
standards in, you know, a higher standard of AI use framework
in situations like that, and what some of the pitfalls are of
not providing a higher standard for AI use in government than
the private industry?
Mr. Schneier. I am not sure it is government versus private
industry. I mean, we probably might expect the same standards
as, like, banking or in university admissions, right? There are
going to be these instances where the decisions made are
important enough that we have rules and laws prescribing how
they can be made, and it is really a matter of making sure the
AI is suitable for the task, being able to audit what the AI is
doing, correct for mistakes. You know, humans make mistakes all
the time, and we as a society has had thousands of years
dealing with human mistakes, from double entry bookkeeping to
checklists of surgical equipment, make sure they are not left
in bodies. We know when a human makes a mistake how to deal
with it.
To the extent that AI makes human mistakes, we are covered,
but AI makes different sorts of mistakes, so a lot of our
systems for dealing with them just do not work. So, when I
think about an AI being used as something important like a
benefits administration, it is going to be how it is used. So,
I am going to make this up. Like, we will have the AI do the
work, but it is only allowed to say easy yeses. Cannot say no.
Just say yes. Everything it does is reviewed, and all the noes,
that anything that is hard, humans review. We can watch the AI
get better or worse. We might give more responsibility, but
again, with a lot of audit, it is really a matter of replacing
versus augmenting.
And one of the early questions was about AI in procurement.
So, I can think of two things where an AI is a negotiating
assistant, assisting a human in negotiating. Walmart does this
right now. Like, they use AIs because they have to buy a
gazillion things, to help with that, but the humans are in
charge, and also possibly using AI for auditing. There is a
really interesting work done on using AI to figure out which
slaughterhouses human ag inspectors should visit, right? There
are a lot out there. We can pick at random, or we can have some
algorithm. Again, the AI is not making the decision, the human
is, so really is really thinking very carefully about how the
AI and human work together where we can get the benefits of the
AI without losing control. I mean, Mr. Perry talked about that
is really important, but still getting the benefits of the
speed or the scale that AI provides.
Ms. Randall. Thank you. You know, I think that that is a
key difference that many of my neighbors and constituents are
talking about, use technological tools for more information,
for data gathering, for the ability to distill things, but when
we are talking about, say, the life and death of individuals
who depend on NIH-funded research or, you know, decisions that
have to do with termination of individuals because maybe they
wrote an email that says a word that people do not like in it,
you know, this becomes really dangerous and has real human
impact. So, putting in safeguards so that DOGE does not
continue to make mistakes that are unable to be quickly
corrected and that have real long-term impacts, I think, is
very important for us as an Oversight Committee to keep on top
of. Thank you, Madam Chair.
Ms. Mace. Thank you, and I will recognize Mr. Burlison for
5 minutes.
Mr. Burlison. Thank you, Madam Chair, for holding this very
critical hearing on the Federal Government in the age of
artificial intelligence. As we explore the AI's ability to
transform this economy and the potential that it has to enhance
Federal program efficiency, combat fraud, and deliver better
services to the American people, we must address a fundamental
challenge to this, and that is our Federal procurement system.
Our current system is riddled with barriers, lengthy
certification processes, complex reseller networks, and
multiyear contract vehicles that prevent agencies from
accessing innovations or AI solutions, particularly from
smaller companies. This procurement challenge not only hinders
efficiency, but also threatens America's global leadership in
AI.
In the last Congress, the 118th Congress, my bill, the FIT
Procurement Act, was reported favorably out of this Committee
and then successfully passed to the House with bipartisan
support last year. I will be introducing the FIT Procurement
Act again shortly, but let me point out that my bill directly
tackles our outdated Federal procurement system. This act
streamlines Federal acquisition to empower agencies to further
acquire cutting-edge commercial technologies like AI. It does
this by simply raising the simplified acquisition threshold
from $250,000 to $500,000, increasing the micro purchase
threshold from $10k to $25k, and allowing advanced payments for
subscription-based software, which is what we see so readily
today that the Federal Government is not able to take advantage
of. With unanimous bipartisan support in the last Congress,
this act is a proven commonsense reform to modernize government
save taxpayer dollars and reposition the United States as a
continued leader in the AI-driven future, and I look forward to
hearing from our witnesses about that. So, let me get begin.
Ms. Miller, in your testimony you discuss why agencies
should buy commercial technology rather than build custom code.
As someone who was an in-house custom coder, I could not agree
with you more. Could you follow up and elaborate on that?
Ms. Miller. Yes, sure. Well, the point I was making in my
testimony about that was that you create technical debt when
you are creating your code yourself, and because AI is moving
so quickly, frequently it will be obsolete soon after it is
built, and so it makes sense to contract with vendors who have
more updated tools. I would just add that the bill that you are
that you are referring to, given my own small companies'
challenges, getting access to small pilot projects and proofs
of concept, given those very low thresholds, this would be very
beneficial to small innovative technology companies.
Mr. Burlison. Thank you. I do not have to now ask my
follow-up questions. Is it true that AI can be used to detect
fraud in the grant process effectively?
Ms. Miller. Yes, there are so many great use cases for AI
in the fraud space when it comes to grants. We can scan through
information using open-source intelligence. AI models, large
language models can sort that information incredibly quickly,
within seconds, and spot anomalies humans would not be able to
spot. A lot of times what is happening today are grant
processors are pulling up Google and opening up different
screens and looking for things on different sites. Sometimes
they have access to open-source intelligence tools, but they
are still manually checking for that. AI makes that incredibly
quick.
Mr. Burlison. Thank you. Mr. Shah, you have highlighted the
Federal Government's procurement problem in their process.
Based on your direct experience, do you believe that changes to
the procurement process, allowing agencies to pilot products
like yours before going enterprise wide, would allow smaller
companies like you to better compete for contracts?
Mr. Shah. Yes, we believe that strongly. I think there are
a variety of ways to do that: one, creating more POC-type
programs. You know, a good example is the, I think it is called
the investment horizon pilot program that the Office of the CIO
of the Navy has implemented. But the challenge is, every agency
seems to have their own set of ways that they procure new
technology, so to the extent that we can help to normalize that
will help us also navigate. You know, for example, through the
FedRAMP process that I have described, of the $8.5 million,
$600,000 I spent on consultants to help me figure out the
process.
Mr. Burlison. Just to get certified, the program.
Mr. Shah. Yes. So, it is like, you know, I could have spent
that on maybe more innovation or other areas. So, I think that
if you have innovator programs or ways for us to start with
other agencies that are less security-focused will give the
help.
Mr. Burlison. Is it safe to say the Federal Government is a
pain-in-the-ass client?
Mr. Shah. It is a complex web, and if you ever want to do a
secret shopper with me 1 day and you can sort of watch how this
plays out, I would be very happy to show you what it looks
like.
Ms. Mace. Thank you.
Mr. Burlison. I yield back.
Ms. Mace. All right. I will now recognize Mr. Garcia for 5
minutes.
Ms. Garcia. Thank you, Madam Chairwoman. Thank you to our
witnesses for being here today.
I want to just start by noting that we know that certainly
AI can be an important tool for our Federal Government, for our
Federal workforce, that we always want to make sure that we are
centering those that are working in our government and uplift
the work that our Federal workers do every single day. But we
know that AI can navigate and help government become more
efficient and intuitive in many cases. I think we should be
honest about that. It can also help constituents get services
faster. I have seen this happen at the local level. I have seen
it happen, of course, and then happening at the state level and
here in Congress as well. We also know that many agencies just
need to move more quickly, be more efficient, and AI can be a
tool when you respect the Federal workforce and work with the
Federal workforce on how we make these implementations happen.
AI can also be used, of course, on issues around red tape. I
saw this happen, seeing happening in cities across the United
States, and we can analyze data better and faster. We can also
work to empower Federal workforce in ways that they can use AI
to help them do their jobs, and I think that is an important
piece of this.
It is also clear, though, that AI can be incredibly
dangerous, incredibly disruptive, and certainly without
guardrails can cause real harm to the American public and to
the work that we are all trying to do. Now, deployment, of
course, it is going to take investments in work, but I do have
some serious concerns about how AI is used, and it is
particularly in this Administration. And I want to talk about
one of them, which I think is actually really, really
important. So, this, of course, is Robert Kennedy, Jr., someone
who I consider to be an extreme anti-vaxxer. I believe for him
to be a conspiracy theorist, and certainly has no business
being the Secretary of Health and Human Services.
Now, take one step back. We just went through a pandemic
where 1.3 million Americans lost their lives, where businesses
were shut down across the country, where healthcare was brought
to the forefront of the public consciousness. And we know how
important getting real medical vaccine testing information is
to the American public, and right now we are dealing with
measles outbreaks in Texas and other places. Information needs
to be peer reviewed, fact checked, and done in a way that is
responsible. But here, of course we know that there was
recently a Make America Healthy report that was put out by RFK,
Jr. in HHS, and I am going to include in the article just this
quote: ``There were dozens of errors, including broken links,
wrong issue numbers and missing or incorrect authors. Some
studies were misstated to back up the report's conclusions, or,
more damningly, did not exist at all. At least seven of the
cited sources were entirely fictitious.'' What we do know from
many folks that have read this report is it appears that AI
played a role in developing a report about the public health of
Americans. This is something that we should be incredibly
concerned with.
And let us be clear. RFK, Jr. has a long history already of
dangerous, unscientific beliefs. He said 5G and Wi-Fi cause
brain damage. He said that you do not believe that HIV causes
AIDS. He said that water can make you transgender. These are
very concerning positions, and the fact that he puts out this
report, and HHS, which has been long respected as an important
Agency, puts out a report that is full of AI errors and AI-
generated content should concern every Member of Congress on
both sides of the aisle. This is incredibly devastating when
you, on top of that, look at all the cuts that he is making to
NIH and other health agencies across the United States, and so
this is, for me, very concerning, and it should be for all of
us.
Professor Schneier, is it fair to say that these kind of
errors in important government documents and health reports
that could be due to AI are concerning to us, and we should be
concerned about?
Mr. Schneier. They are concerning, but put the blame where
it belongs. Maybe the AI is not suitable for the task, or maybe
the human who used the AI did not, like, check the work, and
that would have been true if the human tasked an intern. I
mean, so, it is really a matter of what is the process by
which, whatever gives you your first draft, do you look at it
and make sure it is correct? We see the same problem with
attorneys submitting briefs into courts that have fake
precedence. I mean, yes, the AI made the mistake, but it is the
human who puts their name to it, who says this is correct. They
are the ones responsible, and I teach this stuff. I tell my
students to use AI, but you are responsible for plagiarism. You
are responsible for accuracy.
Ms. Garcia. I agree with you completely, and, sir, I also
spent 10 years in the classroom teaching as well, and----
Mr. Schneier. It is really weird now.
Ms. Garcia. And so, same, and I think that the idea that we
are now seeing so much AI-generated content produced in our
universities is a different topic, but let me just conclude
with this. I think your point is exactly right. AI can be a
tool, but there has to be the human element. The workforce has
to be a part of it. It has to be a responsible user of AI and
so that we do not end up with medical reports that are false.
And with that, I yield back. Thank you.
Ms. Mace. Thank you. I will now recognize Mr. Grothman for
5 minutes.
Mr. Grothman. Thank you. I will start with Ms. Miller. In
your testimony, you noted that the government loses more than a
half a million dollars to fraud every year. Given that
astonishing amount, what can the Federal Government learn from
the private sector because we know the private sector would not
put up with that, right?
Ms. Miller. Yes, it is a half a trillion, and, yes, it is
an astounding amount. What we can learn from the private sector
is, and I frequently say this when people ask me this question.
You know, a bank invests in fraud prevention technology because
they know that they are going to save more money with that
investment than they are going to spend on the tool. A private
sector company would never spend money on something it was not
going to have a better return on investment, but in the
government, because fraud is deceptive and we do not see it, we
do not invest in anything to try to stop it. And so, it is this
sort of happy sort of story that agency leaders can tell them
because if they do not go looking for fraud, they do not find
it, and then they do not have to invest in technology to
prevent it. And so, then you have GAO coming out and saying,
hey, we are actually losing half a trillion a year or more--I
believe it is more--to fraud, and the agencies are not being
held accountable.
Mr. Grothman. I will give you a kind of related question.
Obviously, the answer is, you know, we are going to build more
technology, and blah, blah, blah. We learned from DOGE a little
bit about the work ethic of some of the Federal employees. As
we spend more money somewhere to stop this fraud, do you think
it should better be spent on hiring new government employees or
maybe hiring out to somebody else who can try to detect this
fraud?
Ms. Miller. Well, I do not think that government employees
themselves have the tools to be able to find this fraud, so
whether you hired more government employees or whether you
contracted it out, I still think that there is such a lack of
focus on it. I do think that concerted investment, whether that
is in people or technology or both, will be needed if we really
want to move the needle on this.
Mr. Grothman. OK. Do you feel, though, that the bulk of the
work is going to have to be done by contracting out to somebody
or other?
Ms. Miller. Do I? Well, I mean, I think, generally
speaking, the Federal workforce does not have the skill sets
that they need to be able to find fraud, and we should be
skilling them up or we should be contracting it out, yes.
Mr. Grothman. OK. What is preventing the Federal Government
from implementing similar AI practices that they have in the
private sector?
Ms. Miller. A willingness to pretend that it is not
happening because agencies are more focused on getting money
out the door, and that is the goal, get the money out the door
quickly and then take a victory lap because you spent the
money, and not recognizing that there are people getting that
money who are, you know, foreign state actors and criminal
rings. But again, because it is underground, because it is not
seen, we do not talk about it. We talk about it in these
hearings, but we rarely talk about it, and the public is not
aware.
Mr. Grothman. Yes, I guess part of the lesson should be,
the less the government does, the better because they will
always screw it up, right? That is what one of the lessons
should be. Now we go to Mr. Thierer. On the first day of
office, President Trump rescinded Biden's executive order on
AI. Could you describe some of the damaging aspects of
President Biden's executive order?
Mr. Thierer. Well, unfortunately, that executive order
sought to do a lot of things that Congress should have been
doing, Congressman, and really did not wait for Congress to
take the lead on, on the questions of AI governance, more
broadly. Some of the best practices that were there for
agencies to follow, and some of the things that were
recommended in terms of AI officers were fine. And Mr.
President Trump has continued those practices in his executive
order that replaced the Biden executive order. But I think,
generally speaking, at a high level, I think the Biden
Administration executive order was just very fear-based and
treated AI as more like a curse to be avoided than an
opportunity to be embraced.
Mr. Grothman. OK. In your testimony, you cited a study that
stated utilizing AI systems would allow the Federal Government
to unleash a productivity windfall for the U.S. Government
worth up to $532 billion annually; not $531 billion, not $533
billion--$532 billion. Can you describe what kind of
productivity gains AI could unlock in the Federal Government
and where the big savings are going to come?
Mr. Thierer. Yes, they are really twofold. As that report
was noting, there are very specific types of targeted reform
and oversight efforts that AI can help with to improve
paperwork reduction, procedural issues, better administration
of government. And then there are more broad-based productivity
enhancements associated with AI itself and its use not only in
the public but private sector that can, ultimately, benefit
government administration.
Mr. Grothman. Could you tell me which agencies you think
would most benefit? I do not know how you came up with $532
billion, but is there any big agency? You grab $200 billion?
Mr. Thierer. Yes.
Mr. Grothman. What are the big----
Mr. Thierer. Yes. Well, a lot of Federal agencies could
benefit from more efficient operations. Obviously, Department
of Defense is one of them, but right now, important things
being done at State and DHS and others just approve paperwork
on things like immigration issues or paperwork issues, FOIA
requests and so on.
Mr. Grothman. Thank you. Very helpful.
Ms. Mace. Thank you. I will now recognize Mr. Khanna for 5
minutes.
Mr. Khanna. Thank you, Madam Chair. I appreciate the
testimony. Mr. Shah, of course you have been such a successful
entrepreneur and leader in Silicon Valley. I appreciate your
being here, so I will ask my first question to you. How do you
think AI can help, particularly government employees, do better
and be more effective in their work?
Mr. Shah. Yes, thank you. You know, the things that we are
seeing in the private sector today with large organizations
like Honeywell and like, you know, Northrop Grumman and the
likes, they are benefiting from AI by making their employees be
able to self-serve, get information that they are looking for
and take actions across a variety of systems. You know, on any
given day, an employee might have to swivel their chair 17
times to use different tools, and that becomes a very wasteful
portion of their day, sometimes many hours. And so, to the
extent that we can start to do that for Federal workers, for
all folks who work for this government, I think we are going to
see some benefits, benefits that will allow them to
troubleshoot, to have better IT, benefits that will accrue to
them from an H.R. perspective in terms of getting the
information they need, making the changes, the benefit
selections that go into that and also be able to work across
different systems without having to be an expert.
I think one of the things that you all have pointed out,
there are hundreds of enterprise systems inside the government,
and you may only touch it once a year, twice a year. And so, if
you can use AI as the overlay, kind of that front door, what
you will see is your staff and all the members of the Federal
Government be able to navigate these systems, which can be very
complex, very easily using the most intuitive interface in the
world, which is which is language and conversation. So, when it
comes to that, you know, we have seen companies and our, sort
of, world of private sector, be able to reclaim--a top five
pharmaceutical was able to reclaim 75,000 hours of employee
productivity by deploying this kind of AI across 44 languages.
We have had a Fortune 500 aerospace manufacturer reclaim 75
percent of its support budget to be able to bring in
technologies and leverage. I think that you are going to see
both happier employees and better outcomes while also making
the government more efficient.
Mr. Khanna. Appreciate that, and this is for anyone on the
panel. How would you address people's concerns about AI
displacing jobs, being something that will lead to layoffs or
people not being able to work?
Mr. Schneier. I am happy to start. I think it is going to
happen, and it is not going to be the jobs as we think. It is
going to be, like, low-level lawyers, and that is going to be
hard, I mean, because when you think about it, a lot of these
professions are apprentice professions, and where do the senior
partners come from if low-level lawyers or low-level doctors
are being replaced by AI. Something society has to think about,
whether there will be new jobs. In all previous revolutions,
there have been new jobs to replace old jobs. This might be
different. That is why you see a lot of talk about AGI, sorry,
about universal basic income, UBI. Like, is this going to be
different where we need to figure out how to give people the
ability to survive, and we do not know. We do not know a lot of
these things. So, those are very real fears, and I think we as
society should think about them before they happen. We are not
good at that, but we should.
Mr. Thierer. Congressman, just briefly, I would just point
out that there have been a lot of government reports on this
issue and a lot of efforts about government retraining, and so
on and so forth. It is very hard to have a crystal ball to
predict the jobs and skills that are needed in the future, but
I will just point out that just 10 years ago, everyone was
predicting a whole bunch of jobs that would be dislocated. The
most famous one, and No. 1, was going to be radiologists. And
just a couple weeks ago, the New York Times ran a piece to
basically say there are more radiologists than ever, and they
are utilizing AI to improve their jobs and free up time to do
better things. And that is the story of technological change in
sector after sector, is that we sometimes do not understand how
complex human machine interactions work and actually can change
jobs for the better and improve the employment output over
time.
Mr. Khanna. I agree with being a technology optimist. I
will say, though, that the unemployment rate for people between
the ages of 21 and 29 these days with a college degree is 15
percent. And so, the challenge, I think, is, as Mr. Schneier
pointed out, is how, especially at the entry level, for people
particularly with college degrees, how are we going to create
these jobs and how are they going to be able to use the tools
of AI to be effective? But I appreciate all of your thinking on
this and appreciate your being here for your testimony. Thank
you.
Ms. Mace. Thank you. I will now recognize Mr. Donalds for 5
minutes.
Mr. Donalds. Thank you, Madam Chair. I think it is actually
a good jumping off point from my colleague's comments about
potential concerns of displacement. I think that, you know, as
we looked at the evolutions of our economy over the last 150
years, we have always hit these inflection points when there
was a disruptive new technology or process for how large swaths
of our economy operated, and there is always a concern about
displacement. I think the difference now is that seeing the
evolutions of AI, which, quite frankly, are just starting, and
there are going to be things we are going to be talking about
in a decade we would not even be thinking about today in his
hearing.
I think it is more about making sure that kids going
through the K-12 system primarily, but then also through higher
education, are learning how to leverage AI and these tools to
actually become far more productive, far more innovative, and
then as a byproduct, having their labor be able to have a
different set of value in the economy to come. And I think that
it is an interesting thing, human beings and human psychology,
we always try to find a way to have purpose regardless of
whatever the dynamics are that are operating on the field of
play of our economy. One of the largest data center operators
in my home state of Florida, Equinix, was the first data center
to partner with an SMR provider, Oklo. These nuclear, small,
modular reactors can generate at least 500 megawatts of power
to our energy grid and to power data centers.
Mr. Bajraktari. I think I got that right. That is not
right. Close. My apologies. Can you share some ideas on how
Congress and the Trump Administration can accelerate the safe
deployment of clean, advanced nuclear energy and AI data
centers to support our growing energy and computing demands?
Mr. Bajraktari. Thank you so much for your question. I
think in the last couple of months, we have seen a really clear
linkage between, you know, advancing AI and AI models and
reliance on new modes of energy. I think my organization really
is focused on the development of fusion energy, for example,
because I think one of the pathways that now is showing real
great success in our country is the emergence of private sector
companies that are showing incredible successful models of how
to get to fusion energy. We also have the national labs that
have made major breakthroughs. You know, the Lawrence Livermore
National Ignition Facility in 2022 have made the first fusion
breakthrough. So, I think, in particular to your question,
fusion is one of the pathways that I think can enable the
energy demand that we will face going forward when it comes to
AI models, but that is not going to be enough alone. So, I
think we should pursue multiple pathways because we want to be
energy independent and we want to have the best models
available to the world.
Mr. Donalds. Mr. Thierer? Thierer? Thierer? I am not having
a good day. Mr. Thierer, with the establishment of the special
advisor for AI and crypto, what are some of the recommendations
or priorities related to artificial intelligence that you would
like to see them pursue?
Mr. Thierer. Well, one of them is what you just asked
about, Congressman, which is making sure that America has
energy independence and a greater diversity of energy sources
to power the algorithmic revolution, and this is something the
Trump Administration has taken steps to address with executive
orders and other things like that. That is crucial. Obviously,
a crucial part will also be the development of a national
framework for artificial intelligence policy, just as we had
for the internet before. We need to address this conflicting
patchwork of policies that is developing. That is something the
Administration and this Congress is obviously concerned with.
And then there are a variety of other things having to do with
national security matters and other investments in various
types of algorithmic and robotic systems that are really,
really crucial, among many, many other things that Congress and
Administration are trying to do.
Mr. Donalds. Last point, I will just throw this out to the
panel. I was sitting with my colleague, Mr. Perry. We were just
kind of theorizing about what type of advantages AI
advancements as well as, you know, I guess you could say
quantum computing, what that could actually do to modernize and
streamline governmental operations, not just Federal, but state
and local. Just in your own expert opinions, what do you think
the possibilities are in terms of how it would essentially help
to redesign, streamline, and bring greater amounts of
efficiency into the Federal Government? Like, where do you
think is a good place to start, I guess I would say.
Mr. Shah. Yes, I would say that, you know, if you think
about this new technology of Agentic AI, it allows for a new
kind of automation. We have been building automation for
several decades, but a lot of it was very brittle, and if
something slightly changed under the hood, all of a sudden, you
know, that automation would no longer work. With Agentic AI,
you can actually overcome a lot of that by having these very,
you know, smart, intelligent reasoners and systems that can
then go and do the levels of automation that we have not been
able to do before. And I think what that is going to do is
really elevate all of us to be able to work on less mundane
things and more sort of advanced thinking and creative ideas.
Mr. Donalds. All right. I am out of time, everybody. Thanks
so much. I yield back.
Ms. Mace. Thank you. I will now recognize Ms. Crockett for
5 minutes.
Ms. Crockett. Thank you so much, Madam Chair. You know it
is interesting because I do not know what all was taking place
thus far in this hearing, but I do recall when I left, we were
talking about Elon, and so I want to make sure that we do not
forget him. I know that some people want us to forget Elon and
act like he never existed, but we are not going to do that.
We are talking about AI today, and I can tell you what real
people in my district care about. I had a few town halls and I
talked to some seniors, and they were concerned about their
data and their information, and how it was being used. And they
wanted me to answer questions that I just did not have the
answers to about what has Elon done with their tax information,
their Social Security information, their health information,
banking information, information about their credit histories.
And instead of demanding answers, Republicans just blocked our
motion to subpoena him to testify before this Committee. And I
would ask at this point for unanimous consent to enter into the
record a picture of who all was in the room when we took the
vote, Madam Chair.
Ms. Mace. No objection.
Ms. Crockett. As well as since our clerk was ready and she
knew who was in the room, the actual record of who was in the
room at that time.
Ms. Mace. No objection.
Ms. Crockett. Thank you so much. Five months into the 119th
Congress, and no one from the Administration or DOGE has
appeared before this Committee to explain why they are invading
the privacy of Americans. And the irony of it all is during a
2023 hearing, I remember Chairman Comer stating that, ``We have
witnessed Big Tech autocrats wield their unchecked power to
promote their preferred political opinions,'' and now he and
Committee Republicans are here helping the world's richest man
to do just that. They are letting him spy on companies
competing for government grants, but worse, they are letting
him spy on you, the American people.
Look, we can all agree that incorporating emerging
technologies like artificial intelligence into government
operations can be effective. However, these technologies should
not be used to weaponize the government against the public. So,
I am going to ask Mr. Schneier really quickly. Is it true that
when it comes to any technologies, AI or otherwise, that they
can be a double-edged sword?
Mr. Schneier. I think of technology as power, and the more
the technology enhances your power, and whether your power is
for good or for bad, depends on you, depends on how you are
perceived. So, yes, technology has affordances. You know, we
know that modern technologies have kind of built for
surveillance, make surveillance easy. I mean, this is the most
complex, effective surveillance device we have ever invented,
and we all willingly put it in our pockets every morning. So,
yes, technologies can be used for both good and evil, but think
of the people behind it as the wielders.
Ms. Crockett. As the wielders. Yes, I absolutely agree, and
I appreciate your answer. That is my concern. My concern is
that these technologies will be used to weaponize the
government against the public because everything that we have
seen out of this particular Administration has been nothing
short but weaponization. So, let us talk about some of the
things that this Administration is doing.
They have announced they are going to use AI to scan the
social media accounts of students who disagree with their
policies. They have threatened nonprofits and universities--
they are threatening their tax-exempt status. They have
threatened journalists for reporting stories that are critical
of the President. They are literally in court arguing to end
birthright citizenship. And a lot of you all think only
immigrants are going to be affected by this, but this
Administration has already deported American citizens. In fact,
in an interview with Time Magazine, Trump said, ``I would love
to do that. We are looking into it.''
Madam Chair, I would ask unanimous consent to enter into
the record an article published by ABC News titled, ``Trump
Wants to Send U.S. Citizens to Foreign Prisons, Legal Experts
Say.''
Ms. Mace. No objection.
Ms. Crockett. Thank you. It will not be long before more
Americans are targets of this Administration. This
Administration has refused to follow the law, and the
Republicans are failing to do actual oversight. We do not know
what they are going to do with their health records and Social
Security information. In fact, we do not know what they are
doing. I do not think anybody does, except we know that they
are breaking the law. And it is frustrating to me because I am
somebody that actually believes in science. I believe that we
should be investing in our young people. I believe that we
should not be taking research dollars away. I believe that we
should be trying to come up with new technologies and that we
should be leading the world. But unfortunately, I cannot trust
these guys to do good things because, as it has already been
said by Mr. Schneider, it is about who it is that is on the
side that is asking for this to be done. And frankly, this
feels like some bad movie that we are all in and that the
villains are doing bad. So, with that, I will yield.
Ms. Mace. Thank you. I will now recognize Mr. Burchett for
5 minutes.
Mr. Burchett. Thank you, Chairlady. As the 435th most
powerful Member of Congress, I have the honor of asking
questions that have been asked ad nauseam. But when I ask them,
answer as if this was the first time you said, and the proper
response is, ``Congressman Burchett, that is possibly the most
incredibly intelligent question that has been asked of me,'' so
at least my clips will show that and the 12 people that are
watching this thing are at home. They are probably not in my
district anyway. They are probably in Marjorie's district.
Anyway. Mr. Thierer? I get that name right?
Mr. Thierer. Right.
Mr. Burchett. All right. All right. The majority of East
Tennesseans and, frankly, Tennesseans and Americans, are
worried about their privacy being violated by government
artificial intelligence programs. What do you think Congress
can do to protect our privacy? And my biggest fear in all this
is Congress. I mean, some of these cats up here still have
eight-track tape players in their 1972 AMC Gremlins, and they
are going to try to regulate AI. They cannot hardly spell
``AI'' and now we are going to regulate it, so.
Mr. Thierer. Well, Congressman, thank you for that
excellent question, and as someone who used to grow up in a
Gremlin with an eight-track player, I feel great affinity for
this question. So, I think that, obviously, we need to make
sure that privacy is protected in the digital age, and this
Congress has struggled with how to get that done. Obviously, we
have had a multiyear, multi-decade debate about baseline
privacy legislation, and a great deal of bipartisan consensus
exists, that that should be a priority before you get to a lot
of the other questions about how to regulate all the various
specific concerns about AI. A lot of them go back to
fundamental data privacy, data security issues. And so, I
think, you know, let us not put the cart before the horse. Let
us think about getting that done first. I know that is
something Congress is turning to next.
Mr. Burchett. All right. Also regards to China, how do we
maintain our dominance in the AI sector? Any of you all can
answer that. I will ask Mr. Thierer first, if that is OK, on
this.
Mr. Thierer. Well, a lot of it comes down to making sure
that America diffuses first. We have wonderful products. We
are, essentially, winning one part of the AI war but losing
another. We are winning the part in terms of developing really
sophisticated, high-powered frontier models and other systems
and applications, but the reality is, is that China may be
winning the diffusion side of things, and this is why it is
very, very essential that we not miss out on making sure we
have partnerships and deals, multilateral, mini-lateral
approaches. The Administration has been trying to do this,
obviously, recently, but this Congress should obviously make
sure that we do everything we can to make sure American
technology and products lead the world because it is not just
about competitive advantage, it is about our values. Our values
are embedded in our products, and, of course, the Chinese
Communist Party values are embedded in their products of, you
know, control, surveillance, and lack of human rights.
Mr. Bajraktari. Yes, sir. We got to do three things. No. 1
is we have to continue to out-innovate and out-maneuver China.
Our platforms should be the dominant global platforms. No. 2 is
we have to work with our allies and partners to spread our
platforms to the global south and rest of the world. And three,
we got to find ways to slow down China and make it impossible
for them to deploy their capabilities and to use their
capabilities against us.
Mr. Burchett. Can you give me that first one again? I can
tell by your accent you are probably not from East Tennessee.
Mr. Bajraktari. I am not, sir.
Mr. Burchett. I am the only person in this room that is not
speaking with an accent right now, so if you could say that
first one again, brother.
Mr. Bajraktari. The first one, sir, was that we have to
make sure that we out-innovate and out-maneuver China.
Mr. Burchett. Right.
Mr. Shah. I will just add, I agree with all the other
panelists here and witnesses. I think that this is an infinite
race, and the technology that is being discussed here today,
AI, can essentially be wielded and constructed by any sort of
group of smart people, and the planet grows smart people
everywhere every day. And I think that to the extent that we
can be the ones that others follow, that we can be the ones
whose software, whose hardware, whose foundation models others
use, we will have the sort of dominant leadership position of
how this whole planet evolves.
Mr. Burchett. Ma'am?
Ms. Miller. Well, I would say that China is doing a very
good job of figuring out how to defeat some of our security
vulnerabilities. And I think that is one area that we have a
ways to go is to protect the infrastructure because North Korea
and China and others are getting better and better at
infiltrating and exploiting our vulnerabilities.
Mr. Burchett. Sir?
Mr. Thierer. I am going to be contrary here. I think the
U.S.-China arms race is an overblown metaphor. This is not the
1960s. This is not the cold war. Science happens globally. The
transformer model invented by Google, it was published,
everyone uses. DeepSeek, China comes up with this model. It is
public. Everyone is innovating based on it. Ideas are flowing
cross-border. The dominance is more companies than technology,
so do not think about regulating the hardware, the software,
the ideas. We cannot do that, and this is not the kind of world
where that happens. It really is the companies and what they
are doing. It could be U.S. companies, European companies,
Chinese companies, other companies.
Mr. Burchett. All right. Thank you all. Chairlady, I yield
back nothing.
Ms. Mace. OK. Thank you. I will now recognize Mr.
Subramanyam for 5 minutes.
Mr. Subramanyam. Thank you, Madam Chair. I appreciate us
having this hearing. I served in the second term of the Obama
Administration as a technology advisor in the White House. I
worked a lot with a lot of great groups who are dedicated to
modernizing government, making it more efficient, bringing AI
to government. And one of them was the U.S. Digital Service,
which has now become the U.S. DOGE service. Another one was
18F, which has been completely set apart, completely gotten rid
of it, and I will talk about them in a second. But one of the
concerns I had about DOGE was that there is this notion coming
in that they would come in and fire all these Federal civil
servants and that AI would replace them, and that we would save
trillions of dollars as a result, and so that is wrong for a
bunch of reasons. One, even if you fire every Federal civil
servant, you are actually not saving that much money in
comparison to the deficit and the debt, but two, the Federal
workers are so essential when it comes to helping utilize these
technologies, and especially the technologists that we have
hired in recent years.
And so, we have heard a lot about this already from people
testifying and even from Members here, that it is not just
about having AI, it is about having to deploy AI in a
responsible way and in a manner in which we are not letting AI
make mistakes. I will give you an example. In Michigan about,
let us say, 10 years ago now, they deployed AI to try to find
fraud, and they ended up finding a lot of fraud where there was
not any fraud. And a lot of people got investigated or went
bankrupt trying to defend themselves from getting sued, and
then they realized that the AI had made a big mistake. And
there are so many other examples of AI being deployed in an
irresponsible way because we simply were not pairing them with
a human element that could help sort of edit and deploy it
responsibly.
And so, 18F was a great example of this. They were actually
doing a lot of great work to modernize our Federal Government
and bring AI to the Federal Government, and going agency by
agency and helping us procure great tools to do that, and
instead, they were all fired, basically. DOGE came in and fired
the entire staff at 18F, and now we are trying to figure out
how to deploy AI in government, but we are chasing away the
very technologists who can come to the Federal Government and
help us do that.
And so, if you want AI and modernization in government, we
need the best and brightest to be coming to government, but
instead, what we have is a brain drain on our Federal
Government because we are trying, to quote the OMB Director,
``put Federal workers in trauma.'' Who wants to come and work
for the Federal Government if the people running the Federal
Government wants to put you in trauma? Nobody. So, we are not
going to attract the best and brightest technologists to help
deploy AI, and we are certainly not going to have what we need
to have pairing up with AI in order to deploy responsibly.
And then there are a lot of other things going on that I
think are bipartisan, like, you know, making sure we update our
legacy systems. I was just working with air traffic control,
and I think their software is Windows 95. I have talked to many
others in government who do technology, they still use COBOL,
right? They still use really antiquated technology and coding.
So, how are we supposed to deploy modern AI tools when we have
these legacy systems, right? The other thing is, how are we
supposed to recruit and retain great people to help transition
and modernize our government when we are chasing them all away?
And finally, this AI human collaboration. It is already
running into problems in the private sector, but it has to be
even better in the Federal Government because like in that
example with Michigan where people's lives were ruined because
they deployed an AI technology irresponsibly, here, if you mess
with an AI tool and it does not get people's checks on time for
Social Security or Medicare or Medicaid, then people are going
to go without healthcare. People are going to go without being
able to pay their utility bills. It is going to have a terrible
effect on the American people, and that is what this is really
about.
It is not about just deploying AI so you can say you
deployed AI. It is about making sure we are deploying
technology for the benefit of the American people, and so I
hope we can reverse course. I hope we can work in a bipartisan
way to make sure that we are taking advantage of the advances
in AI but doing so in a responsible way. I yield back.
Ms. Mace. Thank you, and I will now recognize myself for 5
minutes. I want to thank our participants and panelists and
witnesses today for being here.
One of my concerns, coming from a technology background,
like, I learned COBOL back in 1999, and a lot of our systems
today, they are on legacy systems. We waste a lot of money on
that, and then with the advent of AI and how fast technology is
moving, you know, the Federal Government has real challenges
with moving forward in a quick way and competing against our
adversaries. So, this is my first question to all the panelists
this afternoon, is if you could just do one thing that could
advance the Federal Government in any way, any agency, any
capability, if you could just do one thing today to help us
utilize AI and technology to move the ball forward, what is
that? What is the one thing you would do? Does not have to be
something big. It could be a small part that makes a big
difference. But what is that one thing you would do right now,
if we could do it tomorrow, snap our fingers? What is it?
Mr. Bajraktari. Ma'am, I would say enable adoption, any
models at any jobs in Federal Government, for memo writing all
the way to execution, logistics, back office responsibilities,
anything, but without enabling, you know, humans to start
adopting these technologies, I do not think we will get to the
next stage where we need to be as a country.
Ms. Mace. Mr. Shah?
Mr. Shah. Most of the innovation happening in AI is
happening at small startups, and I think that, to the extent
that you can look at a team of maybe 50 people and see how long
it takes for them to get their product discovered and utilized
by Federal Agencies, will determine how fast we can move, and
to the extent that we can solve that, I think it would make a
big difference.
Ms. Mace. Ms. Miller?
Ms. Miller. Pick a high-value process that can be automated
using agents and deploy them. So, for an example, continuing
disability review in Social Security, we have seen a 9 to 1 ROI
on that process of reviewing continuing disabilities that is
using people. That ROI would be at least probably doubled if we
used AI agents, and we would save a lot of money in a very,
very high-value area. So, pick a really high-value area, deploy
agents, and measure the difference so that you can use that
measurement to inform other investments.
Mr. Thierer. Chairwoman, you are going to like my answer
because I am going to say we need the AI Training Extension Act
to be moved.
Ms. Mace. I love that.
Mr. Thierer. We need better literacy and skills government-
wide. We have a lot of smart people working in government, and
sometimes it is just a matter of getting them moving in the
right direction and the right training to get the job done.
Ms. Mace. Mr. Schneier?
Ms. Schneier. I want to see a model that was not built by a
corporation. I want to see a public AI, that democratic
processes are different, they have different ordinances, they
have different requirements, and to get a model that will be
trustworthy for the U.S. Government. I think it needs to be
U.S. Government built. I do not think it can be corporate
built. So, funding and research into a public AI will give us a
different sort of AI. I think it is going to be interesting.
Ms. Mace. It is interesting. You know, we have all this
technology at our fingertips, but, like, going back to Ms.
Miller's comments about fraud and abuse, we have this
technology, but it is almost like we write these checks and
then we cross-check it later. So, there is a System Called
DoNotPay. It is called continuous monitoring. It still gives
agencies match results the next business day, for example, and
in 2025 that is dial up speed, right, when you can have it
within seconds. And taxpayers are footing the bill for a 24-
hour lag when there are fintech companies that are out there
that can flag fraud in milliseconds, like that technology
exists. The March 20 executive order would force agencies to
share all unclassified records with authorized officials, yet
most agencies are still swapping spreadsheets as opposed to
using APIs and technology that is, again, readily at their
fingertips.
The Secretary of Labor now has statutory access to state
unemployment insurance files, but those fees are not wired into
DoNotPay as of today, right now, and it seems like we are still
paying benefits first and then cross-checking later, which,
again, makes it rampant for fraud that we have had year over
year over year. And then the IRS individual master file is
written in COBOL in the 1960s, and it still drives every tax
refund. Like, I cannot even wrap my brain around. I mean, if I
saw a line of COBOL right now, I would probably throw up, and
then to know one of the most important agencies for American
taxpayers, you can get your money back from the government, it
was written in COBOL from the 1960s. So, how can an algorithm
catch refund fraud in real time when the source of the data
lives in a mainframe computer?
And then the last thing I want to say is that GAO says that
the Federal Government blows 80 percent of $100 billion IT
budget just keeping 1970s systems on life support. I would love
to know after this, and I will send a question follow-up, but
if there was one legacy IT system you could cancel today, what
would it be? And I do not have time, but that is a follow-up
question I would love to ask all of you. What is the one legacy
IT system you would just cancel tomorrow, if you could, to make
our systems better?
Ms. Mace. Thank you so much, and I yield back. I will now
yield to Ms. Lee for 5 minutes.
Ms. Lee. Thank you, Madam Chair. As we talk about AI, I do
not think we can afford to ignore what is going on right now in
our Federal Government. So, the latest out, we know that Musk
and the DOGE children went into every department and accessed
their datasets. Initially, they were digging around and using
AI to cut government funding, but it has become clear now that
the goal has been to amass and merge these different pots of
data into one data base.
Trump has now tapped tech billionaire Peter Thiel's
Palantir to head up that master data base. We are not talking
about just inter-government memos here in the data base. We are
talking about Americans' Social Security numbers, home
addresses, credit card numbers, medical diagnoses, and more.
All that information would just be in one place, and then they
are going to unleash opaque algorithms that no one understands
but them. So, they will be able to do whatever they want, and
we are supposed to just trust that they are working within our
best interest. And we probably will not conduct much oversight
because we have not been able to bring Elon Musk or anybody
associated with DOGE into the Committee on Oversight and
Government Accountability [sic].
So, Mr. Schneier, looking at the immediate risks of this,
is there a concern that having all of this information in one
place can make it easier to hack or steal?
Mr. Schneier. Well, that is the fear, that we keep this
data separate for partially security reasons. I mean, there is
a reason why that when data shared between agencies are done
under certain rules and why there are security protections
around the data. So, it is the data being consolidated that is
a risk. Most of the data leaving the government data bases
where they were, that is a risk. They are being moved into
private companies. They are being used by algorithms, well, you
said no one understands, but them. It is actually worse than
that. No one understands, including them. These algorithms are
opaque, and we do not know the quality of the data as data is
combined.
Talking about COBOL, a lot of that is how the data is
presented and making sure that it is the same across. In my
initial statement, I talked about that story about 150-year-
olds getting Social Security. That was a problem with
understanding how the COBOL was not decoded.
Ms. Lee. Certainly.
Mr. Schneier. And if you do not know that, you are going to
have bad inferences that no one is going to understand and no
one is going to check.
Ms. Lee. Thank you. So, beyond the concerns with the data
all being in one place, one less secure place, we also need to
be concerned about with what Trump and these tech billionaires
are going to do with that data. Just very quickly, and ``yes''
or ``noes.'' Mr. Schneier, could an AI system theoretically be
used to decide if someone qualifies for Medicaid?
Mr. Schneier. Certainly.
Ms. Lee. How about used to make decisions around veterans'
benefits?
Mr. Schneier. Yes.
Ms. Lee. Could AI be used to make hiring decisions?
Mr. Schneier. It can make any yes or no question you want.
Ms. Lee. Thank you for entertaining that. Trump and his
Administration can take all this collected data and use it to
profile people, to make decisions about them, to deny them
benefits, and we know from his Twitter rants that he is
incapable of being unbiased. Then they are trying to prevent
any state-passed guardrails on AI through the provision in the
reconciliation bill that you guys unseriously called the Big
Beautiful Bill. That provision would ban any state regulation
of AI for 10 years.
Our government cannot just play loose and fast when it
comes to people's personal data and lives. We have already seen
the consequences of improperly deployed AI. Michigan had to
refund $21 million to residents who were wrongfully accused of
unemployment fraud from their AI system. A tenant screening
system called SafeRent was accused of violating the Fair
Housing Act because their AI algorithm disproportionately
scored Black and Hispanic renters lower than White applicants.
Most of this technology is still so new, it needs time to
be vetted and improved. Look at what happens when AI is used
just in government reports. The Make America Healthy Again
reports cited to multiple studies that did not exist. Can you
imagine if ChatGPT was deciding if you qualify for housing or
not? That is not to say that all AI is bad, but it is dangerous
to work quickly. When you move fast and break things in the
government, people get hurt.
Carnegie Mellon University, which is in my district, has a
Center for AI Standards and Innovation, and has been working
with the National Institute of Standards and Technology to make
a methodical approach to assessing the benefits and risks of AI
models. Thinking about the ethics of AI, work like theirs is
vital to ensuring that AI programs are used safely, accurately,
and without bias to the best of our ability, recognizing that a
society without biases is far off. But that is the approach the
Trump Administration should be taking, not packaging up all of
our personal data and handing it over to unvetted tech
billionaires. I thank you all for your time, and I yield back.
Ms. Greene. [Presiding.] The gentlelady yields. I now
recognize myself for 5 minutes.
We just passed the One Big Beautiful Bill through the House
of Representatives, and now it has gone on to the Senate where
they will be making changes to the bill. If there are any
changes made to the One Big Beautiful Bill, it comes back to
the House for a vote and we get a second bite at the apple.
Now, while this bill was going through committees and being
discussed in the House of Representatives, no one on either
side of the aisle that I know of, Republican or Democrat,
brought up this particular clause on one single page in an over
1,000-page bill, and I want to bring this up because I think AI
is incredibly important. I support AI and many different
faculties. However, I think that at this time, as our
generation is very much responsible, not only here in Congress,
but leaders in tech industry and leaders in states and all
around the world have an incredible responsibility of the
future and development, regulation and laws of AI.
It is such an important responsibility, it literally
affects our children and grandchildren's future. So, I take
this responsibility incredibly sincere, so much so that this
past week I have come out in full transparency and said when I
voted for the One Big Beautiful Bill, I did not know about this
clause. I thought I was voting on taxes, energy, and border
security, and this clause right here says a moratorium. In
general, ``Except as provided in Paragraph 2, no state or
political subdivision thereof may enforce during a 10-year
period, beginning on the date of the enactment of this act, any
law or regulation of that state or a political subdivision
thereof, limiting, restricting. or otherwise regulating
artificial intelligence models, artificial intelligence
systems, or automated decision systems entered into interstate
commerce.''
That is a comprehensive 10-year moratorium. Actually, what
that is, is it is a pause for 10 years in federalism. This
right here, if passed into law, will tell states they cannot
regulate or make laws regarding AI. That means the Federal
Government is the only governing body that can regulate or make
laws for AI in the United States of America. And I would like
to ask each of you, do you support federalism? Yes or no. I
will start with you. Do you support federalism? Yes or no?
State rights. It is a yes-or-no question.
Mr. Bajraktari. I do not think I am able to answer this
question based on my background, to be honest with you.
Ms. Greene. OK. Then I take that as a no. Mr. Shah, do you
support federalism?
Mr. Shah. I do support state rights, and just from a small
company standpoint, I would say whatever we can do to make it
easier for a young company to navigate the 50 states will be
very important, because otherwise----
Ms. Green. Well, either you support federalism or you do
not, one or the other. It is a yes or no.
Mr. Shah. Yes.
Mr. Greene. OK. Ms. Miller, do you?
Ms. Miller. Yes.
Ms. Greene. Yes, Mr. Thierer?
Mr. Thierer. Yes. In line with the White House and with
Speaker Johnson, I agree that federalism is important, but
federalism is a two-sided coin of both states' rights and
interstate commerce that we need to protect.
Ms. Greene. The Constitution is clear on federalism. It is
one side, state rights or none. Mr. Schneier?
Mr. Schneier. I think that provision is nutty.
Ms. Greene. OK. So, do you support federalism? Yes or no.
Mr. Scheiner. In general, yes.
Mr. Greene. Oh, in general, so that is vague. I am not
sure.
Mr. Scheiner. It is vague, but that provision is nutty.
Ms. Greene. OK. So, I still did not get an answer clearly
on federalism. Can you predict the future of AI in 1 year, 5
years, or 10 years? I am sorry. I apologize.
Mr. Bajraktari. Yes. No, no problem. No, no problem. No, I
do not think anybody can predict that. I think every model has
become bigger and better and more transformative. I think----
Ms. Greene. Right, so the answer is no. I am short on time.
Mr. Shah, no or yes?
Mr. Shah. I think it is hard to predict 5 years.
Ms. Greene. Yes or no. Can you predict that period?
Mr. Shah. No.
Ms. Greene. OK. Ms. Miller, can you predict AI?
Ms. Miller. No.
Ms. Greene. Mr. Thierer?
Mr. Thierer. No.
Ms. Greene. No? Mr. Schneier?
Mr. Schneier. Not a chance.
Ms. Greene. No, you cannot. In the state of Georgia, jobs
are extremely important, and AI is, whether we like it or not,
and it is helpful in many ways, and it is not helpful, will
replace jobs. I have a manufacturing district, and the
manufacturing companies are important, but so are the people's
jobs. And if our state cannot regulate or make laws to protect
people's jobs, people are going to go hungry. They are not
going to have paychecks. They are not going to put roofs over
their families' heads. And I think this is such a serious
issue, that we need to look at this through--I will state it
very clearly, I am pro humanity. I am not pro transhumanity.
And when it comes to AI and regulation, when we get to vote on
this bill again, I will be voting no because of this clause,
and we will be working on this further. Thank you so much for
being here today. I now yield to Ms. Simon.
Ms. Simon. Thank you, Madam Chair. Thank you all for coming
today and sitting in this very chilly room. I think we have all
learned so much. I come from the Bay Area, which is, as you all
know, the technological gold mine for the artificial
intelligence community. I am so proud of that. In March,
actually, I participated, I actually convened a table with AI
innovators and leaders in my congressional district on how AI
is and will continue to enable access to high-quality
healthcare, data sharing, and, of course, opportunities for
physicians and folks in the industry to expand access to the
many wonderful folks who need it and deserve it.
We already know that AI is advancing lifesaving, early
detection for cancer, Alzheimer's, heart disease, and I can go
on and on, and can save clinicians time in charting and note-
taking and better spent on patients and actual working with
families and folks who need them. This technology helps
patients self-manage their diseases with suggested resources
and reminders to take medications. Again, as you all know, I
can go on and on. And as somebody, myself, with a visual
impairment, I was born legally blind. I cannot tell you how
artificial intelligence has led the disabled community to be
able to live with more dependence and more self-determination,
and it has exceeded, in this moment, our wildest expectations.
The guardrails are needed, absolutely regulation is needed, but
again, so many of us from impacted communities are so excited
about what this technology allows for us to do.
The Federal Government has this responsibility and duty not
only to set a standard around data privacy and security, but to
follow the best practices. I have a lot written, but time is
short, so I just actually have a couple of questions that I
will just go straight to, and again, I want to thank you all
today, and I am sure all of our offices will be working with
not only you all, but leaders in the industry moving forward
forever. So, my question is actually for Mr. Schneier.
You know, I have met with, as I mentioned, AI companies in
my district who manage very sensitive health data, very
sensitive health information with great care and sensibility
and sensitivity, and these are the companies that I have met
with. They only use de-identified patient information, data
sharing, and encryption, and otherwise control access. Can you
describe how our current Administration's actions differ from
the industry standards and best practices around data privacy
and security?
Mr. Schneier. Oh, in so many ways. So, we have seen
examples of DOGE dismantling security protections inside
organizations. We have seen them taking data out of secure
enclaves and taking it off premises, and we do not know where.
We have seen them combining data in ways that are contrary to
law. If you think about it, healthcare is controlled by HIPAA.
There is a regulation about health care privacy. De-identified
data is a science. The census has a whole program on how to de-
identify data to be used by researchers without revealing
private information. So, we can do this if we choose to. DOGE
chose not to.
Ms. Simon. Well, I have a little bit more time. Continue.
Mr. Schneier. All right. So, I would want to add, moving it
into AI models has its own risks, and it is several risks. It
is the risks of the model being trained on data that is private
and then could make it public. We see researchers being able to
pull private data out of a model, so without safeguards, the
training process fails in privacy. Then the data will be used
as an input to whatever the model is to produce decisions, and
that also can be a violation of privacy. So, all through these
processes, we can build and we can build safeguards. I think
that is your point.
Ms. Simon. Yes.
Mr. Schneier. We just have to do it, and there is a reason
why you do this slowly and deliberately within organizations
and not just pull the data out.
Ms. Simon. Thank you, sir, and actually, this is a question
that, as you are talking, just derived in my mind. Anyone on
the panel want to talk about the innovations in healthcare and
AI, particularly around diagnostics? I only have about 30
seconds, but I think it is an exciting innovation and folks
will live longer, I believe, from what I have heard. Anyone
want to jump in?
Mr. Thierer. Yes. Congresswoman, thank you for your
leadership on this front. I will submit for the record a series
of reports that my organization has put together on AI and
public health where we cite specific ways that AI is being
utilized in drug discovery, medical devices, and so on and so
forth. I will just mention that we have had a 50-year war going
on cancer and we have not made as much progress as we need to,
but last year, 608,000 people died of cancer, and a lot of
these things can be addressed now with the power of artificial
intelligence.
Ms. Simon. I appreciate that, and, Madam Chair, thank you
for your time and allowing me this time. My husband died of
cancer. I know that he could have lived longer if we had these
technologies. We got 2 years with him. Again, that was 10 years
ago. Changed our lives forever. Thank you for your work, your
innovation, and clearly, we need safeguards, but the future is
bright. Appreciate you all.
Ms. Greene. The gentlelady's time has expired. I now
recognize Mr. Timmons from South Carolina.
Mr. Timmons. Thank you, Madam Chair. The greatest national
security threat we face is debt. We have $37 trillion in debt.
We have a $1.8 trillion annual deficit. The work this
Administration has done over the last few months has been
designed to try to find waste, fraud, and abuse to help
mitigate that. They are also working to help grow the economy
through tax reform, through renegotiating trade agreements,
through deregulation, through streamlining permitting. The list
goes on and on. But I think that AI has a huge opportunity to
grow the work that DOGE has done as it relates to seeking out
waste, fraud, and abuse.
So, the main thing that they did early on was to get all
the government agencies to communicate using the same language,
to code the same, to then be able to create transparency and
track data from all the different government agencies. And that
effort was really the beginning because now that everything
goes through Treasury, now that we can track payment systems,
we now have to layer on technology, and some of that is fairly
simple.
For example, I have introduced the TABS Act, Timely
Accurate Benefits, and that what that act does is it uses a
pilot program that was done in Missouri where they, instead of
using antiquated ways of confirming income eligibility for
Medicaid, they use third-party website to verify eligibility
immediately. And then the second you are no longer eligible,
you are removed from whatever benefit it is, and so that
resulted in a 17-percent reduction.
So, the way this technology works is you say, all right, I
am on hard times. I want to get whatever government benefit,
and they say, all right, well, you got to log into your bank
account. And then when you log into your bank account, it says,
OK, you have a Venmo, you have a Cash App, you have a Zelle,
you have to log into that, too, and then immediately they say
you are eligible or you are not eligible, and so you are
actually getting benefits faster than normal. And this has a
huge potential to save money and that 17 percent of Medicaid is
about $150 billion if you do it over the entire Federal
Government.
And, you know, one thing that the One Big Beautiful Bill
did was instead of confirming eligibility once a year, which I
think is ridiculous, we are now confirming eligibility twice a
year. Well, why don't we confirm eligibility every day using
technology? This is not hard. The technology exists. All we
have to do is implement it. So, that is one way that we can
further find waste, fraud, and abuse because I think we can all
agree that if you are not eligible because you make too much
money or you are not eligible because, and this is the next
step, because you are not the person who you say you are, then
you should not get government benefits.
So, I am working on TABS 2, which is using technology to
confirm you are who you say you are, and this is just using
billions of publicly available pieces of data that, you know,
for example, if you have an email address that was created in
the last year and you are 45 years old, you are probably not
who you say you are. If you have three different phone numbers,
that is not normal. So, they are able to use technology,
publicly available information, to confirm an individual
requesting benefits is who they say they are. And honestly,
that would not only help means-tested programs, it would help
any government program. If you are saying you are who you say
you are for Social Security, I mean, that is not means-tested,
but we could use it to seek out waste, fraud, and abuse.
So, those are two things, two ways we can use technology to
reserve government benefits for American citizens that deserve
them. How can we grow that effort? How can we build on that
effort using AI? Mr. Bajraktari?
Mr. Bajraktari. Thank you. Sir, I think those two examples
indicate the potential that we have in AI adoption for
government services. I think we have got to do three things.
The underlying issue that we have here is that we need to
upgrade or modernize our IT infrastructure because you cannot
deploy AI if you do not have the necessary infrastructure
underneath that, and so to do that, we need to do three things.
No. 1 is we need to upgrade our platforms, No. 2 is we have to
update our policies, obviously, and No. 3 is we have to work on
retraining and re-skilling our people.
Mr. Timmons. Ms. Miller, what are your thoughts?
Ms. Miller. There is a lot we can do in this area. I mean,
this is what the software company that I created just does
this. And mining open-source intelligence to identify whether
somebody is who they say they are, is the lowest hanging fruit
and the best use cases we can do in the fraud space for
government programs, and not just means-tested programs, every
grant program. We put a trillion dollars in Federal money out
every year to states for grants, and we have very little
understanding of where that money is going, but we can use AI
and open-source intelligence to very quickly identify if it is
a shell company or even an organized crime ring that is
actually getting that money.
Mr. Timmons. But on the organized crime front, a lot of
these involve changing addresses once they request benefits and
things of that nature, and AI could easily say, oh, it is
weird. There is a P.O. box in Eastern Europe that gets 43
checks a day. That is weird.
Ms. Miller. A hundred percent. There is geolocation. There
are all kinds of behavioral biometrics that AI can flag. You
can use, you know, probably dozens of different data inputs to
identify fraud today using AI.
Mr. Timmons. And this is not a partisan issue. I think
everyone can agree that only people that deserve benefits, that
are entitled to benefits, should get them. So, again, I think
using technology to make sure that we are being wise with our
taxpayer dollars is something that is a no-brainer, and I look
forward to working with my colleagues across the aisle to do
just that. Thank you, Madam Chair. I yield back.
Ms. Greene. The gentleman yields. I now recognize Mr. Min
from California.
Mr. Min. Thank you, Madam Chair. I appreciate the
opportunity to discuss artificial intelligence in the Federal
Government. Of course, AI has been around for a while, but it
is recently hit, I think we all agree, a tipping point of
exponential and profound growth. Innovation and advancement are
now being measured in days or even seconds or minutes as
opposed to years, and this rapid development of AI is poised to
radically transform our society and our economy. And that is
going to create a lot of great benefits, I believe, but also a
lot of challenges, including to our workforce, our national
security, data privacy, and a whole host of other issues, and
these benefits and challenges of AI are also important to
consider as we contemplate bringing more AI into our Federal
Government.
And one important point, and I know it has been made, but I
just want to reemphasize this, AI is only as good as the inputs
it receives, inputs it is trained on, the people who train it,
and the objectives is trained to try to achieve, and that is
something that I do think is being lost right now. In the rush
to think about AI, I think we have to think about what some of
the goals are. And a lot of mention has been made of the rash
and reckless actions made by the Trump Administration,
specifically Elon Musk and DOGE, around AI, and again, I think
what are we looking at as far as goals? I think we want to make
sure that the missions of our different agencies, of the tasks
they are meant to accomplish, is what is in mind, and not just
simply cutting for the sake of cutting.
And so, for example, you know, when we think about AI, DOGE
may be using AI right now to surveil communications at Federal
agencies, performing potential cuts, including staff and
programs. They have stated that they want to use generative AI
to automate much of the work previously done by civil servants,
but there are a lot of potential mistakes there, as well as the
data privacy aspects that have been touched on in previous
questions. But this data privacy, and we do not actually know
at this point what Elon Musk and DOGE have been able to scrape.
They certainly have had access to a lot of our data, illegally
so I would add, and what have they done with that? We have been
told that they have downloaded them onto private servers. This
Committee has not done any oversight on this question, but I
think it is an important one. As we think about AI, is Grok or
Elon Musk's other AI models, are they being trained on data
that is only available because of his position at DOGE? This
also creates a lot of national security risks as well.
And so, I guess my first question is to Mr. Schneier. From
a national security perspective, what safeguards should we be
thinking about implementing to make sure that we are protecting
sensitive data while still promoting artificial intelligence,
design, and innovation?
Mr. Schneier. I mean, we have the safeguards. The
regulations are in place that protect the data. We know how to
do it. We have been doing it for years, for decades, and it is
a matter of implementing them. The problem, as you said, was
what DOGE did was illegal. It was contrary to the laws, and
when I think about using AI in organizations, it is a matter of
bringing the technology in and using it in a way that makes
sense. There is a lot of talk about using it to find fraud.
That seems like a good use.
I would want to add, I see Mr. Timmons left, but how about
tax fraud? That is, like, the for-profit arm of the government,
and we get a lot of bang for our buck for looking for tax
fraud. I think AI would be great at a first thing, but it is a
matter of keeping the data in the organizations with the laws
that protect it.
Mr. Min. And I think Isaac Asimov wrote a book about this a
long time ago, but something similar that we ought to be
training AI, as we think about it, to follow basic rules and
principles that we want them to follow first and foremost
instead of just thinking about the end goals in mind. Now,
again, I want to get back to the point of goals being
important. When we look at Social Security, when we look at
programs like Medicare and Medicaid, it is easy to simply just
cut costs. But if we are diminishing service, if we are just
looking at cutting costs and not at the service being provided
or the efficiency of the service being provided, we end up with
huge mistakes. At the same time, if you have a bunch of coders
out there who do not have any context about the program, how it
is been administered, you end up with huge problems, as we saw.
You know, we saw that DOGE flew into the Administration
without even understanding the basics of how Social Security
was coding its different entries in the COBOL coding process.
And so, in that review, they noticed entries that made it seem
like the Administration was paying individuals benefits who
were aged 150 or older. Trump and Musk actually claimed this
falsely, and again, it was because you relied on computer
wizard geniuses, like, the 22-year-old coder named Big Balls,
instead of actual Administration professionals who understood
what was going on. Similarly, with Medicare and Medicaid, the
goal here should be to maximize healthcare in an efficient way,
not simply to cut costs. The cheapest way to administer
healthcare, of course, is just to let everyone die. That is
cheaper than administering healthcare, and so we cannot lose
the forest before the trees here. So, I see I am out of time.
The last point I just want to make is the 10-year
moratorium on AI regulation, something that is outrageous,
something that was slipped in, in the dark at night, it is a
shocking provision, and it was shocking that my colleague,
Madam Chair Taylor Greene, said she would vote against the
entire Big Beautiful Bill if this provision was not set out.
Ms. Greene. The gentleman's time has expired. I now
recognize Mrs. Luna of Florida.
Mr. Min. Thank you.
Mrs. Luna. Thank you, Madam Chair. You know, I want to
thank you all for being here today. One of the biggest concerns
I have with this whole discussion on AI is a failure to admit,
first and foremost, that it does appear that in the next 10
years, that the way that we have grown up and know life with AI
will be forever changing. With that being said, there are a few
differences that I think most people acknowledge and that AI
lacks a few things, one being a soul, also empathy. And we are
not gods or God, and so I think that we are playing a dangerous
game, especially moving into a future that is simply unknown
right now, especially if we do not essentially develop the
first AI super weapon. Essentially, that is what this is.
Whoever does will essentially control the world, and I think
that that is a very serious topic that needs to be discussed
right now.
But specifically, to transhumanism and coupling of AI with
humanity, what regulatory frameworks can be established to
ensure AI-driven transhumanist technologies, like brain-
computer interfaces prioritize human safety and consent? I am
opening this up to everyone on the panel. I have a few
questions, so I would like to get through this, so just please
limit your responses. Mr. B, if you want to go first, or down
there, either/or.
Mr. Thierer. I will just briefly say, Congresswoman, that
there has been a lot of work done on AI governance by multiple
bodies and entities. I just want to cite something that
actually came from the Biden Administration on this about the
applicability of so many existing laws and regulations to AI.
And their leading regulators, or four large agencies, said
before leaving office that, basically, we have the ability to
enforce the respective laws and regulations to promote
responsible innovation and automated systems, and that,
basically, AI does not get an exemption from civil rights laws,
from consumer protections, unfair and deceptive practices,
fraud, so on and so forth. So, we have a lot of policies that
do regulate many of the fears you are raising.
Mrs. Luna. Is there the possibility of regulation, though,
that exists that would enable transhumanist enhancements that
exacerbate social inequalities or basically financial
inequalities, creating an elite class of enhanced individuals?
I bring that up because we are talking about low-level jobs
potentially being removed from the workforce, people that have
and can afford the ability of, for example, implementing or
implanting chips, given access to unknown amounts of, you know,
knowledge, and essentially creating the first superhuman, and
this is a real concern. I know it might come across as funny to
other people, but I am genuinely concerned about this.
You know, from a bipartisan perspective here, we are
talking about humanity versus machine. Hopefully, in a flowery
world, we would have it set up to where it could, you know,
improve society for the best, but I am also in politics, and I
have a very, unfortunately, sometimes negative perspective on
the world because I have seen the worst of humanity in this
job.
Mr. Schneier. So, science fiction speaks of this a lot. I
mean, there are a lot of people who think about this notion of
haves and have-nots, and how technology will exacerbate that,
and that is going to depend on how it is available and how much
money it costs and who can get it. Nita Farahany writes a lot
about AI and brain interfaces and what we can do to protect
humanity in that world, so I do recommend talking to her. She
is the smartest person I know on that particular question.
Mrs. Luna. OK. And also, what role can AI itself play in
monitoring and mitigating risks associated with transhumanist
technologies, such as detecting biases and enhancement
distribution?
Mr. Shah. I think a lot of the conversation so far since
ChatGPT came out is about foundation models and how they
operate and how they hallucinate, and some of the shortcomings
of them and the risks with them, but a lot of these problems
can be overcome at the application layer. And Ms. Miller and I
represent companies who build applications on top of many
models, and so if you think about that framework, what you
understand then is the responsibility of some of these
outcomes, the responsibility of some of these behaviors is not
just left to the model. It is left to the vendor and the
developer of----
Mrs. Luna. So, how you train it.
Mr. Shah. How you train it, how you package it, how you
check for issues and guardrails and all of that.
Mrs. Luna. Yes. So, real quick, I have one more question,
just real quick, yes or no. Would either or any of you be in
favor of the U.S. Government through either the DIA, NSA, all
agencies involved, Pentagon, even creating a super AI--I would
maybe term it guardian AI--to actually train up, based on our
principles, to defend the United States and potentially world
against maybe a more nefarious actor like China, creating the
same super weapon, because AI can only fight AI, I think you
guys would all agree with that. In an effort to defend the home
front?
Mr. Schneier. I think we cannot do that yet.
Mrs. Luna. Not yet, but maybe in the future?
Mr. Schneier. Maybe in the future.
Mrs. Luna. OK.
Mr. Thierer. Maybe, but I need more about the details about
what that means because there is a lot to unpack there.
Ms. Miller. Unqualified, probably, to weigh in on that one.
Mr. Shah. I am supportive of any sort of sovereign AI that
we think makes sense.
Mrs. Luna. Sovereign AI would be a good name.
Mr. Bajraktari. I think government alone cannot do it. I
think public-private partnership should be required in this
case because, otherwise, we will not be able to confront China.
Mrs. Luna. Thank you.
Ms. Greene. The gentlelady's time has expired. I now
recognize Mr. Frost from Florida.
Mr. Frost. Thank you, Chair. You know, the first few months
of the Trump Administration have been unhinged. At several
events in my district, I have heard from outraged constituents.
People, as I have traveled across the country, have heard the
same from people in many different districts. And there are two
common themes I keep hearing about: No. 1, people that are
angry at the Administration's attempts to degrade vital
services, like, Social Security, Medicaid, FEMA, and the VA;
and also fear about what Elon Musk and folks at DOGE have done
and could be doing with the data of American citizens and the
American people, things like health records, banking
information, and Social Security numbers. These conversations
are really just the tip of the iceberg. Hundreds of
constituents have also called into my office with the same
fear, that their private information is falling into the wrong
hands.
Mr. Schneier, reporting in May revealed that DOGE is now
feeding unknown amounts and unknown types of sensitive
government data into Grok, the AI program owned by Elon Musk. I
find this extremely concerning. What are the potential harms
that individuals now face from DOGE doing something like this?
Mr. Schneier. Well, we do not know. So, it is the two
things I would worry about, feeding it to Grok and also giving
it to Palantir to train their models. So, once it has now left
the government, it is going to different companies for
different things. And I think that the fact that we do not know
what the data is, what is being done to it, how it is being
used, how it is being protected, is a grave danger, right? We
secure that data because it is important and now it is
somewhere else also. Also, we do not know the accuracy. I have
not talked about it. I would worry about data poisoning. If you
were an American adversary, one of the things you might want to
do is go to that ill-protected data and change stuff because
now the model is being trained on false data and we talked a
little bit about hallucinations and making bad decisions.
Mr. Frost. In what way could the dumping of Americans'
private data into personal and corporate AI programs create
conflicts of interest or create personal enrichment for certain
people?
Mr. Schneier. Now, I mean, data is power and whoever has
the data has the power, and for us to give that data to
corporations, really with, you know, not even charging them for
it, gives them an enormous amount of power. How they can use
that is going to depend on what they are doing, and future
technologies are going to make different things possible, but
think of it broadly as it increases the power of those who have
the data.
Mr. Frost. And it is important for everyone to, you know,
remember that we have seen many different conflicts of interest
throughout this Administration. We have also seen Trump, Musk,
and other folks dismantle data privacy protections and ignore
law after law. Since we cannot rely on the President or his
Administration to obey the law, what other safeguards currently
exist to keep our data safe?
Mr. Schneier. No, that is all you got, I mean, you as a
citizen. I mean, my data, that is at the IRS. I can do nothing
about it, right? I am relying on you, the government, to
protect my data. If you are not doing that writ large, I am
kind of screwed.
Mr. Frost. For 130 days, we have seen Musk and other actors
lurk inside our most sensitive data bases, grabbing whatever
private data they want with zero to little oversight,
especially no oversight from the U.S. Congress, including
Republican Members of Congress as well. How can we as the
Oversight Committee assess the damage that has been done and
hold people accountable?
Mr. Schneier. I think you actually have to do the work.
There is an enormous amount of investigation that has to
happen. We do not know what data was taken. We do not know
where it went. We do not know where it is. And that is going to
be work figuring it out, but I think it is work that has to be
done because I think the first step to fixing the problem is
understanding what it is. I mean, we can stop the flow, you can
close down the loopholes, you can get these people out of the
organizations, you can start protecting the data, but now the
data is gone. You have to figure out where it is, who has it,
and then take it back.
Mr. Frost. Interestingly enough, you know, you brought up
the word ``accountability,'' which is really important. When
the Administration changed over from Democrats to Republicans,
I think little things matter. I think words matter, and, you
know, there is a word that changed in the name of this
Committee. The word ``accountability'' used to be in the name
of this Committee, and they changed it from government
accountability to ``government reform.'' It seems like a lot of
my Republican colleagues have no interest in holding this
Administration accountable for the misuse of data, and Congress
has a horrible record in being able to legislate as it relates
to technology. I think there are huge educational barriers. We
bury our heads in the sand until the problem gets out of hand,
and then someone comes up with a grand solution of banning it.
This is what happened with social media, unfortunately, what I
think might also happen with AI unless we get our stuff
together here. Thank you. I yield back.
Ms. Greene. The gentleman yields. I now recognize Mr. Crane
from Arizona.
Mr. Crane. Thank you for holding this hearing today. I want
to say thank you to all the panelists for coming. I think we
are all aware that the age of AI is here and expansion is
absolute and unstoppable, but I do want to talk about what I
feel to be the elephant in the room when we talk about AI, and
one of my major concerns as a representative of the people to
this country is our workforce.
Recently, Dario Amodei, CEO of Anthropic, one of the
world's most powerful creators of AI, has a scary warning for
all of us. He says this: ``AI could wipe out half of the entry-
level white-collar jobs and spike unemployment 10 to 20 percent
in the next 1 to 5 years.'' Other experts have said that number
is closer to 50 percent in the next 5 years. He also said that
``AI companies and the government need to stop sugarcoating
what is coming, the possible elimination of jobs across
technology, finance, law, consulting, and other white-collar
professions, especially entry-level gigs,'' he says.
Steve Bannon, who was a top advisor to President Trump in
his first term, recently told Axios, ``I don't think anyone is
taking into consideration how much administration,
administrative, managerial, and tech jobs for people under 30,
entry-level jobs that are so important in your 20s, are going
to be eviscerated, and it is already happening.'' Research that
we have done shows that IBM just laid off 8,000 employees as of
last week; McKinsey, a consulting firm, 5,000 employees as of
May 28. Dell has laid off 6,000 employees in 2024. Intel has
laid off 15,000 employees in 2024. And all of these were
credited to AI and efficiency. Estimated total jobs lost,
214,000 total jobs lost to AI.
I recently asked AI, what are the careers that will be
replaced first by AI, and AI told me data entry clerks,
telemarketers, retail cashiers, warehouse workers, paralegals
and legal assistants, basic customer support representatives,
bank tellers, fast food workers, IT support technicians, tax
preparers. So, we are talking about tens of millions of jobs,
and I think that that is very scary, even though we have been
talking about the efficiencies of AI, how much money it can
save the Federal Government, corporations, and how we have to
keep up with China.
One of my questions to you guys, and I am not blaming any
of you, but does it bother you at all that the technology you
are developing is set to destroy tens of millions of jobs? I
want to start with you, Mr. Shah.
Mr. Shah. You know, technology over time has generally
raised the quality of living for all humans, and, you know, I
know a lot of my fellow CEO AI folks out there try to grab
headlines with some of their claims. I have not seen Dario's
math behind all of this, but I think the point there is
correct, that low-paying entry jobs will be taken on by
computers, but it gives us a chance to reposition those
individuals, re-skill them, and I do not think this is going to
end. I think it is going to continue. For the next, you know,
every 7 years, we are going to be talking about this.
What we have found is that for customers of ours, if you
look at the private sector, they have taken those efficiencies,
turned them into profits. Profits have then led to more
ambition to grow their businesses. In fact, the majority of our
customers have actually increased their employee headcount over
time because they want to endeavor to do more things. So, I
think when we start to automate mundane tasks, we can start to
do more interesting work, and I think organizations will take
full advantage of that.
Mr. Crane. And you think that applies across the board to
all those careers and industries that I just listed off, not
just the ones in tech?
Mr. Shah. I think that there will be changes to many
industries because of this, and I think we are going to have
to, as a species, respond to what this brings us and be able
to, if we want to continue to lead the global race in AI, our
agility is going to really matter.
Ms. Miller. I would just add----
Mr. Crane. Go ahead.
Ms. Miller. I would like to just add that, I mean, Mr.
Schneier mentioned this earlier, this is a serious issue. We
are going to displace a lot of people, and we do not have the
training right now to give them new jobs. We are talking about
unemployment rates of 15 percent of college graduates, to say
nothing of those who have not got a college degree. So, I do
believe this is a serious issue, and it is not going to get any
better.
Mr. Crane. I have one more question. I want to shift gears
real quick and talk about AI and defense. Mr. Bajraktari, I am
sorry, I looked at your resume. You were the Executive Director
of National Security Commission; Chief of Staff to National
Security Advisor, Lieutenant General H.R. McMaster; former
Chief of Staff to Deputy Secretary of Defense. Recently, the
Ukrainians launched a drone attack on Russian soil. Ukraine
said over 40 bombers, or about a third of Russia's strategic
nuclear bomber fleet, was damaged. It was reported that this
was done with AI. Do you think that this attack could have been
possible without the U.S. help with AI intelligence?
Ms. Greene. The gentleman's time has expired, but go ahead
and answer quickly.
Mr. Bajraktari. So, I do not have any insights into that. I
left the government a couple of years ago, but it only shows
how technology is changing the battlefield today, and I think
how Ukrainians are adopting to this new battlefield is really
an example that they have showcased last weekend.
Mr. Crane. Thank you. I yield back.
Ms. Greene. I now recognize Mr. Bell from Missouri.
Mr. Bell. Thank you, and thank you to our witnesses for
being here today to speak on this important issue.
Americans are concerned about AI and want protections from
the risks it poses to their privacy, security, rights, and
freedoms. Key AI guardrails that provided these protections at
the Federal level were repealed. Now House Republicans are
pushing legislation that bans states from enforcing laws or
regulations to protect Americans from AI for a full decade. Mr.
Schneier, why is it important that the Federal Government adopt
AI responsibly and with necessary oversight?
Mr. Schneier. Because doing it badly is going to harm
people, and these are powerful things. So, these are
technologies that are going to, in a sense, replace human
activities. Now, whether they are going to be assistance to
humans or they are going to replace humans for low-level
decisions, they are going to do the things that humans did. And
we know that when humans make bad decisions in government,
people get hurt, so AI is going to do the same thing, and the
problem is they can do it faster. They can do it more
efficiently, so they could more efficiently hurt people, they
could more efficiently help people, and that is why you need to
do it responsibly.
Mr. Bell. When DOGE pressured Secretary Hegseth and the
Department of Defense to quickly purge any mention of race
across its web pages, the Agency used AI. The result was the
Pentagon removed a tribute to Jackie Robinson, an American
sports legend, civil rights icon, and World War II Army
veteran, who served his country honorably, simply because it
acknowledged his race. We are talking about American history
that was flagged and removed in the blink of an eye. Mr.
Schneier, without such safeguards, what risks do we face in
ensuring the prevention of purges, distortions, or
discrimination in Federal AI?
Mr. Schneier. So, again, it is less the AI and more the
people that used it, and the weirder one is we removed the page
about the Enola Gay bomber, which was an aircraft and not
actually gay. It just had the name. So, when these technologies
are used without human oversight, as humans, we could recognize
that is ridiculous, but an AI does not have that context. An AI
is going to do, like any computer, exactly what it is told, and
if it is given bad instructions, if it is not monitored, if
there is not audit, there is not integrity in the system, you
are going to get bad outcomes. So, the AI did it, but blame the
humans who asked the AI to do it.
Mr. Bell. Removing access to information is just the first
step toward deeper consequences. DOGE is creating lasting
impact in our expert Federal workforce alone. We have seen a
purging of nonpartisan Federal workforce, even though the
Government Accountability Office has for years been sounding
the alarm that a lack of skilled workers will lead to more
waste, fraud, and abuse in government. Understanding how AI
works and its possible biases in programming requires a very
specific skill set, one which we are losing every day. Mr.
Schneier, again, why is it important to retain skilled workers
with institutional knowledge who can carefully oversee the
integration and use of AI across our Federal agencies?
Mr. Schneier. Because the AI is not ready to do that yet.
Without the people, the AI is going to run without context, and
you are going to get mistakes like the ones we just talked
about. And these are mistakes that are, in some ways,
theoretical, they are about information, but they also harm
people. We are talking about using AI to reduce fraud. Now you
are going to have problems with false negatives. It is fraud
you are not going to catch. You are going to have problems with
false positives. There is fraud you are going to catch that is
not real. A really good example comes from the U.K. Horizon
Software was used to audit rural post offices and flagged
hundreds of instances of fraud that was not real, costing
people their jobs. There were suicides. This is the problem of
getting it wrong, and humans can make these same mistakes. We
just make them slower. AI makes the pace of this much worse.
Mr. Bell. In the name of efficiency, the government has
recently slashed funding for Job Corps, a program that offers
free career technical training for young people aged 16 to 24.
Instead of investing in our future workforce, they have
eliminated a program that could help fill the shortage of
skilled workers in emerging tech careers. When it comes to the
Federal Government, it is not a question of just efficiency,
but also of equity, fairness, and safety. I believe in
efficiency, and I believe that technology can do enormous good
for the American people, but this is a country that has a
history that AI cannot erase. I look forward to ensuring that
when the Federal Government employs AI, it actually serves the
needs of those who have too often been left behind.
Ms. Greene. The gentleman's time has expired.
Mr. Bell. I yield back. Thank you.
Ms. Greene. I now recognize Mr. McGuire from Virginia.
Mr. McGuire. Thank you, Madam Chair. Thank you to our
witnesses for being here today. I want to start by saying that
I agree with Madam Chair, Congresswoman Marjorie Taylor Greene,
that somehow we missed this 10-year ban on AI state laws, and I
am hoping that we will fix that in the U.S. Senate. The
President Trump and the White House, they said that we must
ensure American AI dominance, and just yes or no real quick--
just yes or no real quick--do you agree with the Trump
Administration that the U.S. must have AI dominance? Just yes
or no real quick.
Mr. Bajraktari. Yes.
Mr. Shah. Yes.
Ms. Miller. Yes.
Mr. Thierer. Yes.
Mr. Schneier. No, I do not think it works that way anymore.
Mr. McGuire. Thank you. All right. So, DOGE has done
outstanding work in identifying waste, fraud, and abuse. They
have brought to the public's attention extreme inefficiencies
in our government, especially involving outdated technology. In
one example, DOGE highlighted the OPM Retirement Operations
Center. This facility is located in an underground mine that
holds over 400 million individual paper documents. It should
come as no surprise that using outdated technologies leads to
more mistakes. During the Fiscal Year 2024, the Federal
Government reported an estimated $162 billion in improper
payments. Banks and credit card companies use AI and machine
learning to prevent fraud, oftentimes before any money changes
hands.
So, I guess I will ask each witness, do you think the
Treasury payment system could use similar technology to lower
waste, fraud, and abuse? And I am talking AI.
Mr. Bajraktari. Yes.
Mr. Shah. In general, yes. I am not qualified on this
particular topic.
Ms. Miller. Well, there are a number of executive orders
and legislative proposals that are aimed at trying to improve
Treasury's payment verification services and the data that
Treasury is being able to gather in. But one of the problems
with focusing entirely on Treasury, though I do believe
Treasury is the right entity for a centralized system, is that
Treasury is making a decision right before a payment is made
where we can actually put these technologies upstream within
the agencies before an application is approved because many
times we are approving people, we are approving the shell
companies or entities that do not exist. So, if we are using
better data across the agencies, not just at the Treasury, but
absolutely modernizing Treasury system and getting more data,
and I am very supportive of the executive orders that have come
out to date and some of the legislative bills that are being
proposed.
Mr. McGuire. I got a couple more questions coming. Let me
ask the last two witnesses, yes or no.
Mr. Thierer. Yes.
Mr. Schneier. I am with Ms. Miller that you want to push
this upstream.
Mr. McGuire. OK. Do you think the Federal Government is
losing money by not using modern technology, including AI and
machine learning? That is just a yes or no.
Mr. Bajraktari. It is more complicated. So, you need the
digital infrastructure to use AI, so I think we need to first
modernize the government IT infrastructure so you will be able
to use AI.
Mr. Shah. The private sector is doing this, so I think the
government could do as well.
Ms. Miller. Absolutely, without question.
Mr. Thierer. Yes, I agree.
Mr. Schneier. Yes, especially tax revenue.
Mr. McGuire. OK. Mr. Shah, as AI becomes more integrated in
decision-making systems, are you concerned that these tools
might reflect political bias? Specifically, are today's AI
models biased toward left-leaning perspectives, and what
implications might that have for public trust or government
use?
Mr. Shah. There is a proliferation of AI models now being
trained on all sorts of data sets, and hallucinations are a
thing, bias are a thing, and we can never really overcome them
forever. I think there are going to still be traces of that
given the type of technology that they are, but you can build
applications that can control for this. You can absolutely look
and have models checking models, checking models, and have a
structure that ensures that everything it decides can be backed
up and that can be referenced. And while that will not solve
every particular case, nor can we do that with human beings, I
think the models and the AI systems can get very, very
accurate.
Mr. McGuire. President Biden let in, well, 15 million
illegal immigrants into our country and allowed deadly drugs,
like fentanyl, to pour into our communities. In just under 6
months, President Trump has secured it with daily encounters
dropping by 95 percent. We have seen Department of Homeland
Security successfully deploy AI to stop illegal immigration
from cameras and sensors along the border to drug detection
equipment and ports of entry. However, severe damage has
already been done. There are an untold number of illegal aliens
currently reside in our country--and I apologize if I pronounce
your name wrong--Mr. Thierer, do you see any ways we can
leverage AI to help facilitate deportations?
Mr. Thierer. I have not followed that one closely,
Congressman, but I do know DHS is using AI tools for these
purposes.
Mr. McGuire. Awesome. Well, thank you for all your
testimony today, and I yield back.
Ms. Greene. The gentleman yields. I now recognize Ms.
Pressley from Massachusetts.
Ms. Pressley. Thank you. Just 2 weeks ago, Republicans
passed a reconciliation bill that would ban all states from
regulating AI for 10 years. Since most did not actually read
the bill, they were unaware of that, so this is a timely
hearing as Republicans are learning the error of their ways in
real time. They are offering the American people up as guinea
pigs for corporations to conduct AI experiments with no
oversight, no regulation, and no enforcement.
Mr. Chair, I ask unanimous consent to enter into the record
a letter by the Massachusetts Joint Committee on Advanced
Information Technology, the Internet and Cybersecurity from May
30, 2025, which details the dangers of the Republican
moratorium. Unanimous consent?
Mr. McGuire. [Presiding]. Without objection.
Ms. Pressley. All right. Now, let us talk about why
regulating AI matters. Artificial intelligence reflects the
assumptions of those who build it and the priorities of those
who use it, and too often, those priorities fail to include the
safety, the rights, and economic opportunity of Black
communities and other marginalized groups. In my district, the
Massachusetts 7th, we are proud to be a national hub for
technology and innovation. For example, Northeastern
University's Center for Inclusive Computing is leading the
charge to make sure that AI opportunities are accessible to
all. At the same time, we cannot allow AI to be the latest
chapter in America's history of exploiting marginalized groups,
namely the Black community. The government must invest in an
approach rooted in equity that protects the rights of everyone.
That is why, Mr. Chair, I ask unanimous consent to enter
into the record this report released last year from Color of
Change titled, ``Black Tech Agenda: Advancing Equity and
Reimagining Technology.''
Mr. McGuire. Without objection.
Ms. Pressley. All right. Mr. Schneier, do you know which
government employee or agency or board is responsible for
overseeing AI deployment across the executive branch to prevent
violations of civil liberties and civil rights?
Mr. Schneier. I do not believe there is one, so if this is
a test, I just failed.
Ms. Pressley. Well, the truth is no one knows. The
responsibility of AI civil rights enforcement is not in
anyone's job description, and that has got to change. The
United States Federal Government is deploying AI and it is
impacting our daily lives. There are more than 2,000 AI use
cases across agencies. Right now, there is nothing stopping the
Department of Health and Human Services from using AI systems
to help decide who receives Medicaid, who receives food
assistance, and disability benefits. If an algorithm flags
someone incorrectly, a mother could quite literally lose access
to her child's medication, or a family could lose food support.
There is nothing stopping the Department of Justice from
using racially biased algorithms to decide who gets released
from prison. AI should not be determining who gets care, who
stays in prison, or who can feed their kids without any sense
of oversight or recourse. The government's use of AI cannot be
unchecked and unregulated. If we want the public to trust the
Federal Government, then we must have transparency. I yield
back.
Mr. McGuire. The gentlewoman yields. I now recognize
Congressman Moskowitz for 5 minutes.
Mr. Moskowitz. Thank you. Thank you, Mr. Chair. I
appreciate the waive-on today. You know, obviously, this is an
important topic on artificial intelligence. I got to be honest.
I definitely want to meet the person in the Capitol who was
like, we are going to have a hearing on intelligence. Let us
give that to Chairman Comer. But, you know, we saw some
language wind up in a bill that everyone is complaining that,
you know, they did not read. Remember, they were the ones who
had the 72-hour rule. That was their rule. In fact, they cried
and demanded that every bill come forward and be given 72 hours
to read a bill. They complained when Democrats were in control.
Meanwhile, they passed this bill in the dead of night. Do not
listen to me. That is what Elon Musk is now saying on Twitter.
Yesterday, in the DOGE Subcommittee hearing, Chairman Comer
announced new investigations into the Biden Administration, and
I thought, of course. This makes total sense, right? We have
recently brought back measles and the dire wolf and the tariff
policies of the 1930s, so you know, why not resurrect the
failed Biden impeachment? I do not want to go down memory lane
about what happened last year, but, you know, they said they
would call for impeachment. That is what they said, and they
never did. After hundreds of interviews, fundraising emails,
and speeches, and lying to their base, they never called for
impeachment. So, I want to do the do not listen to what they
say, watch what they do, and this is, of course, all a
distraction, right?
They said they would lower food costs. They said they would
take us into the golden age. They said they would end the war
in Ukraine and Gaza. They said they would not touch Medicaid.
They said they would lower interest rates. They said they would
lower our debt. They said they would release the Epstein files.
Remember that? This is not one of the ones that Pam gave out. I
made my own. It is not hard, right? And the biggest one, the
biggest one, my favorite, they said they would make government
more efficient, DOGE. ``E'' stood for efficiency.
Name one department, one government service, anything that
the government does that they made more efficient. Certainly
not the Newark airport. Definitely not FEMA. We got an
Administrator who is finding out about hurricane seasons, a
little late. Total failure. No new technology, no new
procurement rules. Procurement is a big problem in government.
You got different agencies, different departments making
different procurements, buying technology that cannot
communicate with each other. By the way, we learned that after
9/11 in law enforcement and we still have not fixed that
problem. No new streamlined processes. Zero, OK?
And so, look, sometimes people are visual learners, right?
They are visual learners, and so look, this is what the
Republicans promised the American people. They promised them
this, OK? This was the Willy Wonka thing in London. This is
what they promised them. This is what it was supposed to look
like, right? The golden age, and that is not what they have
delivered. What they have delivered is this. This is what they
have delivered the American people. This is what is left of the
DOGE effort that--she got to keep her job. This is what is
left. And so, they are very good at telling the American people
what they want to do, what they say they are going to do, but
the American people should start watching what they are
actually doing, or what they have done or what they have not
done.
And so, listen, fine, bring in the Biden people again. We
will go through this whole thing. Tens of millions of dollars
wasted in the last Congress. We will do it all again, right,
back to our old tricks, but of course, they want to bury them
in the basement again. It is being proposed that, oh, we will
take their depositions, bury them in the basement. Then they
will leak the transcripts. They tell you, well, we want to
shine transparency on the end of the Biden Administration. We
want to get to the truth, but they want to hold these meetings
in secret, not in front of the cameras, not in front of the
American people, in the bottom of a basement. By the way, I
noticed that the Oversight Hearings are now in the bottom of a
basement. Ironic, after all the basement talk last Congress,
but they just want to suppress information. They want to do
selective leaks and they want to distract that this is the
least productive Congress in 70 years.
So, listen, it is fine. We will do it all again, OK? Will
not help people at the grocery store, will not help people save
their healthcare. It will do none of that, but it will get you
on OANN and Fox News and raise money. Thank you.
Mr. McGuire. I now recognize Congresswoman Trahan for 5
minutes.
Ms. Trahan. Well, thank you, Mr. Chairman. I appreciate you
allowing me to be part of this important conversation. Over on
the Energy and Commerce Committee, which is where I usually
serve, we have a lot of conversations about technology, and one
thing is always clear: data is at the heart of AI. That is why
I believe that any serious discussion about AI has to start
with a conversation about privacy, and that is what I am here
to do today, to sound the alarm about a deeply troubling trend,
our own government's growing appetite for Americans' personal
data.
Let me just give you an example, a hypothetical, of course,
but not a far-fetched one. Sarah is a typical American woman.
She pays her taxes, owns a gun legally, and is raising her
daughter, Emma, on her own. She and Emma rely on Medicaid to
get the healthcare that they need. One day, Sarah shares a post
on Facebook. She is concerned about something the President
said about firearms and she posts so, but in Washington, an AI-
powered monitoring system flags her post. A political appointee
digs into her personal data and sends emails to agency heads
urging them to take action against her. Within days, Sarah's
life falls apart. The IRS audits her and claims she owes
thousands. Emma's doctor says her Medicaid is not active
anymore and now Sarah has to pay out of pocket. Now, to be
clear, this story is made up, but it is not science fiction. It
is an alarm. It is a warning.
Mr. Schneier, you talked in your testimony about coercion
as an adversarial use of data. What kinds of coercion could bad
actors inside the government use if they had detailed profiles
on every American?
Mr. Schneier. I would think of it as selective
investigation. I mean, the government has enormous powers to
investigate people, and the question is who they choose to
investigate. There is a famous book from many years ago, called
``Three Felonies a Day,'' that we, in our normal lives, commit
three felonies a day because there are just so many rules and
we do not know them, and so, given things like that, who you
choose to enforce the law on matters. So, this data can be used
to select people whom to investigate, people whom to charge,
and this could be used selectively by any regime, even not the
U.S., any country that wants to do this.
Ms. Trahan. Yes. Well, unfortunately, this is not just a
hypothetical trend. It is already happening. Under the Trump
Administration, DOGE aggressively collected sensitive data
across agencies, breaking down firewalls that are supposed to
protect us. Then came the executive order directing agencies to
eliminate information silos, basically to share and pool that
data. And just last week, we learned that Palantir, a Silicon
Valley company known for building surveillance tools, is being
hired to build AI-powered profiles on every American using the
data that DOGE collected. It is hard to overstate how dangerous
this is. Mr. Schneier, are you worried that once this data is
centralized, future administrations, no matter their party,
could weaponize it? I mean, are we on the verge of opening
Pandora's box here?
Mr. Schneier. I do not know if Pandora's box has been
opened years ago, but certainly giving this power to a
government is something that feels very un-American. I mean,
there are reasons why this data was siloed. There are reasons
why we did not have these powers. I mean, you can imagine
humans doing this well before AI, but we chose not to. So, AI
can certainly make this more efficient, but yes, this is power
in the hands of a human who wants to wield it for ill, can do
that very efficiently.
Ms. Trahan. Thank you. We need a national reckoning on
privacy. That means strong oversight of this Administration and
its tech partners and real legislation to protect Americans'
rights. You know, I have spent the past 3 months talking with
civil liberties groups, privacy experts, and people across the
country, and the one thing is clear. We need stronger privacy
laws. I believe that we can protect people's data and modernize
government to prevent fraud, waste, and abuse. These goals are
not at odds. They are linked. So, if you are listening and you
are concerned about what is happening about Big Tech, about
government overreach, about your family's privacy, call my
office. Let us have a national conversation. Let us protect the
freedom that our founders fought for and the privacy we all
deserve.
And one last thing I just wanted to mention, because over
the course of this hearing, the Chair has suggested that no one
on the other side of the aisle called attention to the harms of
the Republicans' 10-year ban on state AI regulations. That is
patently false. We had robust debate on the Energy and Commerce
Committee with several Democratic members, myself included,
calling attention to this provision during and after our 26-
hour markup. In fact, Democrats offered an amendment to strike
the language entirely. So, Mr. Chair, I ask unanimous consent
to enter into the record the results of the recorded vote.
Mr. McGuire. The gentlewoman's time is up. Without
objection.
Ms. Trahan. Thank you. I yield back.
Mr. McGuire. In closing, I want to thank the panelists once
again for their testimony today.
With that and without objection, all Members will have 5
legislative days within which to submit materials and to submit
additional written questions for the witnesses, which will be
forwarded to the witnesses for their response.
If there is no further business, without objection, the
Committee stands adjourned.
[Whereupon, at 2 p.m., the Committee was adjourned.]
[all]