[House Hearing, 119 Congress]
[From the U.S. Government Publishing Office]
ARTIFICIAL INTELLIGENCE AND CRIMINAL
EXPLOITATION: A NEW ERA OF RISK
=======================================================================
HEARING
BEFORE THE
SUBCOMMITTEE ON CRIME AND FEDERAL
GOVERNMENT SURVEILLANCE
OF THE
COMMITTEE ON THE JUDICIARY
U.S. HOUSE OF REPRESENTATIVES
ONE HUNDRED NINETEENTH CONGRESS
FIRST SESSION
__________
WEDNESDAY, JULY 16, 2025
__________
Serial No. 119-31
__________
Printed for the use of the Committee on the Judiciary
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
Available via: http://judiciary.house.gov
______
U.S. GOVERNMENT PUBLISHING OFFICE
61-182 WASHINGTON : 2025
COMMITTEE ON THE JUDICIARY
JIM JORDAN, Ohio, Chair
DARRELL ISSA, California JAMIE RASKIN, Maryland, Ranking
ANDY BIGGS, Arizona Member
TOM McCLINTOCK, California JERROLD NADLER, New York
THOMAS P. TIFFANY, Wisconsin ZOE LOFGREN, California
THOMAS MASSIE, Kentucky STEVE COHEN, Tennessee
CHIP ROY, Texas HENRY C. ``HANK'' JOHNSON, Jr.,
SCOTT FITZGERALD, Wisconsin Georgia
BEN CLINE, Virginia ERIC SWALWELL, California
LANCE GOODEN, Texas TED LIEU, California
JEFFERSON VAN DREW, New Jersey PRAMILA JAYAPAL, Washington
TROY E. NEHLS, Texas J. LUIS CORREA, California
BARRY MOORE, Alabama MARY GAY SCANLON, Pennsylvania
KEVIN KILEY, California JOE NEGUSE, Colorado
HARRIET M. HAGEMAN, Wyoming LUCY McBATH, Georgia
LAUREL M. LEE, Florida DEBORAH K. ROSS, North Carolina
WESLEY HUNT, Texas BECCA BALINT, Vermont
RUSSELL FRY, South Carolina JESUS G. ``CHUY'' GARCIA, Illinois
GLENN GROTHMAN, Wisconsin SYDNEY KAMLAGER-DOVE, California
BRAD KNOTT, North Carolina JARED MOSKOWITZ, Florida
MARK HARRIS, North Carolina DANIEL S. GOLDMAN, New York
ROBERT F. ONDER, Jr., Missouri JASMINE CROCKETT, Texas
DEREK SCHMIDT, Kansas
BRANDON GILL, Texas
MICHAEL BAUMGARTNER, Washington
------
SUBCOMMITTEE ON CRIME AND FEDERAL
GOVERNMENT SURVEILLANCE
ANDY BIGGS, Arizona, Chair
TOM TIFFANY, Wisconsin LUCY McBATH, Georgia, Ranking
TROY NEHLS, Texas Member
BARRY MOORE, Alabama JARED MOSKOWITZ, Florida
KEVIN KILEY, California DAN GOLDMAN, New York
LAUREL LEE, Florida STEVE COHEN, Tennessee
BRAD KNOTT, North Carolina ERIC SWALWELL, California
CHRISTOPHER HIXON, Majority Staff Director
JULIE TAGEN, Minority Staff Director
C O N T E N T S
----------
Wednesday, July 16, 2025
OPENING STATEMENTS
Page
The Honorable Andy Biggs, Chair of the Subcommittee on Crime and
Federal Government Surveillance from the State of Arizona...... 1
The Honorable Lucy McBath, Ranking Member of the Subcommittee on
Crime and Federal Government Surveillance from the State of
Georgia........................................................ 3
WITNESSES
Dr. Andrew S. Bowne, Professor, George Washington University
Oral Testimony................................................. 5
Prepared Testimony............................................. 8
Zara Perumal, Chief Technology Officer, Overwatch Data
Oral Testimony................................................. 24
Prepared Testimony............................................. 26
Cody Venzke, Senior Policy Counsel, National Political Advocacy
Division, American Civil Liberties Union
Oral Testimony................................................. 40
Prepared Testimony............................................. 42
Ari Redbord, Global Head of Policy, TRM Labs
Oral Testimony................................................. 72
Prepared Testimony............................................. 74
LETTERS, STATEMENTS, ETC. SUBMITTED FOR THE HEARING
All materials submitted by the Subcommittee on Crime and Federal
Government Surveillance, for the record........................ 104
Materials submitted by the Honorable Lucy McBath, a Member of the
Committee on the Judiciary from the State of Georgia, for the
record
A letter to Speaker Mike Johnson, Minority Leader Hakeem
Jeffries, Majority Leader John Thune, and Minority Leader
Chuck Schumer, from the National Association of Attorneys
General, May 16, 2025
An article entitled, ``Inside Congress Live,'' Jun. 27, 2025,
Politico
A testimony from Barry Friedman, Jacob D. Fuchsberg Professor
of Law and Affiliated Professor of Politics, Faculty
Director, Policing Project, New York University School of
Law, Jul. 16, 2025
A letter to the Honorable Andy Biggs, Chair of the
Subcommittee on Crime and Federal Government Surveillance
from the State of Arizona, and the Honorable Lucy McBath,
Ranking Member of the Subcommittee on Crime and Federal
Government Surveillance from the State of Georgia, from
PublicCitizen, Jul. 16, 2025
A testimony from Keith Kupferschmid, Chief Executive Officer,
Copyright Alliance, Jul. 16, 2025
An article entitled, ``The countdown to artificial
superintelligence begins: Grok 4 just took us several steps
closer to the point of no return,'' Jul. 12, 2025, The Blaze,
submitted by the Honorable Andy Biggs, Chair of the
Subcommittee on Crime and Federal Government Surveillance from
the State of Arizona, for the record
APPENDIX
A statement from the Honorable Jamie Raskin, Ranking Member of
the Committee on the Judiciary from the State of Maryland, Jul.
16, 2025, for the record
ARTIFICIAL INTELLIGENCE AND CRIMINAL
EXPLOITATION: A NEW ERA OF RISK
----------
Wednesday, July 16, 2025
House of Representatives
Subcommittee on Crime and Federal Government Surveillance
Committee on the Judiciary
Washington, DC
The Subcommittee met, pursuant to notice, at 10 a.m., in
Room 2141, Rayburn House Office Building, the Hon. Andy Biggs
[Chair of the Subcommittee] presiding.
Members present: Representatives Biggs, Kiley, Lee, Knott,
and McBath.
Also present: Representative Raskin.
Mr. Biggs. The Subcommittee will come to order. Without
objection, the Chair is authorized to declare a recess at any
time. We welcome everyone to today's hearing on Artificial
Intelligence and Criminal Exploitation.
I now recognize the gentlewoman from Florida, Ms. Lee, to
lead us in the Pledge of Allegiance.
All. I pledge allegiance to the Flag of the United States
of America, and to the Republic for which it stands, one
Nation, under God, indivisible, with liberty and justice for
all.
Mr. Biggs. Thank you. I now recognize myself for an opening
statement. I appreciate everyone being here today, our
witnesses, and those in the audience. This is an important
hearing which focuses on artificial intelligence and how it is
being exploited by criminals. The conceptual roots of AI can be
traced to British mathematician Alan Turing when the 1930s
theorized about a machine being capable of performing any
computable task. Today, AI is best understood as a branch of
computer science that leverages large scale data processing,
algorithmic modeling, and modern hardware to enable machines to
perform tasks typically requiring human cognition.
Unfortunately, like most technical innovations, the
criminal element has begun to use AI to enhance their elicit
activities. The AI-enabled threats continue to evolve as bad
actors use AI technology in a wide spectrum of criminal
enterprises. From deepfake scams and synthetic identity fraud
to financial crimes and child sexual abuse material, CSAM, the
landscape continues to evolve at a rapid pace as AI provides
users with enhanced capabilities. The AI-based or AI-driven
threats and schemes can cost businesses millions of dollars a
year including both prevention and falling prey to them. In one
case, fraudsters used AI to clone a CEO's voice and authorized
a wire transfer. Among corporations that experienced a rise in
deepfake incidents, 75 percent of deepfakes impersonated a CEO
or another C suite executive.
Generative AI enables the criminal exploitation of victims'
emotional vulnerabilities through tactics such as sextortion,
pig-butchering scams, phishing, and elder fraud. Senior
citizens are increasingly targeted through voice phishing scams
where an AI-generated replica of a grandchild or military
officer claims to need urgent funds. In one case, a Colorado
mother received a call from what sounded like her daughter
pleading for help. The voice was AI generated, closed from a
short, online clip and used to demand ransom. The voice was
indiscernible to the mother who wired $1,000 to scammers in
Mexico.
AI is also fueling a rise in sextortion and synthetic CSAM.
New AI tools can generate highly realistic, but entirely
fabricated explicit images often used to extort minors or
damage reputations. Some sextortion scams exploit the trust
associated with platforms like Apple's iMessage by
impersonating classmates or romantic interests via recognizable
blue bubble interfaces. Criminals now deploy apps like Muah to
fabricate child abuse images at scale. Stanford University
researchers have uncovered evidence that generative models were
trained on real exploitative content.
Terrorist groups now utilize AI to target, recruit, and
indoctrinate vulnerable individuals. Generative AI provides a
degree of separation, allowing actual terrorists to maintain
anonymity in their public facing recruiting practices.
Generative AI also allows terrorists to produce propaganda,
fake news stories, and emotionally resonant messages tailored
to specific psychological profiles.
Reports of generative artificial intelligence enabled scams
between May 2024-May 2025 rose by 456 percent. The use of
exploi-
tive generative AI allows criminals to produce human-like text,
code, images, and videos allowing criminals to use the
technology for further criminal activity such as creating more
realistic phishing lures or generating deepfakes for extortion.
On average, phishing attacks cost $4.9 million per breach.
On the other hand, AI is increasingly integrated into police
investigations offering new tools and capabilities for law
enforcement agencies into the backdrop of rapidly expanding
digital data sources and increasing demands on law enforcement
agencies. AI provides a more adaptable and comprehensible
approach to solving crimes, compared to traditional methods
leveraging data analytics, machine learning, and pattern
recognition to enhance investigations and assist with
administrative tasks. This is also potentially a problem, as
well, as we seek to balance curbing AI with our civil rights.
AI can also help process large volumes of data, identify
patterns, and generate actionable insights, and in turn, these
applications can improve efficiency, accuracy, and resource
allocation with investigative processes. However, to fully
benefit from AI applications, law enforcement entities need
reliable data, human oversight, while also tackling issues
related to privacy, bias, and ethical considerations.
Addressing the continued misuse of AI will require a varied
approach while also raising public awareness about the risks
associated with AI-generated content.
Law enforcement agencies must engage openly with community
stakeholders, legal experts, and the public to communicate the
intended uses, benefits, and limitations of AI technologies.
The collaborative effort to both prevents the misuse of AI
while encouraging lawful application is required to effectively
navigate this evolving landscape.
I am excited about today's hearing. I think this is the
first of its kind. I believe it will be only the first of its
kind as we consider AI and its continued expansion of influence
on our lives. I am looking for a very substantive discussion--I
anticipate a substantive discussion that we are going to have
today and with that, I yield back and recognize now our Ranking
Member, Ms. McBath, for her opening statement.
Ms. McBath. Well, thank you so much, Mr. Chair, and thank
you to our witnesses today. Thank you so much for taking
moments out of your day to come before us. Thank you for
convening this hearing to discuss AI-enabled crime, efforts to
detect and combat such crime, and how law enforcement deploys
AI tools.
Like so many new technologies, AI is not inherently good or
bad. The AI-enabled tools can find patterns, sort through vast
amounts of information, and may even help law enforcement solve
their crimes. In the wrong hands, the same tools can be used to
commit financial fraud, breach national security systems, and
to harm our children. When used by law enforcement, this
technology has the potential to empower our investigators,
while also carrying the risk of serious errors with life-
changing consequences. That is why it is critical that we
proceed thoughtfully and put appropriate guardrails in place so
that everyone in our criminal justice system that is using AI-
enabled tools, they are using them responsibly, not to the
detriment of law-abiding members of our community.
You have already seen what can go wrong when those
safeguards are missing. A woman and her family experienced the
dangers of using AI enabled facial recognition technology.
Detroit police used a facial recognition tool in an attempt to
identify a carjacking suspect using an image from a
surveillance camera. The tool matched the surveillance image
with a picture of Porcha Woodruff, a nursing school student.
One morning, as Ms. Woodruff was getting her two children ready
for school, the police knocked on her door and they told her
that she was under arrest for carjacking. She knew right away
there must be some kind of mistake, and she gestured at her
body as she spoke to law enforcement to point out the obvious,
she hoped, to law enforcement that she was eight months
pregnant. Though the police had not been looking for a visibly
pregnant woman, they still handcuffed Ms. Woodruff, took her
away from her crying children, held her for 11 hours, searched
her phone, and they charged her. After her release, she went
straight to the hospital and was treated for dehydration. The
charges were dismissed a month later.
This case is especially troubling because facial
recognition tools have been shown to perform worse on Black
individuals, increasing the risk of misidentification and
contributing to over criminalization. AI is only as good as the
data is it trained on and when that data is biased, it
exacerbates racial disparity, long embedded in our criminal
justice system, and an inaccurate tool is dangerous for every
single one of us. Not one of us is immune to these mistakes.
Thankfully, and in due part to cases like this one, the
city of Detroit has adopted new rules to direct the use of
facial recognition technology within its police department, and
they are simply not alone. Many cities and states have put
sensible guardrails in place to limit potentially harmful uses
of AI. That is why it was alarming when some of my Republican
colleagues recently attempted to pass a moratorium on State and
local AI regulations in the big ugly bill, a move that
generated bipartisan opposition so much that 40 State Attorney
Generals and 17 Republican Governors, including the Governor of
my State, Georgia, wrote letters to the Senate in opposition to
the proposed moratorium. The Governors warned that, and I am
quoting them, ``People will be at risk until basic rules
ensuring safety and fairness can go into effect.''
As you will see behind me, Sarah Huckabee Sanders, the
Republican Governor of Arkansas and former press secretary to
President Trump, took to Twitter to quote,
I stand with the Majority of GOP Governors against stripping
States of the right to protect our people from the worst abuses
of AI. The U.S. must win the fight against China on AI and
everything else, but we won't if we sacrifice the health,
safety, and prosperity of our people.
While this most recent proposal was ultimately stripped
from the bill by a 99 to one vote of the Senate, the Republican
Chair of the House Energy and Commerce Committee has already
vowed to continue to pursue a moratorium, even while
acknowledging that Federal regulations on AI are still light
years away.
I stand with those seeking to protect the health and the
safety and civil rights of our communities from the abuses of
AI and I hope that we can come together and follow the lead of
the States to explore what those guardrails should look like
and put them in place.
I look forward to learning more from our experts here
today. Hearing from you is going to be extremely critical for
us on this very important issue.
Before I yield, Mr. Chair, I ask unanimous consent to enter
into the record two letters. The first is a letter from 17
Republican Governors in opposition to a moratorium on State and
local regulation on AI; and the second, a letter from 40 State
Attorneys General, both Republicans and Democrats, in
opposition to a moratorium on State and local regulation of AI.
Mr. Biggs. Without objection.
Ms. McBath. I yield.
Mr. Biggs. The gentlelady yields. Without objection, all
other opening statements will be included in the record. We
will now introduce today's witnesses, and we are very grateful
for our witnesses.
Dr. Andrew Bowne.
Dr. Bowne. Bowne.
Mr. Biggs. Bowne, OK. Dr. Bowne is a Professorial Lecturer
in Law at the George Washington University Law School where he
teaches courses on artificial intelligence law and policy. He
has served in the United States Air Force Judge Advocate
General Corps since 2010 and previously serves as the Chief
Legal Counsel of the Department of the Air Force Artificial
Intelligence Accelerator at the Massachusetts Institute of
Technology.
Thank you, Doctor, for being with us today.
Ms. Zara Perumal is the Co-Founder and Chief Technology
Officer of Overwatch Data, an artificial intelligence company
focused on threat intelligence and cybercrime tactics on the
dark web. Prior to founding Overwatch Data, she worked at
Google on matters involving machine learning and cyber security
threats.
Thank you for you being us today, Ms. Perumal.
Mr. Ari Redbord is the Global Head of Policy at TRM Labs, a
company focused on preventing illicit financial activity. He
previously served as Senior Advisor to the Deputy Secretary and
Under Secretary for Terrorism and Financial Intelligence at the
U.S. Treasury. Prior to Treasury, he was an Assistant U.S.
Attorney where he focused on terrorism, espionage, financial,
child exploitation, and human trafficking cases.
Thank you, Mr. Redbord, for being with us today.
Mr. Cody Venzke is a Senior Policy Counsel in ACLU's
National Political Advocacy Department where his work focuses
on surveillance, privacy, and technology. Specifically, he
works on matters related to artificial intelligence, privacy,
children's privacy, and civic uses of data.
Thanks, Mr. Venzke, for being with us today.
We appreciate all of you being here and now ask that you
please rise so you can be sworn in.
Would you please raise your right hand? Do you swear or
affirm under penalty of perjury that the testimony that you are
about to give is true and correct to the best of your
knowledge, information, and belief so help you God?
Let the record reflect the witnesses have answered in the
affirmative and thank you, you may be seated. Please know that
your written testimony will be entered into the record in its
entirety. Accordingly, we ask that your testimony be summarized
in five minutes and what is going to happen, just so you know,
is about 15 seconds before the end, you will start hearing
this, something like that, and then at the magic moment, I will
start getting a little bit louder, but it will kind of help you
wrap up on time, so we can work this out. We are so grateful
that you are here.
Mr. Bowne, you may begin with your five minutes.
STATEMENT OF ANDREW S. BOWNE
Dr. Bowne. Thank you, Mr. Chair, Ranking Member, and the
distinguished Members of the Subcommittee. Thank you for the
opportunity to testify to the intersection of artificial
intelligence and criminal exploitation.
My name is Andrew Bowne. I serve as a Professorial Lecturer
at the George Washington University Law School where I teach
courses on AI law and policy. I have served in the United
States Air Force Judge Advocate Generals Corps since 2010
including assignments as a prosecutor, a staff Judge Advocate
General Counsel Air Force installation, and as you heard, a
Chief Legal Counsel for the Air Force's AI Accelerator at MIT.
I do appear today in my personal capacity. The views I
present are my own and do not necessarily reflect those of the
Department of Defense, the Department of the Air Force, or the
Judge Advocate Generals Corps.
AI is both a catalyzing and transformative technology
enabler, a solver of traditional processes, but also creates
entirely new ones. When a task AI is used for is criminal or
harmful, the nature of AI becomes a threat multiplier. The AI
systems now automate decisions, model environments, and infer
actions of unprecedented speed, and scale. While they are
designed to benefit society, their dual-use nature means they
could also facilitate exploitation, fraud, and abuse. Even AI
systems designed with legitimate use in mind can create harm if
not carefully designed and deployed with safety, ethics, and
accountability built in.
Today, I would like to briefly highlight how AI enables
criminal activity, the gaps in current criminal law, and
proactive steps Congress might take. First, how AI enables
criminal activity. I am seeing really three categories of AI
developments that are of particular concern in this area:
Computer vision, generative adversarial networks, or GANS, and
large language models, or LLMS.
Computer vision systems which interpret visual data are
used to automate surveillance, identify targets, and even
harvest personal data from breached documents for identity
threat and fraud or enable real time threat detection for
public safety that can be repurposed to stalk or blackmail
individuals with chilling efficiency. GANs are capable of
generating synthetic images, videos, and audio that are known
in public discourse as deepfakes. These tools allow the
impersonation of public officials and private citizens alike.
Multiple watchdogs and law enforcement agencies that have been
conducted longitudinal analysis on AI generated child sexual
abuse material or CSAM are reporting rapid rises in the number
of generated images shared online. Perhaps more concerning is
that the increased capabilities of sophistication of generative
AI models enable them to produce very realistic images.
Generative AI tools are also used for sextortion, grooming,
and emotional coercion, often targeting children. Large
language models like those powering chatbots revolutionizing
phishing, fraud, and social engineering involve misinformation.
They engage victims and extend realistic conversations, target
elderly people and vulnerable people in scams or overwhelming
financial institutions of thousands of tailored loan
applications. They are also used to generate malicious code,
making cybercrime accessible to individuals with no technical
background. In short, deepfakes, AI generated CSAM, and
automated fraud are not theoretical threats. They are real,
growing, and causing harm now. The barrier to entry to using AI
to perpetuate or perpetrate is low. Anyone with a few seconds
of your voice or image can create convincing synthetic content
without coding or expensive hardware. The tools are cheap,
accessible, and often unregulated.
Tragically, gaps in current Federal criminal law allow bad
actors to use AI to profit at the expense of others, prey on
the vulnerable or create mistrust with impunity. While there
are statutes for wire fraud and child sexual abuse material,
they do apply to many of the AI-enabled crimes, there are
several significant gaps.
More alarming still are emerging threats such as autonomous
criminal activity, cross-board AI enabled crime, and
algorithmic market manipulation in which criminal liability is
unclear or altogether absent.
There are a variety of strategies that the Committee, as
well as Congress, can consider, including criminal law reform.
They could define new offenses from malicious use of AI,
particularly deepfakes and AI CSAM. They could also provide
sentencing enhancements for crimes aggravated by the use of AI
tools when those tools augment the scale or impact of the harm
or make it more resource intensive to investigate and
prosecute.
You could also look at enabling AI safety and transparency
requirements and in conclusion AI is a revolutionary enabler
but does not self-regulate. The bad actors who choose to
exploit the law must be ready. Thank you, Mr. Chair, for the
time.
[The prepared statement of Mr. Bowne follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Mr. Biggs. Thank you, Dr. Bowne, and we have your written
testimony, which is more expansive, so I remind everybody we
have that, so appreciate that.
Ms. Perumal, we give you your five minutes now.
STATEMENT OF ZARA PERUMAL
Ms. Perumal. Chair Biggs, Ranking Member McBath, and the
Members of the Committee, thank you so much for the opportunity
to testify today and for creating this forum to discuss how AI
is changing the landscape of cybercrime. I am honored to share
my perspective on how technology is making these threats more
accessible, more personalized, and more difficult to detect.
My name is Zara Perumal. I am the Co-Founder and CTO of
OverWatch Data, a cyber threat intelligence company that use AI
to identify and analyze emerging threats in the cybercrime and
fraud ecosystems. Through our work, we see every day how AI is
used to both prevent and also to facilitate criminal activity.
I would like to focus my remarks today on how we see AI
changing the threat landscape and what we can do about it
through both education and innovation. AI is a powerful,
general-purpose tool with a broad range of applications for
criminal activity and for a broad range of threat actors. It
can be used to learn how to commit crimes, to write code and
craft realistic scam text messages, generate audio or voice
clones, and of course, create deepfake images or videos.
Across this wide range of malicious use, three trends stand
out.
First, AI is reducing the barriers to entry for
cybercrimes. Users can ask a chatbot including ones explicitly
designed for fraud how to commit crimes. They can learn the
technical skills they need to carry them out and they can also
use it to make their attacks more convincing and more
effective.
Second, it is challenging businesses by subverting identity
verification systems. AI can be used to generate a photo for a
fake persona or profile. It can then be used to generate high
quality synthetic fake IDs and then if they are asked to verify
their identity, they can join a video called swap bare face
with their fake photo and verify their access to that business.
This is a problem not just for the specific business that is
being targeted, but it is a problem because it gives them a
foothold from which they can hide their identity for the future
next online crime that they will carry out.
Third, we see is the crimes that are becoming far more
personalized. For example, voice clones are used to target the
elderly by calling a grandparent and what sounds like their
grandchild's voice and claiming to be in the hospital and in
need and in need of money. Employment scams prey on young
adults who are looking for their first job. One of the most
disturbing cases is the nudifying apps which often target
children. They turn ordinary photos into fake sexually explicit
images which are then used to bully, harass, and extort
victims, in some cases driving them to suicide. This abuse is
carried out to children, both by their classmates and by remote
criminals.
There is a lot of harm. There is a lot that is changing,
but there is also a lot we can do.
First, education awareness can prevent and deter many of
these harms. These crimes work because people don't expect
them. If we invest in education across schools, workplaces, and
communities, we can make these crimes less effective and more
costly to carry out.
Second, we have an opportunity to combat this with
innovation. The same technology that is enabling these crimes,
can also be used to detect scams, find malware, and destruct
cybercrimes. By supporting innovation and strengthening public/
private partnerships, we can shift the technical advantage to
the defenders.
While AI enables crime to be more accessible, more
personalized, and harder to detect, I remain optimistic. With
the right investment in education and innovation, we can engage
our whole society in building a future where AI expands access
to opportunities, strengthens safety, and helps people spend
more time on what matters.
On a personal note, it is very exciting to me to see that
Congress is addressing this issue and putting a spotlight on
it. Thank you and I look forward to your questions.
[The prepared statement of Ms. Perumal follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Mr. Biggs. Thank you very much.
Mr. Venske, you are recognized for your five minutes.
STATEMENT OF CODY VENZKE
Mr. Venzke. Chair Biggs, Ranking Member McBath, and the
Members of the Subcommittee, thank you for the opportunity to
testify today on behalf of the American Civil Liberties Union.
I will address two issues this morning: First, it is
crucial that our response to criminal uses of artificial
intelligence adhere to the Constitution, civil rights, and
civil liberties. Second, efforts by Congress and the
administration must not inadvertently open the door for AI
abuses such as through a moratorium on State regulation of AI
or continuing consolidation of Federal data. With appropriate
measures, Congress can ensure that AI is safe, effective, and
consistent with our rights and liberties.
First, Congress' measures to address criminal AI must
comport with the Constitution, civil liberties, civil rights,
and privacy. For example, traditional First Amendment
activities do not lose their protection simply because
artificial intelligence was used. Editorial content moderation
using AI is not categorically exempted from the First
Amendment's protections. Neither is commentary on politicians
or candidates for office. Speech about politicians and
candidates, in particular, lies at the heart of the First
Amendment and enjoys special protection. Consequently, courts
have readily and correctly overturned laws prescribing false
speech about politicians and candidates. The emergence and use
of AI does not change the core foundational Constitutional
precepts.
Likewise, privacy concerns may arise from obligations
imposed on platforms that host and distribute AI systems.
Requirements or incentives to search users' communications, to
restrict their publication of models, code, and data, to
monitor a report their online activity or to prohibit or
undermine encryption, all increase governmental surveillance
and, in some circumstances may violate the Fourth Amendment. As
the Committee considers legislation addressing criminal uses of
AI, we urge you to ensure that speech, privacy, and other
important civil liberties are protected.
Second, the recently rejected AI moratorium would have
dramatically increased the risk of AI harms including criminal
and fraudulent activity. Similarly, the consolidation of
Federal data is creating enormous risk of AI harms. The
moratorium that was included in versions of the reconciliation
package was sweeping, preempting State laws and local
regulations that regulate AI for 10 years. Although the
moratorium included limited exemptions for some generally
applicable laws, serious questions about the scope and
workability of those exemptions remain. For example, dozens of
States have passed laws regulating deepfake, nonconsensual
intimate imagery often by amending existing statutes to clarify
their application to generative AI.
Similarly, Tennessee's ELVIS Act extends legal protection
to a person's voice including a simulation of the voice. It is
not clear if such laws with their express application or clear
intent to apply to AI qualify as generally applicable.
Moreover, in many instances addressing AI's harm requires
legislating specifically on AI. Establishment of an AI
moratorium will jeopardize these efforts giving bad actors a
blank check. As Ranking Member McBath recognized, 17 Republican
Governors and members of both parties in both chambers of
Congress oppose the moratorium before Congress stripped it from
the reconciliation package in a 99 to one vote in the Senate.
The more immediate concern, consolidation of Federal data
creates a platform for super charged AI driven surveillance.
While data consolidation sharing could potentially improve
governmental operations and limited circumstances, efficiency
should not be elevated over robust protection of our privacy,
otherwise consolidation could risk the creation of vast and
unaccountable surveillance platform, capable of tracking
citizens' activities, movements, and associations. Such a
platform would be readily analyzable by large language models,
machine learning, and other AI systems such as black box
fleecing algorithms used by Federal law enforcement to ingest
governmental data and predict who is likely to commit crimes.
Data consolidation could lead to biometric information
gathered by Federal law enforcement or during air travel, being
readily accessible by other agencies. Records related to
firearms might be accessible across the Federal Government and
IRS data reflecting contributions to organizations like the
ACLU, the NAACP, the NRA, or The Heritage Foundation could be
accessible to Federal law enforcement without meaningful
process. It is essential for Congress to block the reaction of
centralized government dossiers on each of us.
Thank you for the opportunity to testify before this
Subcommittee and I look forward to your questions.
[The prepared statement of Mr. Venzke follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Mr. Biggs. Thank you, Mr. Venzke.
Now, Mr. Redbord, you are recognized for five minutes.
STATEMENT OF ARI REDBORD
Mr. Redbord. Chair Biggs, Ranking Member McBath, the
Members of the Subcommittee, my name is Ari Redbord and it is
an honor to appear before you today on behalf of TRM Labs,
where we work every day with law enforcement, financial
institutions, and national security agencies to detect,
investigate, and prevent illicit activity in the digital asset
ecosystem.
Before joining TRM I spent about 11 years as a Federal
prosecutor at the U.S. Department of Justice and later as an
official in the U.S. Treasury Department's Office of Terrorism
and Financial Intelligence.
In those roles and now at TRM I've seen one truth borne out
time and time again: Criminals are often the earliest adopters
of transformative technology. They were among the first to
weaponize automobiles to move illicit goods across State lines,
adopt pagers and cell phones to coordinate narcotics networks,
utilize encrypted messaging apps to evade surveillance, and
exploit cryptocurrencies to steal and transfer illicit proceeds
at the speed of the internet. Now they are embracing artificial
intelligence.
We are rapidly approaching a world in which the bottleneck
for crime is no longer human coordination, but computational
power. When the marginal cost of launching a scam, phishing
campaign, or extortion attempt approaches zero, the volume of
attacks, and their complexity will increase exponentially.
We are not just seeing more of the same. We are seeing new
types of threats that weren't possible before AI, novel fraud
typologies, hyper-personalized scams, deepfake extortion, and
autonomous laundering. The entire criminal ecosystem is
shifting. That is why today's hearing matters.
We must recognize that in the same way criminals are
leveraging AI to disrupt and deceive law enforcement and
national security agencies must be empowered to use AI to
defend and respond. This is not optional. It is foundational to
preserve public trust and the social contract itself. If
adversaries are deploying large-scale AI-enabled crime with
impunity, and if the public no longer feels that government can
protect them, we risk a breakdown of that trust. The
consequences are not just individual harms; they are systemic
national security-level threats to our institutions and civic
cohesion.
At TRM we see this shift every day. Through Chainabuse, our
public scam reporting platform, we've tracked a 456-percent
rise in AI-enabled scams, which often use deepfake technology,
just in the last year. Ransomware actors are using AI to draft
realistic phishing emails, identify vulnerable targets, and
deploy malware that adapts to evade detection.
On the laundering side we are seeing bad actors use AI to
automate and accelerate illicit money flows. We are also seeing
fully autonomous fraud agents scraping personal data, launching
scam campaigns, and even coordinating laundering operations.
The most disturbing, we're seeing AI used to generate synthetic
child sexual abuse material, fake, but deeply harmful content
traded online and weaponized in sextortion schemes.
The solution to the criminal abuse of AI is not to ban or
stifle the technology; it is to use it and use it wisely. We
must stay a step ahead of illicit actors by leveraging the same
innovations they use for bad, for good.
At TRM we integrate AI at every layer of our platform to
combat crime using machine learning models and behavioral
analytics to flag complex obfuscation techniques, trace illicit
cryptocurrency transactions in real time, and identify emerging
criminal typologies.
We are developing and deploying AI-powered defense agents
at scale to map illicit networks, triage threats, and surface
early warning signs. These operational tools are already being
used in the field to help global law enforcement agencies move
faster, trace complex laundering schemes, and target the
highest-risk activity, often stopping criminal networks in
their tracks before they can cause further harm.
The future of crime will be defined by AI, but so will the
future of enforcement. With the right investment,
collaboration, and technology we can meet this moment. Thank
you again for the opportunity to testify and I look forward to
your questions.
[The prepared statement of Mr. Redbord follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Mr. Biggs. Thank you so much. I appreciate all of you and
your testimony and very, very interesting. I hope we can get to
some deep substance on it today.
I now recognize the gentleman from North Carolina, Mr.
Knott, for five minutes.
Mr. Knott. Thank you, Mr. Chair.
To the witnesses, thank you for coming to testify today.
This is sort of a novel topic, but as each of you have stated
clearly, this is a very important topic and development around
the blockchain technology, AI technology, it is something that
it is only going to become more prevalent and more dominant in
just about every area of our life. That is true here in
America; that is true overseas. It is going to be one of those
paramount moments that is going to be formative. Where were you
before? Where were you after?
To that end, as a former prosecutor myself, when I was
prosecuting, I came on at the very end of the AI-free crime. I
started to see how blockchain and cryptocurrencies and
artificial intelligence was infiltrating every area of the
criminal world.
To that end, Mr. Redbord, as the use of AI and blockchain
technology becomes more common and it becomes more integrated
in every avenue, there is obviously access from domestic actors
that are criminal and international actors. There is really no
border that is recognized. How does that impact the criminal
infrastructure as it were when you look at domestic versus
international crime and that relationship?
Mr. Redbord. Thank you so much for the question. Yes, I
feel like you would uniquely understand this as a former AUSA
as well. We both prosecuted cases in a world where there were
networks of shell companies and hawalas in high-value art and
real eState used to launder funds. Today, as criminal actors
are looking to blockchains and cryptocurrencies we can trace
every transaction on an open public ledger, right? There's no
more bulk cash smuggling. There's TRM to track and trace the
flow of funds.
I think what we see right now is this really interesting
convergence between AI crime, as we're going to discuss today,
and crypto. Really primarily crypto is often the means of value
transfer in these different crimes, right? You see these
deepfake scams where they are scams trying to get
cryptocurrency from investors or users. They are ransomware
actors who are supercharging attacks where cryptocurrency is
the payment.
The significant difference now as we're investigating
crimes as prosecutors and law enforcement is that we can trace
and track every one of those payments on open public ledgers
which allow us to do financial crime investigation, really,
better than we've ever done it before.
Mr. Knott. Is there the ability as you see it to pinpoint
with specificity criminal activity and, I would say more
importantly, going back to the trust that we need in law
enforcement, pinpointing criminal actors? Because it is such a
foggy space for many people. Can you pinpoint the actual
criminal as opposed to just criminal activity?
Mr. Redbord. It's a great question, and absolutely in many
circumstances. What we're doing essentially at TRM is we're
taking that raw blockchain data, right, those alphanumeric
addresses, those crypto wallets, and we're associating them
with real world entities. Oftentimes it's terrorist financiers,
ransomware actors, and sanctions, for example. That allows law
enforcement to then take that data and track and trace the flow
of funds to build out networks.
Mr. Knott. Is there--
Mr. Redbord. Cartels are a great example of that today. I'm
sorry.
Mr. Knott. Is there a risk that cartels, terrorist states
could use legitimate constructs on the blockchain that are
developed here in the United States for their own benefit,
therefore taking advantage of a legitimate structure or a
legitimate software that is developed here?
Mr. Redbord. Absolutely. The real challenge for regulators
and policymakers is how to ensure that lawful users have access
to those types of tools and yet stop bad actors from using
them. To me the answer the U.S. Treasury Department over the
last few years has done a pretty good job on this--target the
bad actors: The North Korean cyber criminals, the ransomware
actors, and the scammers, as opposed to necessarily the lawful
services that they're using.
Mr. Knott. Is there a risk if we are too zealous in the
prosecution? As a prosecutor I am all for strong law
enforcement, but if we are too aggressive on the front end as
this technology is developing, could we stifle domestic
innovation here at home if we are too aggressive in
prosecuting?
Mr. Redbord. Absolutely. It is critical that we continue to
focus on the bad actors in this space which will allow the
lawful ecosystem to grow as opposed to the lawful services that
are being used by bad actors. Absolutely, really the key to all
of this is to stop bad actors from leveraging the technology to
allow this industry and this technology to grow.
Mr. Knott. Then briefly, what can we do in Congress to
ensure that law enforcement has the resources to target the bad
actors with specificity?
Mr. Redbord. That's exactly right. Today what we really
have across the U.S. Government is a cadre of law enforcement
agents that are really true experts, power users of blockchain
intelligence tools. What we really need is that cadre to grow
significantly. As bad actors are leveraging AI, as they're
leveraging blockchain technology, every Federal agent should
have access to tools and the training necessary to sort of meet
this new moment from a technology perspective.
Mr. Knott. Sir, thank you.
Mr. Redbord. Thank you.
Mr. Knott. Other Members, I ran out of time, Mr. Chair. I
yield back.
Mr. Biggs. Thank you. Without objection, I propose that we
have a second round of questions. Seeing none, we will proceed
in that fashion.
Now, I recognize the Ranking Member, Ms. McBath.
Ms. McBath. Thank you, Mr. Chair. I just have to say that--
and just in listening to each and every one of the witnesses, I
am just really amazed at the depth of the use, criminal usage
of AI. Really thank you so much for what you brought to the
table today, but I do want to talk a little bit about facial
recognition.
The Detroit Police Department reportedly conducted 129
facial recognition searches in 2020, and all on African
American people. The following year 95.6 percent of the
searches targeted Black people.
Mr. Venzke, how does the use of AI-enabled facial
recognition comply with the Fourth Amendment's equal protection
principles?
Mr. Venzke. It raises serious concerns. As far as I know
there's not been a clear holding that equal protection
principles are violated by the use of facial recognition
technology, but it certainly raises those concerns because of
the disproportionate impact we've seen that technology have on
protected classes, particularly as you said, Black people, and
in particular Black men.
We've also engaged in civil rights litigation to defend
individuals who have been wrongly identified by facial
recognition technology. To a large extent this is a matter of
process, ensuring that police departments have appropriate
processes in place so that there isn't reliance solely on an
identification made by facial recognition technology to bring
in a suspect that there isn't bias in lineups and things of
that nature.
Because of the certain threats that we've seen here and the
ways that the technology can struggle in real world conditions
we have long stood by that there needs to be a moratorium for
law enforcement uses of facial recognition Thank you.
Ms. McBath. Thank you for that. This is just so interesting
you should say that because even on my phone I have facial
recognition and sometimes it says I don't recognize you and I
am like, well, you know who I am. Of course, there are still
problems with AI and the technology still needs to be advanced.
Mr. Venzke, I am going to ask you another question: What
warrant requirements and limitations should be applied for
facial recognition tools when they are used by law enforcement?
Mr. Venzke. Well, as I said, our overall stance is that law
enforcement should not be deploying the technology at all
because of the underlying foundational issues of how it can
struggle with a variety of protected classes and correctly
identifying people, especially in real world conditions where
lighting may not be ideal or the surveillance footage may be
grainy. That can result in someone who's 8 months pregnant
being apprehended for a crime where there clearly was not a
pregnant person involved.
The use of facial recognition technology may not
necessarily implicate the Fourth Amendment, but as I said, it
raises very serious concerns about perpetual surveillance, the
ability of the Federal Government to identify individuals in
public spaces going about their daily lives without any
recourse, without any judicial oversight. That is ultimately a
policy question for legislators, city councils, and Congress to
step up and regulate.
Ms. McBath. Thank you for that. Dr. Bowne, in your
experience have you find that AI-enabled tools used by law
enforcement are tested and evaluated before they are deployed
to ensure that they are safe and effective?
Dr. Bowne. In my experience in law enforcement, as a
prosecutor, recently as a supervising prosector the tools that
are being used are certainly going to depend on the
jurisdiction that's using them. States, counties, certainly
Federal Government, from my experience in the Department of the
Air Force, law enforcement organizations that are starting to
use these tools are relatively new users.
In the Air Force any AI tool is supposed to go through
rigorous testing and evaluation standards to ensure that it
results in a certain quantified reliability. That's very
challenging to do with some of these tools, particularly when
you're talking about edge cases like African American men when
they are not found frequently in data sets. Those questions are
being asked.
I don't see in my experience the type of rigorous standards
established across the board by regulators. Until that happens
law enforcement is going to try to keep up. It's tempting to do
that, but there's likely to be some gaps there.
Ms. McBath. Thank you.
Mr. Biggs. Thank you. The gentlelady yields. I now
recognize the gentlelady from Florida, Ms. Lee, for five
minutes.
Ms. Lee. Thank you, Mr. Chair.
Welcome to our witnesses today. As a Member of the House's
Bipartisan Task Force on Artificial Intelligence I had the
opportunity to work with Members on both sides of the aisle to
discuss a national approach to artificial intelligence that
would encourage innovation, strengthen our global leadership,
and also confront serious threats including those like you have
been discussing here today that involve criminal misuse of this
technology.
Today's hearing, among other things, highlights one of the
most urgent of those threats, which is the exploitation of
children through AI, from synthetic child abuse materials to
predatory chatbots, to real time location spoofing, we are
seeing criminals use AI to expand both the scale and
sophistication of their crimes. AI can also, we know, be part
of the solution. In cases like Operation Renewed Hope AI helped
Federal agents identify and rescue minor victims who might
otherwise have never been found.
One of the things that we are interested in doing is
ensuring that law enforcement, child protection, nonprofits,
and trusted partners in the private sector have access to
effective responsible AI tools and the legal clarity to use
them. It is about stopping criminals, saving lives, protecting
children, and ensuring that we are doing our part to help
technology be a force for good.
On that subject, Ms. Perumal, I would like to followup with
you. I would like to know what would you recommend Congress
prioritize, to better equip law enforcement with the tools they
need to stay ahead of AI-enabled threats?
Ms. Perumal. Thank you so much for the question. I think a
few things come to mind. One is strengthening public and
private partnerships. The more that we have the opportunity to
share information across industry and government, and the
threats we're seeing, it is incredibly helpful.
Making it easier to share technology and share the
innovation--frankly, it can sometimes be difficult, especially
as small business are trying to figure out how to share that
technology, which makes a delay. As you have a new way that we
can maybe find something like trafficking or better detect AI-
generated harm, it can be difficult to then deploy that. I
think anything that is to improve that public-private
partnership would be incredibly helpful.
Ms. Lee. Are there specific legislative or funding
priorities that you or your clients have identified that you
think would be impactful?
Ms. Perumal. Yes, few things that we see with our clients.
One is there's a big challenge to identity in terms of online
identity. The AI-generated agents can more explicitly scrape,
use websites for fraud. Then on the other side you also see
things like ID fraud where people are using this to hide their
identity and commit online crimes.
That's an area that the industry is generally trying to
adapt and respond to because that's how so much of fraud and
scams and extortion has been carried out.
Another thing that comes to mind is to the earlier point on
innovation, if it's easier to share and collaborate, that would
be incredibly helpful for us in the private sector.
Ms. Lee. I would like to go back to you, Mr. Redbord. You
said something in your remarks that I thought was really
interesting, that AI is also the future of enforcement. I
believe your words were invest, collaborate, and we can be more
effective.
I would like to hear in your view what are the most urgent
risks posed by AI to national security and public safety and
what would you like to add about what we can be doing in
Congress?
Mr. Redbord. Absolutely. Thank you for the question. Really
what we see AI doing today is supercharging criminal activity
that we've seen exist for some time. Now, you don't need
ransomware affiliates because you can have AI agents that are
automatically deploying malware. We're seeing cyber-attacks at
scale by Norther Korea and other types of cyber actors. Then
we're seeing the laundering of the funds that are stolen move
faster than ever before.
As I mentioned in my testimony we've seen a 456-percent
increase from last year in scam activity involving AI. We have
to move as fast as the criminals. When we think about these
issues at TRM, it's how can we move faster? How can we us AI
the same way they're moving funds to track and trace those
funds to ultimately seize them back?
It's--when you ask, it's the tools and the training to
ensure that every single law enforcement and national security
professional have access to the same tools that many cyber
criminals are using today, and obviously the funding necessary
to support that as well as the training.
Ms. Lee. Thank you. Mr. Chair, I yield back.
Mr. Biggs. The gentlelady yields back. I recognize myself
for five minutes. I appreciate the testimony that we have had.
I had a different line of questions for the next round, but
things you have said in your testimony and what I have heard in
the first round makes me want to ask some specific areas.
You don't have a lot of time to respond, so I am going to
ask because I want a quick response from every one of you.
One of you mentioned the ELVIS Act in Tennessee, which
prohibits the simulation of Elvis' voice, basically. That is
how that generated. What about any other person? I am thinking
what if there is a deepfake of any other public figure and you
have that person say something that is pernicious, something
that is bad, something that is politically inflammatory,
whatever, do we have laws in place that would prohibit that or
it's just the Wild, Wild West?
We will start with you, Mr. Redbord.
Mr. Redbord. Slightly more broadly, I would say that we
have laws in place that absolutely cover a lot of these areas,
but we are going to need to add AI to a lot of them. When I
think about these issues, for example, when you talk about
these types of scam activity that are being supercharged by AI,
we have wire fraud statutes to address them.
Mr. Biggs. Right, right. We will get into that, too, but I
am talking specific. This is a specific case. Let's say you
have a public figure and you have them say something that is
totally outrageous, it is totally deepfaked. You can't tell.
The average person can't tell. If that person was--if that
language is attributed to that person elsewhere, you might have
libel. You might have civil claim of libel or defamation of
character, something like that. Does that in your opinion exist
when someone manipulates a deepfake to do something like that?
Mr. Redbord. It would require adding additive measures to
what we have today.
Mr. Biggs. Mr. Venzke?
Mr. Venzke. That sort of speech lies at the core of the
First Amendment's protections. It's a commentary on
politicians. For example, when the--
Mr. Biggs. If it is not commentary on politicians. Let's
say with maliciousness. Maliciousness? With malice. That is the
word I am looking for. With malice you say that Andy Biggs said
X, Y, this. It is just horrible. You put it in The New York
Times and you did it because you wanted to harm me.
Mr. Venzke. If you take, for example, the Republican
National Committee's deepfake about President Biden announcing
a draft for Ukraine, that is commentary on public events. Of
course, existing exceptions to the First Amendment still apply
to AI. That means--
Mr. Biggs. That is what I want to get at. That is what I
wanted to hear from you, whether something like a defamation,
like--
Mr. Venzke. Yes, defamation is subject to--
Mr. Biggs. --which is why I specifically used The New
York Times.
Mr. Venzke. That's exactly right.
Mr. Biggs. Ms. Perumal, same question?
Ms. Perumal. Yes, I am not super familiar with the legal
side, so I can talk more the technology, but I do see that it
is challenging, detection on it.
Mr. Biggs. Right. Thanks. Dr. Bowne?
Dr. Bowne. There's this inherent friction that we see. I
agree with both Mr. Redbord and Mr. Venzke, that what you
described, sir, is likely protected under the First Amendment.
You have--
Mr. Biggs. Unless there is malice.
Dr. Bowne. With malice. Now, there's legal protection
certainly from a civil right of action if it were to be--
Mr. Biggs. Let's not to interrupt. I don't want to be rude
to interrupt, but I do want to interrupt. What my question is--
let's expand it. A true deepfake is so persuasive you can't
tell the difference side-by-side of me over here and the
deepfake. We have had examples of parents and grandparents.
They can't tell the difference between the deepfake and the
voice of the kid. They look the same. They act the same. They
are remarkable. Now, do I have protection, for instance, from
someone doing that?
This is going to lead into the CSAM, which I was hoping I
would have enough time to get to that, because it is the same
type of deal where you see CSAM, which is so persuasive and
sick and disgusting. That is all AI-generated. You mentioned
the one of the Ukraine thing. What remedy does someone have
when they are a victim of this kind of generative AI?
Dr. Bowne. Mr. Chair, if it were to fall under the statute
of like wire fraud--so you look at the intent of what's behind
it. Those are protections there. If it's to create
misinformation, you certainly articulated the challenge and the
potential harm, that may not be outside of the law. There are
gaps in protection when you're facilitating particularly
harmful activity using deepfakes.
Mr. Biggs. Yes, thank you. We will be talking about that in
the future.
That ends the first round of questioning and now for the
second round of questioning I recognize the gentleman from
North Carolina, Mr. Knott.
Mr. Knott. Thank you, Mr. Chair. I have got to say I am
happy to have a second round so soon. I wished more Members
would show up because this is important, but selfishly I am
enjoying the conversation.
All of you have basically indicated what was stated either
directly or indirectly that computational power and ability
will dictate sort of the new criminal landscape. Obviously, AI
will supercharge this. The landscape is going to be forever
changed. I want to ask each of you, how far are we from having
autonomous criminal behavior?
Doctor, start with you. Go in order down the line.
Dr. Bowne. Those are fantastic questions.
Mr. Knott. Thank you.
Dr. Bowne. My assessment is we're there. We have plenty of
autonomously certainly, from a bad actor that is using AI and
using the autonomous features of those models to perpetuate
crimes we're certainly there. The scale, the sophistication,
and the speed that are created by using AI-enabled models,
certainly, from committing scams at scale, targeting
personally, finding cyber vulnerabilities, that is all
happening already.
Ms. Perumal. Yes, I agree. We're definitely seeing that
now. It's the beginning of what's happening. It's being used
for more simple crimes. We see different uses. You might think
of simple bots that text people and send us these annoying scam
text messages. Those are using AI and automate by pulling
breached or leaked data about you. Then, similarly with
computer-use agents, they're starting to be filling out forms.
There's a large opportunity for those to get much more
sophisticated.
Mr. Venzke. Relatedly on the civil side, we're seeing rapid
advancement of artificial intelligence that's used to make
decisions about who can get a loan, who has access to house,
and things of that nature. Often in many cases that will output
a score that humans are largely deferring to. Artificial
intelligence has reached the point where it is having an
outsized effect on our lives, not just because of criminal
activity, but because it affects so many important sectors as
well.
Mr. Knott. Yes.
Mr. Redbord. Thank you for the question. We are there
today, but AI is not dominating criminal activity. That's in
large part why this hearing is so important at this moment, to
start having this conversation. This is really why at TRM over
the last year or so we have focused a lot of our attention.
Particularly how can we build AI tools that enable us to move
faster? Because while we're not there yet, we're getting very,
very close.
Mr. Knott. What is going to be required to make it
responsive to the threat?
Mr. Redbord. That's exactly right.
Mr. Knott. What will be required? I am asking, like--
Mr. Redbord. Oh, sorry.
Mr. Knott. --in terms of capacity, in terms of private
sector investment, in terms of public investment? Give me a
broad picture of what's going to be needed.
Mr. Redbord. It's all of it. I know public-private
partnerships were talked about. It's often this sort of this
right way as something to discuss. Really what it has to be is
the private sector building the tools that government can
ultimately leverage to move as fast as the cyber criminals.
Mr. Knott. In your opinion is the law lagging this new
frontier or are the existing criminal laws sufficient to
protect the marketplace and victims like Mr. Biggs talked
about?
Mr. Redbord. It'll absolutely be a combination of both. It
will be important when you talk about CSAM, which I know we'll
focus on, on ensuring that the Federal sentencing guidelines
meet the AI moment for that. We will have a need for AI-
specific laws, but I will say that a lot of the laws we have
today: Wire fraud, bank fraud, those types of laws, these types
of disinformation investigations, certainly will include AI.
Mr. Knott. Jurisdictionally how do you see AI factoring
into content actions, vehicles designed overseas that penetrate
into the American market? How do we protect against that in a
jurisprudence sense?
Mr. Redbord. It's a challenge. The nature of crypto, the
nature of AI, and the nature of technology, is global and
cross-border. In large part we want to make sure that the
innovation is happening here. That's why it's so important that
as we have these conversations we're walking that line between
stopping bad actors but not stifling innovation in this
critical moment. Just like the internet was born and created in
the United States we need to ensure that is true for AI
technology as well.
Mr. Venzke. If I may add to that representative, as we
think about ways that existing legal frameworks need to adapt
to this rapidly evolving challenge, a multijurisdictional
approach is the right approach, not just at the Federal level,
but also internationally, and of course with States. We've
talked a little bit about the moratorium that was included in
versions of the reconciliation package. The House did exempt
criminal laws. Often, in many cases, civil penalties will be a
necessary complement, for example, in addressing nonconsensual
imagery at high schools--
Mr. Knott. One more question for the panel really quick.
What can parents do to protect their children from this type of
landscape?
Dr. Bowne. As a parent myself, education on the risk is
certainly important. That's something that the public sector
can lead on, law enforcement can lead on, similar to drugs and
tobacco. The risk of AI, whether it's for scams, whether it's
for CSAM, whatever harms, they really impact children as well.
Ms. Perumal. I definitely agree. Especially for the CSAM
harms, giving their children awareness that something like this
might happen to you, it might have nothing to do with anything
you did and here's how you can reach out. Sharing that
awareness because they target the fact that people feel shame
when they have no reason to. I think that would be really
helpful.
Mr. Knott. Thank you.
Mr. Venzke. As a former teacher parents are an integral
part in helping kids navigate the world, and education and
talking with kids about the new risks that are emerging are
critical.
Mr. Knott. Thank you.
Mr. Redbord. Education is absolutely critical. As a parent
of middle school and high school-aged kids I appreciate this.
One more point that I do think is important here though is that
we need to ensure that they also know how to leverage it, not
that they're just afraid of it.
Mr. Knott. Sure.
Mr. Redbord. Their success in large part and in the rest of
their life will depend on their ability to leverage and engage
with this technology. It's absolutely so critical that on the
one hand we protect them against the bad harms, but also ensure
that they're really able to use it and leverage it.
Mr. Knott. Absolutely. Thank you. Mr. Chair, I yield back.
Mr. Biggs. Thank you. The gentleman yields. We now
recognize now the Ranking Member, Ms. McBath.
Ms. McBath. Thank you, Mr. Chair. Once again, as you can
see, we are quite alarmed as to what we are hearing today, so
thank you very much.
I want to go back and touch on what the Chair was just
asking, the question that he asked about. In specific, I would
just--each of you, if you can just tell us in a nutshell that
you are expressing to us that there are gaps. There are gaps in
legislation. There are gaps in things that we need to do here
in Congress to make sure that there are protections that are
enforcing and preventing these kinds of deepfakes and all that
we are talking about today.
Can you give us an idea? Tell us what kinds of legislation
are going to be extremely important for us to put in place to
prevent these kinds of gaps that you just expressed to us that
there are today?
Dr. Bowne, could you start, please?
Dr. Bowne. Yes, ma'am. One of the gaps may be closed soon.
H.R. 1283, which was introduced by this Committee would amend
Title 18 of the U.S.C. 2252(a), which is the statute that
covers child pornography and CSAM. The bill is intended to
amend that statute to include AI-generated CSAM within the
definition and coverage under that criminal statute. There is
the Title 18, the criminal statutes that may need to be amended
both in content on what is covered, what is criminalized, but
also potentially in the sentencing guidelines, as Mr. Redbord
mentioned.
Then, as the Chair explained and in his question there is a
gap as well on the civil side that there might not be a cause
of action for that noncriminal, because even that amendment
that's proposed in H.R. 1283 still has to be constitutional.
There are First Amendment protections even on things that would
normally be objectionable.
Ms. McBath. OK. Thank you. Mr. Redbord, please?
Mr. Redbord. Thank you very much. I provided about five
suggestions in my written testimony, but I'll just focus on one
for purposes of this answer.
It is absolutely critical that agencies at every level have
access to modern investigative capabilities including the
blockchain analytics platforms integrated with AI, media
authentication tools, as we talked about, and AI-enabled
investigative tools. Congress should allocate dedicated funding
to these tools along with specialized training programs.
Ms. McBath. Thank you. Mr. Venzke?
Mr. Venzke. I think awareness of, first, where existing law
can apply to criminal activity is critical in navigating this
space. For example, 2258(a) has been extended and prosecutions
have been brought for simulated CSAM material created regarding
a specific identifiable child using AI technology. I agree with
Mr. Redbord that education, providing training, defense, and
funding for critical infrastructure for schools and others to
educate students and other vulnerable populations about the
threats of AI will be key, and to shore up their own
cybersecurity infrastructure.
Ms. McBath. Mr. Venzke, I want to go back to something that
you did touch on though. You did touch on the moratorium on
State and local AI, which actually failed. That last attempt
failed. The Republican Chair of the House Energy and Commerce
Committee has already vowed to continue to pursue it even as
they acknowledge that Federal legislation setting standards on
AI is still years away.
As we work to develop that legislation what principles or
proposals should we consider?
Mr. Venzke. One thing I would look to, and Representative
Lee already referenced it, was the House AI Task Force final
report from the last Congress. That was a thoughtful sort of
compendium of the various issues around AI regulation at the
Federal level. It recognized that preemption particularly is a
sensitive issue. It doesn't need to be all or nothing, that
there is a range of tools that can be used in adjusting what is
the appropriate area for preemption? What is the program amount
of preemption to ensure that States have significant latitude
to address these harms in a timely manner, which of course is
beneficial for Congress as this Committee looks at what works
and what does not.
Ms. McBath. OK. Thank you. One last question. Dr. Bowne,
how can public agencies use the procurement process to ensure
that the AI systems they want to acquire are safe and
effective?
Dr. Bowne. They certainly articulate what the problem is
and what the need to create a demand signal for private
industries, for R&D, for academia to do the research that is
needed. The procurement system is a fantastic way for Federal
agencies or State agencies to ensure that the standards are
being met and that the capabilities are there and are being
focused on in the development and the research in the public
sector--or in the private sector and in academia.
Ms. McBath. Thank you. Mr. Chair, I have a unanimous
consent request to enter into the record, a statement from
Barry Friedman, Faculty Director of the Policing Project at New
York University School of Law, dated July 16, 2025. Also,
entering into the record a statement for record dated July 16,
2025, from Public Citizen regarding the growing threat of
criminal exploitation through AI. Last, but not least, a
unanimous consent to enter into the record a statement from
Keith Kupferschmid, Chief Executive Officer of the Copyright
Alliance, dated June 13, 2025.
[The information referred to follows:]
Mr. Biggs. Without objection.
Ms. McBath. Thank you. I yield.
Mr. Biggs. Thank you. The Chair now recognizes the
gentleman from California, Mr. Kiley, for his five minutes.
Mr. Kiley. Thank you, Mr. Chair, for calling this hearing.
It is a hugely important topic. We have often seen this arms
race develop between criminals and law enforcement when it
comes to the use of technology where criminals innovate and law
enforcement has to innovate in turn. It is a matter of just
trying to sort of keep up.
With AI it is a totally different ball game in the sense
that the development of new capabilities is happening so
quickly, and the nature of those capabilities is often emergent
and surprises even the people who train the systems. The ways
in which they are being applied is equally unpredictable.
To me it seems that integrating AI into law enforcement
operations, is not just a tool at this point; it is absolutely
essential. It is a tremendous challenge and requires a lot of
expertise. It seems to me that we need to be thinking very
seriously about how we can have coordination and how we can
make cutting-edge tools available to law enforcement across the
country.
I wanted to ask both Mr. Redbord and Ms. Perumal, if I am
saying that correctly, about your thoughts on this. I know that
you have a law enforcement background. I know that you actually
worked at Google, and I believe it was just a couple days ago
that Google announced that its cybersecurity AI platform for
the first time detected and defeated a software vulnerability
in the wild that was known to malevolent actors. Maybe if you
could both address the role of the public and private sector in
meeting this challenge.
Mr. Redbord. Thank you so much for the question. It's
absolutely critical that the public and private sector work
together on this as the question noted.
Look, every Federal law enforcement agency in the U.S.
today is using tools like TRM to track and trace the flow of
funds, to automate tracing when it comes to cryptocurrency, to
leverage AI tools in that respect. The reality is that it's
still a handful of investigators that have this expertise and
training.
As we move from crime on city streets to crime on
blockchains and in cyberspace we're going to need every agent
and investigator, not just Federal, but State and local to have
access to these types of tools and training. As you mentioned,
cyber criminals are now using this more and more and will
eventually be using this at scale. Every agent investigator who
is investigating these cases, tracking them, needs to be moving
as quickly. I would say tools and training primarily.
Then the other piece is really true public-private
partnership where FBI and IRS-CI and HSI and others are working
closely with the private sector to share information to move as
quickly as possible.
Mr. Kiley. Thanks very much.
Ms. Perumal. Thank you for the question. I love that you're
following the vulnerability discovery. That's awesome.
To your point the criminal ecosystem is using this to scale
their offense. We have to use it to scale our defense and make
that more effective if we're going to keep up. There's a lot of
ways we can do that, from better detecting malware, to better
understanding the tools and tactics they're using, better
detecting the scam messages. If we can enable, as I said, the
public-private partnerships, which we keep repeating and if we
can make it easier and much faster to adapt as the criminal
ecosystems adapts, that makes us a lot more effective.
There are so many opportunities. Even if you think about
all the reports of scams that maybe we already have access to
or law enforcement has access to, if we can just go through
that data and find the trends, using AI to speed up and scale
investigations is a way that we can really keep pace with the
criminal ecosystem.
Mr. Redbord. One more thing I would add to that,
historically, I was a prosecutor for a long time, we
investigate specific cases, right? There's an instance of crime
and we need to investigate that specific case. What this
technology really allows us to do is build out networks, to
understand crime typologies, to understand where the threat
actors are, and how they are engaging and what they would
potentially do next. Really, it's an extraordinary moment when
it comes to not just law enforcement, but how to disrupt
adversaries from a national security perspective.
What this technology--and I think you got to this in your
question--really enables is not just the one-off harm that's
been done to an individual, but how do we build out networks of
cartels, fentanyl dealers, and scam networks in Southeast Asia
and elsewhere? This technology I guess connected to blockchain
intelligence, really allows us to do a lot of that today.
Mr. Kiley. Interesting. You can address crime much more
systematically in a more efficient way and more preemptively?
Mr. Redbord. We even see that today in the actions that are
coming from the U.S. Treasury and Department of Justice where
they're going after networks. They're doing civil forfeitures.
There was a very large civil forfeiture complaint filed about 2
weeks by the U.S. Attorney's Office in D.C., against $225
million involved in pig butchering scams. It was a network.
We're seeing them use it today.
Mr. Kiley. Very interesting. Thanks very much. Mr. Chair,
it seems there is a role perhaps for us to play in supporting
these public-private partnerships and in facilitating the
training and access to this knowledge and these resources in
law enforcement across the country. I yield back.
Mr. Biggs. Thank you. The gentleman yields back. I am going
to recognize myself for my last five minutes here. I would
suggest that as we started off, I said this would be the first
of its kind. I meant that we are going to have to keep pushing
this.
A week--let's see, maybe a week ago Elon Musk announced
Grok 4. He talked about artificial intelligence and Grok 4 is
going to have the intelligence--it already has beyond a Ph.D.,
engineering, science, genius, et cetera, an artificial super
intelligence.
We've touched on it lightly here today using different
terms, but my question is at what point do we no longer see
computational decisionmaking with a human first mover and you
have an algorithmically iterative process that essentially--and
we are there in some extent now, but we are not totally there
because human interaction is still the first mover. At some
point it won't be a human that is the first mover anymore; it
will be the algorithm itself.
How long before we get there? How do we get there and
prevent the crime and provide the deterrence that is necessary?
For the hypothetical I give you, how long before adjudicating
whether there is probable cause or not for a search warrant or
an arrest warrant is merely algorithmically sustained as
opposed to having a human make that determination?
With that bizarre question but acknowledging that we are
actually moving so rapidly that we probably thought by 2050 you
would be getting to artificial super intelligence, but it looks
like maybe before 2030 you are going to be in artificial super
intelligence.
Dr. Bowne? Then we will go down the whole panel.
Dr. Bowne. Thank you, Mr. Chair. Certainly, a thought-
provoking exercise. I am glad we are still at the point where
it is an exercise, and not reality, but I recognize that we may
not be far from there.
Before we get to artificial general intelligence, or before
we get to certainly artificial super intelligence--and those
are still theoretical and not a foregone conclusion--there is
this very powerful and very rapidly increasing agentic AI.
Mr. Biggs. Yes.
Dr. Bowne. We start to see some of what you describe
already taking place at, admittedly, lesser extent than what we
would if it were true AGI or ASI. We still have algorithms that
are making decisions on behalf of humans that are able to do so
at speed and often at a skill that it certainly makes it
difficult for the human agent to observe and to be a part of.
It really depends on how much trust, how much authority we
provide those agents, and what those models are; they fine-
tuned to limit that?
I do believe, as you said, Mr. Chair, this is the first of
hopefully many having some of those industries, some of those
private sector companies describe that so that this
Subcommittee and Congress can ensure that they are able to
predict and know, and set up those guard rails if necessary to
limit what you're concerned about.
Mr. Biggs. Ms. Perumal? Since we only have a minute left,
each of you get 20 seconds.
Ms. Perumal. Thank you. Very interesting question. We see
AI agents making simple decisions today. For more complex
decisions, as the models can do more complex reasoning in the
next few years, it really should come down to how important is
the decision, and then what transparency and explainability we
can get from the model and how much human oversight is
necessary? It really depends on how risky that individual
decision is where we should hopefully see it overlap on these
autonomous decisions.
Mr. Venzke. I'm going to build directly on that. The role
of AI is a choice. Laws passed in Texas, the National Security
Memorandum that governs national security uses AI, say that
certain things are off limits for artificial intelligence.
You mentioned probable cause. That strikes me as a core
foundational tenet of due process that should probably be truly
a human activity. That is the choice that we make of who will
be--what will be human, what will be AI, and where will humans
be in the loop?
Mr. Redbord. I've never seen anything move as fast as this
moved in my lifetime. While I don't have a great answer for how
long, it's certainly coming. If I could leave this Committee
with anything today, it's that we need to move as quickly
building the tools, working with this body to provide the right
laws there. As an old school prosecutor, I'm happy with judges
still making decisions around probable cause, but I do think we
really need to ensure that we are using the tools defensively
to meet this moment.
Mr. Biggs. Great. Thank you all for being here. My time is
expired. There is so much more to cover about this. Like I say
it is just the first, and hopefully we will get back together
soon and continue this.
Please feel free to contact me or the Ranking Member. I
assume that is OK. She says that is OK. Because we want to have
a dialog and see where there are holes, where there are gaps.
Let us know. We want to do stuff that is preventive without
being constrained, if that is possible. Thank you.
With that, we are adjourned.
[Whereupon, at 11:30 a.m., the Subcommittee was adjourned.]
All materials submitted for the record by Members of the
Subcommittee on Crime and Federal Government Surveillance can
be found at: https://docs.house.gov/Committee/Calendar/ByEvent
.aspx?EventID=118467.
[all]