▶ 0:22:37You're good. The subcommittee on digital assets, financial technology, and artificial intell artificial intelligence will come to order. Without objection, the chair is authorized to declare a recess of the committee at any time. The hearing is titled Unlocking the Next Generation of AI in the US Financial System for Consumers, Businesses, and Competitiveness. Without objection, all members will have five legislative days with which to submit additional material to the chair for inclusion in the record.
▶ 0:23:05I now recognize myself for four minutes for an opening statement. Artificial intelligence is rapidly changing industries across the world. Few sectors are more prepared and more impacted than financial services. For decades, our financial institutions have been at the forefront of development and deploying AI technology.
▶ 0:23:26From algorithmic trading to machine learning systems used in risk management to combating fraud in better and faster ways, the rise of generative AI represents the next transformative step which could bring new efficiencies, new opportunities, and potential new risks to our financial markets. This subcommittee has demonstrated strong leadership in charting a path forward for transformative technologies including digital assets most recently through the clarity act in the genius act.
▶ 0:23:57Today we turn our attention to artificial intelligence examining how it is being deployed in financial markets and assessing whether the current regulatory framework is prepared to keep pace. The United States has long been a hub for financial innovation, and we must ensure our policies support responsible AI adoption, not stifle it, as seen during the Biden Harris administration.
▶ 0:24:20Recently, the Trump administration released its AI action plan, which emphasizes American AI leadership across all sectors. We must look to how we can enhance American leadership and competitiveness in financial technology. It is essential that regulations strike the right balance, fostering innovation while ensuring investor protection and market integrity.
▶ 0:24:43Our financial regulators must be cognizant of this that this technology is it uses and Congress must provide the clarity to encourage responsible development here at home. It's paramount our markets are not left behind in the global race for AI leadership. We're fortunate to have with us today a panel of esteemed experts who bring a wealth of knowledge and experience in artificial intelligence and its deployment in the financial system.
▶ 0:25:09The information we learned today will build upon our prior work and assist the subcommittee in shaping policies that encourage responsible innovation in financial markets while cementing American AI leadership. I want to thank our witnesses for being with us today and I look forward to today's discussion. I'll now recognize the ranking member of the subcommittee, Mr. Lynch, uh, for four minutes for an opening statement.
▶ 0:25:33Thank you very much, Mr. Chairman, and thank you for your courtesy. I had a vote on a subpoena in another hearing and, uh, that caused my delay, but but thank you, and I want to thank this panel of witnesses for your willingness to come before the committee and help us with our work. Uh this hearing continues this committee's oversight work to examine the use of AI in the financial services and housing sectors.
▶ 0:25:55As evidenced by the final report issued by the bipartisan working group that chairman Hill and I co-chair co-chared last year, financial institutions, fintech companies, and even financial regulators are already deploying AI to maximize operational efficiency and achieve cost reduction across their core functions from personalized customer services and consumer lending to fraud detection and financial crime monitoring.
▶ 0:26:21At the same time, the rapid development of AI based technologies has introduced serious risks into the financial service space. The trans, excuse me, the Treasury Department, the Federal Reserve, and Consumer Financial Protection Bureau have all expressed concerns.
▶ 0:26:36Other financial regulators have repeatedly cautioned that the development of robust and trustworthy artificial intelligence is dependent on our ability to to encourage innovation that maximizes oversight, consumer protection, data privacy, as well as workforce protection and marketplace fairness.
▶ 0:26:55In view of these considerations, I'm concerned by the Trump administration's decision to rescend some of the common sense federal directives that sought to advance the safe and responsible development of artificial intelligence.
▶ 0:27:08This includes the reversal of an executive order directing all federal agencies to enforce existing consumer protection and to develop additional regulatory safeguards against fraud, unintended bias, discrimination, and privacy violations only when absolutely necessary.
▶ 0:27:28President Trump subsequently issued his own AI action plan, a strategy that largely reflects the divergent view that consumer protection stands as an impediment to AI innovation, a position with which I strongly disagree.
▶ 0:27:43Regrettably, the AI action plan undermines state level AI regulation by requiring federal agencies to quote consider a state's regulatory climate close quote when allocating funding and to outstrip federal funds from any state with a regulatory framework that it considers burdensome. I would note that recent legislation to impose a 10-year moratorium on state AI regulation was soundly defeated in the US Senate by an overwhelming bipartisan vote of 99 to1.
▶ 0:28:14The AI action plan similarly impedes federal oversight of AI based systems by directing the Federal Trade Commission to stand down from ongoing investigations that unduly burden AI innovation. It is the fundamental mission of the FTC to protect the American public from unfair or deceptive business practices.
▶ 0:28:34President Trump also recently issued an executive order to limit federal procurement of AI large lang language models to those that are developed with ide ideological neutrality as reported by the independent Brennan Center for justice. Compliance with the order will likely require technology companies to degrade model performance and undermine public trust in their technology.
▶ 0:28:57Not surprisingly, our committee has received regular reports from AI stakeholders who have put AI development on hold over concerns that their integration of fairness metrics will run a foul of the executive order. Innovation will also not be well served by the dismantling of the CFPB, a federal agency that has sought to advance the development of trustworthy AI systems in financial services through consumer protection enforcement actions and regulatory clarity.
▶ 0:29:23In stark contrast to these actions, the work of our bipartisan AI working group stem from a genuine commitment by members on both sides of the aisle to foster AI innovation in the financial services industry while also ensuring that regulators are equipped with the authorities and resources necessary. In closing, to this end, I look forward to continuing to work on a bipartisan basis with Chairman Hill, Chairman Style, Ranking Member Waters, and our committee colleagues to advance our US leadership in this important area. Thank you, Mr.
▶ 0:29:53Chairman, for your courtesy once again, and I yield back the balance of my time.
▶ 0:29:56The gentleman yields back. I now recognize the chairman of the full committee, Mr. Hill, for one minute for an opening statement. Thank you, Chairman Style, and I appreciate our panel being with us. AI is, of course, rapidly changing the way Americans live, work, and engage with our financial system. Last Congress, we explored AI practices at financial services firms, regulators, and supervisors. Congressman Bill Foster and I served on Speaker Johnson and Minority Leader Jeffrey's bipartisan congressional AI task force.
▶ 0:30:26Miss Garcia and Mr. Lynch and I had an excellent work session at MIT labs on their research in AI and financial services as well. In all these efforts we saw demonstrated how AI has the potential to boost efficiency, cut costs and strengthen the tools used to protect consumers from fraud detection to But as with any innovation, they're risks. AI systems have to be trustworthy, fair, and secure.
▶ 0:30:54Today, we'll start that exploration in this Congress on how AI, particularly the emergency capabilities of generative AI, can reshape our financial system. I look forward to the discussion and I yield
▶ 0:31:07Thank you, Mr. Chairman. Uh, today we welcome uh an esteemed panel. We welcome the testimony of Mr. Dr. Dr. David Cox, vice president of AI models at IBM Research. Uh Dr. Christian Laauo, co-founder and president of Dynamo AI. Uh Mr. Matthew Reeseman, director of privacy and data collection at the Center for Information Policy Leadership. Uh and Daniel Gorfine, CEO of Gatka Horizons LLC, as well as Dr.
▶ 0:31:36Nicole Turner Lee, senior fellow and director of the Center for Technology Innovation at the Brookings Institution. We thank each of you for taking your time to be here. Each of you will be recognized for five minutes to give an oral presentation of your testimony. Without objection, your written statements will be made part of the rec of of the record. Uh Dr. Laauo, you're now recognized for five minutes for your oral remarks.
▶ 0:32:00Chairman Style, Ranking Member Lynch, members of the subcommittee and staff, I thank you for inviting me to testify today and I'm honored to participate in discussions focused on advancing AI across the financial services ecosystem. In 2021, I founded Dynamo AI alongside my co-founder and CEO Vikun McGunthan during our PhDs at MIT.
▶ 0:32:23We started off as a small group of researchers deeply interested in AI but also keenly aware that AI risk pose fundamental challenges to real world adoption. Since starting as a small group of PhDs, we quickly found ourselves working handinand with some of the largest financial institutions, the most cutting edge fintech companies as well as regional banks across America to help them navigate compliance and governance challenges posed by AI.
▶ 0:32:50Today we not only provide the security and governance layer for many of the largest deployments of AI and banking we but we are also now proudly backed by 40 of the top 100 US financial institutions and a consortium of community banks across Our mission has always been to help enterprises navigate complex regulatory environments particularly where legal and compliance requirements pose open technology challenges that institutions struggle to solve.
▶ 0:33:20We found that when faced with new technology regulations or internal compliance requirements around new technologies, enterprises are often left paralyzed asking themselves not only how do I comply with these new requirements, but is it even technically possible for us to comply with these new requirements? Nowhere do we see this to be more prevalent than with the struggles of financial institutions striving to adopt AI.
▶ 0:33:46Every day, our team has the ability and opportunity to witness exciting new AI proof of concepts that bring new efficiencies to the bank. But for every new exciting AI proof of concept that we encounter, we also see another AI proof of concept fail to make it into production and deliver meaningful value.
▶ 0:34:04We commonly see that these projects fail not because the underlying AI technology cannot deliver, but rather because financial institutions struggle to answer open questions about managing AI risk in heavily regulated, high impact and consequential environments. Some common and quite frankly sensible questions about AI risk include, what new security and data leakage risks does AI bring to my organization?
▶ 0:34:31As we hand over more autonomy to AI agents, how can we enable those agents to comply with established banking protocol and procedure when carrying out tasks? And what happens when an AI assistant inevitably hallucinates or fabricates facts in its response? While financial institutions often struggle to answer these questions, these are not intractable problems.
▶ 0:34:53At Dynamo, we've worked with a multitude of financial institutions to establish effective AI risk management that accelerates rather than blocks their AI To truly manage these new risks and unleash AI innovation, financial institutions and regulators must embrace technology solutions that can help risk and compliance teams scale their oversight, including controls like AI guard rails, red teaming evaluations, and auditability over AI usage.
▶ 0:35:22Importantly, the AI landscape is evolving at breakneck speed, and regulators and policy makers need to adopt governing frameworks that keep up with the pace of innovation rather than falling behind on new opportunities and risks.
▶ 0:35:36Regulators should expand the use of AI sandboxes as called for in the administration's AI action plan in the bipartisan HR4801 to not only enable technology teams to experiment with AI on high impact use cases but also bring leading evaluation and red teaming technology to rigorously test these experimental AI against real world risks.
▶ 0:36:01Drawing on global examples such as Singapore's successful sandbox programs, the US can strengthen competitiveness while promoting secure and compliant AI across financial institutions and governing bodies that we speak to every day. We're starting to see comprehensive AI risk management take shape. We believe that this will be key to ushering a new and exciting era of advancement for the industry.
▶ 0:36:25On behalf of the entire team at Dynamo AI, thank you for the opportunity to testify and I welcome the opportunity to work with this subcommittee to create a competitive, secure, and compliant AI ecosystem within financial services. Thank you.
▶ 0:36:38Thank you very much. Uh Dr. David Cox, you're now recognized for five minutes.
▶ 0:36:44Chairman Style, Ranking Member Lynch, and distinguished members of this committee. Thank you for the opportunity to testify. My name is David Cox and I serve as the vice president for AI models at IBM research and the IBM director of the MIT IBM Watson AI lab. While the excitement around AI has intensified in recent years, artificial intelligence has fascinated researchers for many decades.
▶ 0:37:08Um to give you a sense of our historical place in this, IBM has been on the vanguard of this ongoing revolution from the very beginning when an IBMER co-authored the proposal for the 1956 workshop uh Dartmouth conference that coined the term artificial intelligence and gave the field its name. IBM also has a long history of helping enterprises, including in the financial sector, use artificial intelligence technologies to unlock business value and doing so in ways that are responsible and engender trust.
▶ 0:37:38Our work spans a broad spectrum from helping to identify appropriate use cases to providing tooling to govern both the development and deployment of AI systems to inventing new technologies for trustworthy AI. From the start, IBM has tested AI internally before offering it to others. An approach we call client zero. By deploying AI internally, we ensure our models are stress tested in the real world environments before they ever reach clients.
▶ 0:38:07For example, by leveraging IBM's Watson X AI and and automation tools, we help to augment the skills of our own workforce by eliminating repetitive tasks. This has helped has enabled uh our employees to focus on more challenging, rewarding and impactful work uh with the time they save. For the financial industry, the implications of AI, particularly generative AI and large language models, are transformative. LM are not merely chat bots.
▶ 0:38:34They are multifunctional tools that can be adapted across a wide range of financial services use cases. These opportunities come with challenges. Large models consume enormous computing resources, raising both costs and concerns about energy use. Transparency is vital. Enterprises and regulators must understand the provenence and quality of the data underlying deployed systems. Security is also paramount as organizations seek to safeguard sensitive information when using cloud-based systems.
▶ 0:39:03But perhaps the greatest risk is hesitation. If industry delays too long, consumers in the US economy may miss out on the early benefits of adoption. Responsible governance is not a break on innovation. It is a mechanism that ensures that innovation can be deployed securely and sustainably in regulated environments. Firms must understand exactly what data underpins their models and be able to audit those systems over time. That's why IBM's enterprise AI efforts are grounded in three principles.
▶ 0:39:34Open, trusted, and secure. Open source AI strengthens transparency, reduces dependence on proprietary vendors, and enhances US competitiveness. Trust is built through transparency in data curation, training processes, and lineage between models and data. And finally, security must be embedded throughout the AI life cycle from data collection to deployment backed by continuous oversight rather than reactive compliance.
▶ 0:40:04With this in mind, I urge policymakers to support open ecosystems to regulate applications rather than technologies in the abstract and to promote transparency requirements that allow enterprises and regulators alike to test models, evaluate accuracy, and understand safeguards. These steps will foster an environment where innovation can thrive responsibly. AI is not about replacing people but augmenting them. It's about empowering professionals and enriching consumers and expanding opportunity.
▶ 0:40:34By embracing openness, insisting on transparency, and embedding security, we can ensure that AI strengthens both the financial system and America's global competitiveness. Thank you, and I look forward to your
▶ 0:40:48Thank you very much. Uh, Mr. Mr. Reeseman, you are now recognized for 5
▶ 0:40:53Thank you, Chairman Style, Ranking Member Lynch, and members of the subcommittee for the opportunity to speak with you today. I am Matthew Reeseman, director of privacy and data policy at the Center for Information Policy Leadership or Sipple, a data and privacy policy think tank within the Huntton law firm whose mission is to advance best practices for the responsible and beneficial use of data.
▶ 0:41:18Simple facilitates constructive engagement between business leaders, data governance experts, regulators, and policy makers around the world. AI plays a critical role in the financial services industry as this committee documented well in the staff report of its bipartisan working group on artificial intelligence.
▶ 0:41:37For years, machine learning has strengthened financial services institutions ability to combat fraud, provide richer and more tailored services to existing customers, and extend services to new ones. More recently, generative AI has boosted productivity across functions, from software development to customer service. And we are now in the early days of Agentic AI, which shows promise for enhancing the experiences of businesses and customers alike.
▶ 0:42:06Potential applications are extensive from streamlining know your customer processes to back office operations like payroll and invoicing to online banking and agentic commerce. AI's use in financial services also carries risks to consumers, institutions, and the financial system. The bipartisan staff report documents these risks well.
▶ 0:42:28Agentic AI may accentuate some risks while at the same time enhancing riskmanagement capabilities in areas such as privacy and cyber security. To secure the advantages of AI within the financial system, we have three broad sets of First, with respect to regulation, pursue a risk-based approach that focuses on the outcomes to be achieved. Avoid overly prescriptive measures.
▶ 0:42:54Build upon existing foundations, including regulations, guidance, and standards that are already in place. When necessary, clarify or adapt their application to emerging technologies. Consider risks and benefits in equal measure. Incentivize organizations to adopt accountable practices. Build trust through meaningful transparency.
▶ 0:43:15Simple has long underscored that these concepts of organizational accountability are central to smart governance of data and Second, enable the responsible use of data for model training and development. To function safely, effectively, and fairly, models must be trained and tested using rich data sets. Regulators should apply data protection principles in ways that ensure the quality of AI systems while preserving individuals privacy.
▶ 0:43:44Privacy-enhancing technologies or PETs can reduce risks associated with the use of personal data. Examples include synthetic data, which is artificial data that mimics the value of real world data, and differential privacy, where random noise is added to data sets to prevent identification of any individual's data. Policymakers should encourage continued research on and use of pets. Third, engage in cooperative dialogue among regulators, technologists, and industry.
▶ 0:44:13As AI continues to evolve rapidly, stakeholders must learn from each other. Fostering such exchanges is the is the heart of Sipple's mission and regulatory sandboxes offer an invaluable avenue for such dialogue. Numerous jurisdictions have established regulatory sandboxes over the past decade from the UK to Singapore to US states like North Carolina and Delaware. We commend steps to promote sandboxes under America's AI action plan.
▶ 0:44:41the proposed bipartisan bicameal unleashing AI innovation and financial services act and other proposed legislation. Simple has published numerous papers on the affformentioned topics and is continuing our research on them. We look forward to the discussion today and to supporting your efforts to secure the benefits of AI and financial services for everyone. Thank you.
▶ 0:45:02Thank you very much. Uh Mr. Gorfin, you're now recognized for five minutes. Thank you, Chairman Style, Ranking Member Lynch, and members of the subcommittee for the opportunity to testify before you today. I am the founder and CEO of Gatka Horizons, an adjunct professor at the Georgetown University Law Center, and the former chief innovation officer at the USCFTC. Today's topic on unlocking the next generation of AI in our financial system is critically important.
▶ 0:45:29The US holds significant competitive and first mover advantages and we must foster continued development through thoughtful policy approaches. AI presents tremendous opportunities to further expand access, lower costs, and increase efficiencies and competitiveness while also enhancing compliance and regulatory oversight. To level set, it's important to recognize that AI and financial services is not new. It's part of a steady progression of automation that began decades ago.
▶ 0:45:59Today, recent advances in generative AI and Agentic AI are creating new possibility possibilities ranging from generating code and new content to planning and executing complex tasks. Responsibly developed AI is indeed already yielding tremendous benefits. AI can provide more accurate and efficient decision-making, detect patterns that traditional approaches would miss, help regulators keep pace with digital markets, and promote financial inclusion by unlocking credit for historically underserved populations.
▶ 0:46:30It is further enhancing customer service, helping to identify patterns of financial crime and market manipulation, and making financial advice more accessible and lower cost for Americans. As with any area of innovation, however, there are risks associated with AI, including the potential for perpetuating bias, relying on poor quality data, failing to operate as expected, and advancing fraud and scams.
▶ 0:46:55The mere speculative potential or fear of future harm, however, should not broadly block development of AI, including by those small firms and community banks seeking to remain competitive in an increasingly digital economy. To this end, as a key guiding principle, I would encourage everyone to assess new AI based models are on their ability to improve off of a highly imperfect status quo.
▶ 0:47:18This principle should apply across AI applications since a singular focus on risk can blind us to the greater benefits as compared to legacy approaches. With respect to existing policy frameworks, the financial services industry is well equipped to manage new technologies and should be a model for other sectors. For decades, a robust technology neutral regulatory framework has governed the adoption of emerging technologies.
▶ 0:47:43For example, consumer protection laws bar discrimination in lending, whether the decision is made by a human or by an algorithm. In our capital markets, rules against fraud and manipulation along with circuit breaker technologies help to mitigate risks related to potential AIdriven trading activity. And our financial regulators have long provided robust riskmanagement guidance, principles, and frameworks that address model, IT, third-party, and enterprise risks.
▶ 0:48:11The overarching framework to ensure safe and responsible adoption of AI is in place. The challenge now is thoughtful application through sound, informed and consistent regulation. To ensure the US maintains its leadership and competitiveness, I accordingly offer five recommendations. First, regulators must support, not block, responsible AI adoption.
▶ 0:48:33We often hear that innovation is blocked by regulators at the examination and supervisory levels due to vague expectations and endless inquiries that lack clear paths to compliance. Examiners should be well-versed in the benefits and risks of AI and provide the marketplace with clear expectations. Second, we must ensure financial regulators have the in-house expertise and tools to keep pace with technology.
▶ 0:48:58A recent GAO report found a significant lack of technology skills at federal financial regulators, which is why codifying innovation offices and equipping regulators with their own AI expertise, tools, and capabilities is essential for effective oversight. Third, Congress should establish a federal uh data privacy framework and ensure open access to quality permissioned financial data. The quality of AI model outputs is inherently tied to the quality of data inputs.
▶ 0:49:26Congress should work to establish a national framework that governs data privacy and ensures that consumers have control over how their data is being shared and used. Fourth, Congress should prevent state laws from undermining federal financial regulatory frameworks. Given the national nature of AI development and its application in the well- reggulated financial services industry, a patchwork of state laws can create conflict, ambiguity, and confusion.
▶ 0:49:52Finally, regulators should further clarify risk management frameworks and promote the value of well-crafted standards. This includes making clear that the mere use of AI, even Gen AI or agentic AI, does not inherently make an activity higher risk. Regulators can further help their efforts to keep pace with technological change by supporting well-crafted industry standards through regulatory rec recognition and safe harbors. Thank you and I'm happy to answer any questions you may have.
▶ 0:50:18Thank you very much. Uh we now recognize uh Nicole Turner Lee for five minutes.
▶ 0:50:24Thank you Chairman Style, Ranking Member Lynch, and distinguished members of the subcommittee. Uh thank you for this invitation to testify on the use of AI in financial sector. Uh my name is Dr. Nicole Turner Lee. I'm a director of the center for technology innovation at the Brookies Institution and I'm also the co-author of a report that was done on AI in the global markets by the CFTC. Artificial intelligence brings a variety of opportunities to financial sector and for years has been used in banking, fraud detection, mortgage applications, credit underwriting and data analytics.
▶ 0:50:53Many of these use cases are promising, often the potential for greater accessibility and improved customer service, while others may introduce challenges, including concerns about racial bias and discrimination. Given that the financial services industry is one of the country's most highly regulated ones, adoption and use of AI requires careful review. Without such oversight, risk abound that can undermine consumers ability to be economically resilient, especially under changing economic conditions.
▶ 0:51:17Further, inaccurate or discriminatory information that is used to train AI models can put marginalized populations at even greater risk, threatening to widen the racial wealth gap, limit home access to home ownership, credit, and other financial transactions. In other words, when algorithms make poor decisions, the quality of life for black and brown communities, seniors, and even some of us are placed in reverse.
▶ 0:51:39For these and other reasons, Congress must continue to foster responsible and ethical use in the financial regulation sector by providing safeguards that protect consumers from AI risk now and reinforce algorithmic accountability, safety, and security, especially when we look at incidents of fraud.
▶ 0:51:57Congress can also safeguard consumers from both the unintended intended unintended consequences through legislation, which starts with comprehensive national privacy standards and reinforcing the jurisdiction of independent federal agencies and state attorneys generals who enforce consumer protection regulations in the age of AI. For example, the former director of the Consumer Financial Protection Bureau clarified that algorithmic decision-making is held to the same standards as human decision-making.
▶ 0:52:25the Department of Justice, the Department of Housing and Urban Development. These agencies also previously leaked documents stating the compliance with many of the provisions that apply to either tenant screening or fairness overall. But recent actions taken to defund and compromise the independence of federal agencies like the CFPB, like the FTC, which have oversight over deceptive consumer practices, can have major implications on protecting consumers from the opaque technology of AI.
▶ 0:52:54State and state attorney generals are critical in enforcing these protections, but challenges to their state rights and laws are not productive. Even with the rejection of the 10-year moratorium on states abilities to develop their own laws, the current AI action plan still suggests that AI related funds should not go to those with burdensome AI regulations.
▶ 0:53:14In the absence of a national framework, states must continue to protect the millions of seniors, children, marginalized populations, and even farmers who are being bared by algorithmic discrimination, deep fakes, and other malicious attacks on their financial well-being. Recent proposals from legislators seek to establish regulatory sandboxes for AI developers where companies can apply for waiverss or modifications.
▶ 0:53:37But without strong regulatory or enforcement consequences, companies will have another avenue to exploit the personal information and behaviors of consumers. As my Brookings colleague Aaron Klein suggests, we should be creating more green houses which allow for more transparency and sunlight into these processes, promoting collaborative partnerships between business, government, and consumers to see where issues arise and how the the three of them can foster trust with one another.
▶ 0:54:04These approaches can also assist in antibbias discrimination, which is also very much imperative to building trust. I'd like to just close with five proposals for members of this committee to consider as we embark on the inquiry today. Ensure that the industry is compliant with existing legal statutes and remedies which reinforce algorithmic accountability. The financial sector is already regulated by numerous federal and state organizations.
▶ 0:54:27Is imperative that it stay that way and that they also comply with acts like the federal the fair credit reporting act and others to seek the prevention of discriminatory results. mandate transparency guidelines and full disclosure for all consumers. Companies need to tell people publicly when they are using AI to make decisions. Those matter output should be explainable and accurate and reliable. Encourage not discourage responsible and ethical development of financial models. Innovation and regulation can be complimementaryary.
▶ 0:54:57And I think that's to the benefit again of those green houses which show transparency in the ideas and processes that we we embrace in this area. Brace for the adoption of Agentic AI, but do the first two. Make sure there's consumer protection in place that we're poised to benefit from the autonomous nature of Agent AI, but we still have firm oversight. And finally, I would say invest in AI financial literacy programs so that consumers know how this industry and sector is evolving.
▶ 0:55:24These and other questions I place before this committee, and I look forward to your questions and
▶ 0:55:29Thank you very much, Dr. Turner Lee. Uh we'll now turn to member questions. I'll recognize myself for five minutes uh for questioning. I want to start with you, Dr. Cox, if I can. Um let's do a real quick stage setting. Um algorithms have been utilized in the financial services space since the 1980s, but we've now seen an explosive investment into AI. Um what is sparking this massive investment in a in a short answer?
▶ 0:55:56Yeah, I I think what is sparking this massive excitement and investment is that this technology is general purpose in a way that previous technologies have not been. So it was a lot of work to train a system before and now that the same system can be used in many different settings
▶ 0:56:12and it can pull through massive data sets in a way that that was not possible in the 1980s.
▶ 0:56:17Although the 1980s as well as today there's already a regulatory framework in place. Fair.
▶ 0:56:22Okay. So, let me let me continue this and I I want to come over to you, Mr. Gorfine, if I can. Um, we're faced with something new. As we focus on kind of risks and threats, the human mind's really good at thinking about what the risks are. It's not as good at thinking about the potential upside. Um, if we think about what the Biden administration did, they really took an attitude to the extreme in his handling of artificial intelligence.
▶ 0:56:46Fortunately, uh, President Trump has reversed course on this, issuing executive order, uh, 14179, removing barriers to American leadership in AI, uh, and releasing America's AI action plan in July of 2025. So, this approach can really accelerate AI development through a try first approach that removes some of the red tape and ownorous regulations. I want to come to you, Mr. Gorfine.
▶ 0:57:10In your opinion, how can the AI action plan or how is the AI action plan an improvement over the previous administration's AI approach?
▶ 0:57:20Thank you. Thank you for the question. Um, I think that what we're seeing right now is an effort to look where there's existing regulation in place and taking an approach of let's allow this to develop so we can actually identify risks and determine whether anything more is needed. Um I think as many of us have outlined the financial services industry is a heavily regulated industry. We have existing laws, regulations and guidance that have been able to adapt and incorporate emerging technologies for decades.
▶ 0:57:50Um there is there are some approaches including around the world that are looking to do things kind of preemptively and in a more prescriptive fashion. uh verse an approach where we say hey we have this scaffolding in place we have principles in place to guide adoption of technologies in a responsible way let's allow this to actually grow and develop and I think it's critically important that the US does remain a the global leader in AI including in the financial services context so the environment seems set to unleash that
▶ 0:58:20thank you very much Mr. Reese I want to build on what Mr. Garfine just said um in particular let's let's dive into uh data privacy laws something that's heavily regulated in the financial services space but with the advent of AI there's additional concerns that are coming online so as financial services uh firms are increasingly deploying uh AI in areas like lending fraud detection customer services uh what role will Americ's personal financial data play and how do we use AI u to implicate existing data
▶ 0:58:50privacy laws and of course the logical A followup would be what should we be looking at from an AI perspective as it relates to data
▶ 0:59:01Thank you very much for this important question and it's been heartening to me to hear how much we've talked about data privacy today because I think what we would say is that a sound data privacy framework is an very important foundation for uh the development of AI and for America's AI leadership. I think data is the raw material on which artificial intelligence depends and uh AI model developers depend on having rich data sets to develop highquality models.
▶ 0:59:29So one thing I think that often gets put into the discussion is the idea of a binary between privacy or innovation and we would absolutely say it's privacy and innovation and privacy as an enabler. So a good data privacy law with modern interpretation holds on to classic principles we've had for decades in privacy law but allows for flexible use of data for purposes such as model training and development.
▶ 0:59:54So building on what you're saying, do you believe that we can regulate and address AI under the current regulatory framework or do you need a new regulatory framework which has been proposed by some?
▶ 1:00:05What I would say is that as a first step we should look at the framework that we have look at it closely and a lot of the elements as has already been discussed today are there and try and be very precise about identifying any lacuna. But first let's look at what we have.
▶ 1:00:20Dr. Cox, you're nodding. you you agree that we can in many ways regulate AI through the existing framework.
▶ 1:00:26I I think certainly the starting place should be what are the risks what are the outcomes that you're trying to to to govern and control and and then any modifications that come from the technology can come from there but starting fresh.
▶ 1:00:39Thank you. Thank you very much. I I I agree. I yield back. I'll now recognize uh the gentleman from Massachusetts, Mr. Lynch, who's also the ranking member of this subcommittee. You're now recognized for 5 minutes.
▶ 1:00:50Thank you, Mr. chairman. Uh this morning a number of us uh uh had a a meeting with Jack Clark, the co-founder and head of policy at Anthropic. Um and and it was it was uh I think enlightening to hear him say how uh for anthropic um that the the investment, the advancements, the deployment of AI had exceeded
▶ 1:01:21by about five years their expectations of of the development and advance of AI. Even based on the projections that they made three years ago, they said we're five five years beyond where we we thought we would be today.
▶ 1:01:37And I I just point to the velocity of change here in this in this sector uh and and what a problem it creates up here in terms of trying to trying to regulate that and and protect markets as well as protecting consumers.
▶ 1:01:52Um, I know a couple of you have mentioned the sandbox model and uh uh Miss Turner Lee uh the Singapore model and we actually had a a congressional delegation from this committee go to s go to Singapore and uh kind of review the model that they had uh for for their sandbox uh back when they were doing fintech a fintech sandbox and uh crypto sandbox.
▶ 1:02:18So uh interestingly enough their their their sandbox has an ethical framework uh which begins with a principlesbased approach uh so that there's there's elements of fairness and accountability and u and it sets an ethical foundation for all AI projects.
▶ 1:02:40It has a bias mitigation element to their their sandbox uh and and mitigates bias in those AI systems and developers are encouraged to assess their algorithms for potential discrimination. Uh there are data uh transparency practices that that must be uh complied with. Organizations must be clear about how data is collected, processed and shared.
▶ 1:03:08uh explanability uh requirements within their their sandbox model. uh so they must assure ensure that their AI systems can provide understandable justifications for their decision making and there's also very strong uh stakeholder engagement encouraging a wide range of stakeholders and and massive user privacy protections that are very important not only for Singapore but but globally and
▶ 1:03:38certainly for the United States as well. And there are regular reviews and audits of that that sandbox process sandbox process uh among these uh participants in in the sandbox as well as blueprints and guidelines and knowledge sharing.
▶ 1:03:56Is that the is that the the the the formula that we should use if we're going to go I know uh the full committee chairman has an idea about sandboxes and and uh we're we're going back and forth here and I about uh you know what that should look like. I I just like to get your your feedback. You've touched on a lot of the sensitive issues um and I'd like to hear you extrapolate a little bit
▶ 1:04:22Thank you so much for that question. I mean I am a fan of sandboxes. I wrote a paper in 2018 when we were just beginning this bubble on what algorithms was going to do when it came to consumers and regulatory sandboxes are a great way to experiment. In fact, here in the United States, we use sand sandboxes to cultivate the fintech marketplace. But because of the velocity of speed in which AI is gathering our personal information, the velocity in which we are going into transformational models where we can't determine the beginning and the end.
▶ 1:04:51When we actually deploy a sandbox model in this day and age, it is important to have many of those variables that the s Singapore government has put into place. It allows for accountability, continuous monitoring, transparency, and it also allows us to ensure that consumers are baked into the process as opposed to providing waiverss and exceptions to companies to experiment and then come back and tell us how it's all going to work, which is the current model from which we exercise right now. when it comes to AI.
▶ 1:05:22Thank you. One one problem that I noticed in the Singapore model was um the number of participants was limited. So it never scaled up. So some of the problems that we see with AI is is when you get that uh scalability and and is there a way for us to you know I know the auditing and review after the sandbox is is ongoing but is there any way we can um sort of make sure that that uh that scalability effect is is implemented in a inside of a sandbox?
▶ 1:05:52Well, I think one of the areas in which many of us who have fought for consumer protections are interested in
▶ 1:05:58the gentleman's time is we'll ask you to to write that for for the record cognizance. No problem of the time. The gentleman yields and the gentleman from Michigan, Mr. Heisenga, who's also the chair, the vice chair of the full committee is recognized for five
▶ 1:06:12Thank you, Chairman Style. Um, Dr. Laauo, we're talking about sandboxes. uh do you have anything I know you've have been commenting on this in the past so do you have do you have anything that additional that thoughts that you would like to
▶ 1:06:27definitely we are uh we are actually participants in the Singapore AI sandbox that they've set up the AI verify foundation and it's actually been working exceedingly well in the sense that these types of uh government agencies they need to bring the latest technologies to actually evaluate the latest risks that are emerging with AI agents and new AI technologies and so by bringing in innovative technologies to do that type of red timing and testing. They're able to stay ahead. Think about here in the US where we have regulatory agencies that are still being educated about the technology.
▶ 1:06:56Whereas in Singapore, they're able to unleash a lot more in terms of AI potential through bringing in new technology to test and evaluate it. So I think there's a lot of benefits there. And also to the point that Congressman Lynch mentioned around scalability, red teaming is a very labor intensive problem, right? So you have thousands of tests that enterprises have to perform. If the uh federal agencies can provide more guidance and tooling around that, it's going to accelerate AI
▶ 1:07:20Okay, that's helpful. Thank you. Um staying with you here, we often hear of an a AI arms race with uh with with China that if lost would threaten US national security and the United States global economic uh dominance. In your opinion, is the United States currently uh uh how are we currently fairing, I guess, in this in this in this arms race? And and are we failing at it? Um if if so, how?
▶ 1:07:50And more importantly, how can Congress and the uh US government at large ensure that the United States outpaces China in the AI
▶ 1:07:58I think there there's been different paradigms emerging here. one is where you have a federal government that can choose and and pick and invest in winners in the space versus creating an open marketplace where different say large language model providers or vendors can compete within this marketplace right and I think what we're seeing when we talk to folks within the DoD or department of war now uh we see actually uh a movement towards opening up these marketplaces where many different vendors can compete and the best solution rises to the top again I'll
▶ 1:08:28I'll move back where actually evaluating which are the best solutions for use cases that is going to drive America's national security interests really comes down to can we do the right evaluation of those large language model providers or vendors and so combining open marketplaces that foster innovation allow different folks to compete and then we get to choose the best ones for our use cases is absolutely essential here for us to be competitive uh in in this arms race you're talking about
▶ 1:08:56by the way I deeply appreciate how you are succinctly wrapping up some very complicated and detailed things and allowing me two minutes yet to ask other questions. So that's uh that's important up here when we only get five minutes. Um uh Mr. Goreine, I I want to talk a little bit about the impact of um of US regulation and how it may threaten the uh the use of AI in financial services uh space.
▶ 1:09:21You know, we had we had had a uh the SEC's predictive data analytics proposal under the Gendler SEC. um I I believe would have had uh uh some significant impact. You are there other things like that that we need to make sure that we're aware of and avoiding.
▶ 1:09:38Yeah. So I I appreciate you raising the predictive data analytics rule. I mean that was rescended and I think that was an example of of a
▶ 1:09:46rightly so but as we saw with that SEC chair, he was expansive in his pursuit of of covering every territory he possibly could.
▶ 1:09:56Right. And I would agree it was it was not technology neutral and that should be kind of a guiding principle. Uh it had incredibly broad
▶ 1:10:03so it should be techneutral
▶ 1:10:05absolutely should be uh technology neutral and importantly like you the soft power too in the way that regulators communicate to the marketplace is important. So with the
▶ 1:10:15oh in other words you don't want to be pounded into the ground and threatened with being put out of business. That that doesn't foster innovation.
▶ 1:10:21Um that's a that can be a challenge.
▶ 1:10:24Allow the sarcasm to be mine. That can certainly be a challenging uh a challenging environment. I think especially when you're talking about smaller firms and when I when I think about regulatory messaging, it's the small firms, it's the community banks that don't have large armies of compliance teams that are able to parse uh certain types of messaging that comes from regulators. So, it can serve as a big deterrent to adoption.
▶ 1:10:47in the last 30 seconds, how can you know these smaller community uh financial you mentioned community banks or credit unions, how can they use AI to compete? Well, so I'm a big believer and I say this in my written testimony in standard setting uh organizations like industrydriven standards that regulators can effectively recognize or even create safe harbors would allow small firms to know that if you're compliant with these best practices or standards that is a from a diligence perspective a route you can pursue and
▶ 1:11:17I think that exploring those types of models in the US is going to be critically important as uh I thank
▶ 1:11:24gentleman yields back um the gentleman from Illinois. Dr. Foster, the ranking member of the financial institution subcommittee is recognized for five
▶ 1:11:31Uh thank you, Mr. Chairman, to our witnesses. Um the I guess Mr. Riceman and maybe others have mentioned the uh the importance of uh privacy enhancement techniques. Um several years ago when I was chairing the AI task force on this committee, we had I dragged in a witness to talk about homorphic encryption and some of the uh differential privacy techniques. Uh, and I noticed just recently, and it was clear back then that these were not ready for prime time.
▶ 1:11:57There was a huge penalty for performance and um and the privacy wasn't actually that great. I noticed recently Google just announced this uh Google Vault Gemma, which sound like they're actually implementing training with differential privacy. And so I was wondering if anyone on the committee could say something about, you know, are these techniques really ready for prime time?
▶ 1:12:18And is there the possibility of a regulatory safe harbor for firms that commit to using um high quality differential privacy and uh type tools? Thank you for the question. It's been quite extraordinary to see the progress on privacy-enhancing technologies over just the last few years. I think you're right that not so long ago, many of them were not ready for prime time, but the curve has been quite steep in terms of many of them becoming much more feasible.
▶ 1:12:45Part of it is that many privacy enhancing technologies are compute intensive, but our computing power collectively is growing a lot stronger. So, we're able to handle that. I think there was also a sense that a lot of them were complex and maybe out of reach, especially for smaller companies. There's a whole ecosystem now of expert companies that are able to consult and offer what you might call off-the-shelf privacy-enhancing technologies to make them much more available, much more broadly. So, um,
▶ 1:13:12it would make sandboxes easier to
▶ 1:13:15Any other sort of comments on the state-of-the-art? Would you anyone disagree that this is something that's pretty promising at this point?
▶ 1:13:23Well, things like fully homorphic encryption, uh, they provide very strong guarantees and you're you're correct that they're slower than conventional computing, but the leaps and bounds that they're making in that technology, it's orders of magnitude, thousand times better. You know, uh, you know, every time I turn around, it's gotten better. So I think that's moving very fast and the hardware is going to ultimately start to close that gap as well. So I think we're going to we're going to benefit from that wave.
▶ 1:13:46Yeah, that could be a key component to a safe sandbox in a lot of these things.
▶ 1:13:51Um in relating to one of the things we struggle with is uh the small bank versus large bank trying to level the playing field we can and you know it's one of the things that AI can potentially help us with. You know soon all of us will have in our pocket the best team of lawyers that's ever been assembled. So it means if you ever get in a fight with a billionaire, a legal fight, you're he's the billionaire will not be able to get a better legal team than you have essentially for free.
▶ 1:14:15Now similarly a small bank should through AI be able to have access to a really good AI risk advisor and a team of risk advisor AIs so that even if it's a pretty small bank this team will know every way that any bank has failed in the history of this country and they just have a tremendous amount of knowledge and make it easier to operate a small bank. Um similarly uh regulations and and uh reg I think is another real advantage.
▶ 1:14:44I I did an interesting experiment couple weeks ago where I said okay Claude or whoever I was using. Um write give me the balance sheets for three banks that have just failed uh for typical reasons. And it just knocked it out of the park. And then I said now give us a resolution plan for each of those three. And it was great. I mean it just said okay in this case set up a bridge bank. in this case try to try to merge it in this case just liquidate it. They very sophisticated things and I I was impressed.
▶ 1:15:13I was wondering if any of you have a a feeling for whether that's really kind of the future of regulation that even small banks have all of their records electronic. If there was a standard interface for the the accounting software that that gets run that could report up to risk management and to the regulators, you could have realtime stress testing against dozens of of scenarios every single night for the smallest bank and that would be a real leveler. Um any thoughts?
▶ 1:15:42Is there reason why things can't evolve that
▶ 1:15:47Dr. Low look.
▶ 1:15:50Thank you. Yes, certainly. So we work with a lot of community banks and regional banks backed by a number of them and we see this every day right so this new AI technology can empower them to automate or streamline a lot of these compliance workflows that are very manually intensive that they just don't have staffing to do but also as you mentioned unlock really new types of auditing and risk control. So 24hour continuous monitoring and observability into certain types of workflows that are highly regulated.
▶ 1:16:19That's all really exciting and and actually being realized materialized today across a lot of these these smaller banks. But I'd say on the flip side, you also want to make sure that as these AI technologies enter in these into these compliant uh regulated workflows, they're also having the right guard rails in place and they're being used for those workflows appropriately in alignment with proper bank policies and procedures. So that's something that actually requires technical expert.
▶ 1:16:43The gentleman's time has expired. We ask you to complete that in writing. Uh the gentleman from Ohio, Mr. Davidson, who's also the chair of the subk national security, elicit finance and inter in international financial institutions is recognized for five minutes. Thank you, chairman. Uh clearly AI represents a transformative force, one that can exponentially enhance productivity, uh bolster our financial systems global edge, and it will no doubt influence our culture. Technology never exists in a vacuum.
▶ 1:17:12And I've long warned about the erosion of our Fourth Amendment protections in the digital age. The Patriot Act, for example, massively expanded domestic surveillance. The Bank Secrecy Act had long ago obliterated any real claim to privacy in your financial dealings. Most state bureaus of motor vehicles are monetizing the personal data citizens are required to provide just to drive or obtain ID.
▶ 1:17:37Now, the government is outright buying data that would otherwise require a warrant or a subpoena, sidest stepping the Fourth Amendment entirely. Because we are serious about fostering innovation, and we are serious about our Constitution, we also need to recognize that AI should serve the American people without turning it into another tool for unchecked surveillance or data exploitation. Mr. Chairman, I'd like to submit this document for the record.
▶ 1:18:03In my recent op-ed uh for the Daily Caller, I emphasize privacy is a foundational layer for ethical AI.
▶ 1:18:10Without objection.
▶ 1:18:12Thank you, chairman. Without it, we're handing the keys over uh to surveillance state on steroids. I mean uh so we need to update the regulatory framework for AI or build it in Congress. And we certainly need to address privacy because that is the data set that AI is using. urgent questions including who's liable if AI you if AI is misused.
▶ 1:18:36Um who who profits and how when someone else's data or intellectual property is indexed or shared with AI? Um when does law enforcement need a warrant if ever? Uh these aren't abstract questions. They're urgent and Congress needs to be proactive, not reactive to set clear boundaries. We can't let AI become another excuse for big brother to pry into our lives. Uh so what safeguards do we need? Uh we should learn from other jurisdictions.
▶ 1:19:04Uh but we should of course chart our own course. The EU's AI act for instance takes an approach with outright bans on real-time biometric surveillance and social scoring measures that align with protecting individual liberties from invasive tech. But in other ways the European Union has become you know pretty Orwellian with some of their privacy approaches and speech limitations. So positively framing uh safeguards around AI is important. Freedom surrendered is rarely reclaimed.
▶ 1:19:33So I'm encouraged by the Trump administration's AI action plan with its emphasis on accelerating innovation, building infrastructure and leading globally. It promotes open source development, cuts red tape, and streamlines permitting without smothering the private sector. Uh so Mr. Mr. Reeseman, how are financial regulators protecting the vast amounts of sensitive financial data that the government already collects from AI uh
▶ 1:20:04Thank you for the question. Um first of all I just want to to your point about raising data privacy and this very important foundation for a sound environment not only for keeping people safe but for innovation and would underscore our our sense that having a sound privacy framework in the United States is actually not only consistent with but important for achieving the goals of the AI action plan. So it's encouraging to see progress in that area.
▶ 1:20:28I I think others may have more experience than me with what's going on inside the agencies for what they're doing to keep data safe. So, I would defer to others. Anyone else?
▶ 1:20:37I I'm I'm happy uh to take that. And you know, I would I would agree that data is the one place screaming for federal legislation um to create a proper baseline for how we secure uh data that will be consumed by AI models. With respect to your question, how what can regulators do? Uh we were just talking about privacy enhancing technologies, things like encrypting data.
▶ 1:21:00Ultimately, financial regulators need to be recognizing that there are these developing privacy-enhancing tools and making sure that regulated financial institutions can use those technologies so data can be encrypted. We can avoid creating honeypotss of information that are being sent every which way. I mean, the reality is today all of our information, driver's license photos have been emailed, sent to every single provider.
▶ 1:21:25there is a better way to move encrypted information and limit access to who actually gets it.
▶ 1:21:31Thank you. I think the computing architecture is really important. I've always liked blockchain technologies. I like uh uh zero knowledge proofs and ways to protect data that way. Um the government in some cases has artificially limited that. But I will submit written questions for the record. I'm really curious how you protect intellectual property, copyrights, trademarks, and maybe monetize that for other people down the way. Maybe blockchain does it, maybe there's other ways. But thank you for your expertise and attention. Thanks for this hearing, chairman. I yield back.
▶ 1:22:00Gentleman yields back. The gentleoman from California, the ranking member of the full committee, Miss Waters, is now recognized for
▶ 1:22:06Thank you very much, Dr. Turner Lee. We have seen the development of new AI technologies like aentic AI which can operate independently and Agent Tech AI is capable of self-directed and complex behavior and can also carry out multi-step tasks.
▶ 1:22:32Financial services companies are now uh and experimenting with Egentic AI in investments, credit decisioning, and more. However, I'm concerned about the new scale of risk and vulnerabilities from all of this. Dr.
▶ 1:22:54Turner, can you elaborate on the potential risks for agent agent AI and whether a liability framework can help us mitigate those risk? Who's responsible for the actions of an AI agent if something goes wrong? Thank you so much congresswoman for that question.
▶ 1:23:15I mean I think much of the conversation we've had today is on this promise of AI as it relates to efficiency in the financial sector and agentic AI is an extension of that. The more and more financial institutions can become more autonomous through chat bots and other tools that allow people to not necessarily talk to a financial counselor but to talk to AI to be able to get through their steps. It makes sense on the productivity side but it doesn't make sense for the consumer. the consumer has a lot of uh reputational uh risk that comes with this, a lot of financial risk.
▶ 1:23:45And in an industry where we know that consumers at the heart of any type of harm or potential harms that come through misadvice or miscalculation, it can have a huge effect, you know, not only on people who are wealthy, but people are experiencing the wealth gap. So, I appreciate your question.
▶ 1:24:00I do think we need a liability risk structure uh for this uh one in which Congress thinks through the same type of consumer protections that we're defunding right now and diminishing even with uh many of the actions today to append uh those regulatory agencies like the FTC and the CFPB. As we move into aic without those guardrails and the ability of states in particular to control how people respond to this new technology, it is actually going to go further than we can catch up with it.
▶ 1:24:28Wow. Um, Dr. Turner Lee, the Republican regulatory sandbox appro proposal for AI we're considering today appears to be just more deregulation framed as innovation and lacks requirements for public disclosure, harm mitigation, and other important protections with an unlimited scope and virtually no limitation putting American consumers and others uh market partitions participants at risk.
▶ 1:24:57I'm deeply concerned that regulatory sandboxes may remove safeguards from a rapidly developing AI market that is already lacking meaningful federal regulations and oversight. In fact, we have yet to fully understand the consequences of deploying this technology without any safeguards and are already dealing with countless uh public safety issues.
▶ 1:25:22Despite the fact that you've alluded to some of this, what risk do regulatory sandboxes pose? If Congress were to consider regulatory sandboxes for AI, what kind of standards should responsible sandboxes meet?
▶ 1:25:39Thank you for that question as well. I mean, I think we've mentioned already the role of uh sandboxes and helping us to cultivate new products and services within this sector and among other sectors. We actually see sandboxes in healthcare. The challenge is without the right variables that we're actually constructing to make sure they're safe, ethical, fair, inclusive, as it's been mentioned, sandboxes will turn out to be uh an exceptional exploitation of consumers.
▶ 1:26:04And what that means is without any guard rails, which the AI action plan is suggesting there should be modifications, there should be waiverss. We really need sandboxes to be very transparent. We need clear goals and questions of what it's trying to solve. We need protections against consumers who are part of those sandboxes to ensure that any harm that they face is, you know, there's retribution for that as well.
▶ 1:26:25Um, in my opinion, we can actually create this without putting people at risk in terms of the end product and we can do it in the light as opposed to in the dark when it comes to companies and government working together on these policy
▶ 1:26:39I'm so pleased that you're here today.
▶ 1:26:40Oh, thank you. And I'm pleased you're here today, too. I'm going to call on you uh to continue the kind of education that's needed with members not only of this committee but this entire Congress and I'm particularly concerned about the possibility uh for wrongdoing uh the possibility for discrimination all of that and I want to learn more about the databases that are being used and how they are you know significant in determining the outcomes. Thank you so very much.
▶ 1:27:10Thank you. The gentleoman yields back. The gentleman from Tennessee, Mr. Rose, is recognized for five minutes.
▶ 1:27:15Thank you, Chairman Style, and thanks to Ranking Member Lynch for holding uh this important hearing, and thank you to all of uh our witnesses for taking time to be with us today. I want to start with you, Dr. Cox. Uh could you please describe the concept of technological singularity and share your perspective on whether you believe AI will achieve singularity and provide an estimate on the possible time frame for this
▶ 1:27:43Um thank you for the question. Um so the idea of a singularity is this we we reach a point where the technology is is able to advance its own progress faster and faster and faster such that it can get sort of a positive feedback loop and then we we suddenly have an explosion of capability. Um one of the things that's interesting that you know I I just dusted off a copy of um the singularity is near by Ray Kerszfile that happened to be on a shelf that I was cleaning up.
▶ 1:28:09Um, what one thing you'll see about futurists is they they often get the shape of things right, but the the the years are are difficult to predict. And I wouldn't hazard to make a guess about, you know, when that's going to happen or if it's going to happen. But I I would just say that I don't think we're anywhere near, as somebody who works with this technology day in and day out, I don't think our biggest risks come from sort of some eclipse of AI is better at everything than humans than humans are.
▶ 1:28:36But certainly our labor market issues that we have to deal with as ta subtasks of a job are displaced, but I don't think we're in imminent danger of anything so extreme as a
▶ 1:28:48Thanks. I appreciate the insight. Dr. Lee, I found an interesting article from the nonprofit uh Cash Essentials that states, quote, "Some AIdriven Systems may treat cash transactions as suspicious, leading to increased scrutiny and regulatory pressure that discourages cash use," unquote. How can we ensure that AI does not discriminate against individuals who use cash?
▶ 1:29:16Additionally, what measures can be taken to prevent AI systems from falsely flagging cash transactions as suspicious, thereby avoiding undue pressure on companies and regulators to favor cashless payments over cash
▶ 1:29:33I do appreciate that question because I think we are seeing based on the regulatory sandbox that we used to create the fintech industry a lot more cash transactions especially among the unbanked or underbanked. To your point, here's where I think AI could be an interesting tool to help us combat fraud. Uh, improved analysis through AI detection systems could actually be helpful there.
▶ 1:29:54Uh, using AI in ways that the AI sort of cleaves with the data about the underbanked or unbanked and combines that with traditional legacy systems could also be helpful. We often talk about AI as just taking people's data, but it also considers other externalities like where you live, what's your zip code is, these other proxies. oftentimes those proxies are for the worse in terms of bias and discrimination, but they could actually also be helpful in helping us understand what the tradition what the new modern economy looks like in banking.
▶ 1:30:23So I would suggest that there are some techniques on the technology side, but there's also room for people like me as a sociologist to come sit at the table and help determine how we do this
▶ 1:30:33We already see uh uh without the benefit of AI, we see cashless or cash transactions being discriminated against in a number of ways. So, it's a very real concern to me. Mr. Riseman, uh, with the increasing prevalence of artificial intelligence, it is clear that AI assisted fraud will likely escalate rapidly, employing novel tactics to deceive consumers.
▶ 1:30:57In my view, it is essential to warn the public about emerging AIdriven fraud schemes as soon as they are identified. How can private companies and industry stakeholders collaborate proactively to anticipate and combat the evolving threat of AI assisted fraud?
▶ 1:31:18Thank you for the question and I share your concern about fraud and I think I've heard a lot of folks say we're in a moment where the defense has to keep up with the offense, right? We want to make sure that what our financial institutions are empowered to have the same tools to detect fraud that the fraudsters are using to advance it. That's most important is that our institutions are whether they're small banks, whether they're large ones, that they have access to that state-of-the-art screening technology.
▶ 1:31:44I think, you know, one thing that I talked about in my opening testimony was the importance of dialogue. And I think making sure that we have spaces whether it is through sandboxes or whether it's through other mechanisms that are set up by the Congress or by the regulatory agencies to constantly be sharing information about these uh advances in technologies and about these fraud capabilities so that we are all collectively working together on
▶ 1:32:09Thank you very much. I appreciate those insights and I yield back.
▶ 1:32:12Gentleman yields back. The gentleman from California, Mr. Licardo is recognized for five minutes. Thank you, Mr. Chair. Uh, thank you all for your testimony u, and for taking the time to help educate us. Uh, myself particular. Um, I know there's a this is a fastmoving area and I really had questions um, talking really primarily Dr. Cox and Dr. Lee. I'd be interested in your views.
▶ 1:32:39Um, we have several concepts that have been presented in legislation here before us and I I wanted to see if there was I'd like to tinker a little bit uh with a couple of these bills u from Chairman Hill the sandbox concept uh and from my colleague uh Congresswoman Person uh the task force concept and combine them in a way if I if we could. I I represent Silicon Valley.
▶ 1:33:07Obviously, we have a lot of concerns or I hear lots of concerns in my neck of the woods about having government involved in regulating the code and the algorithms. Um particularly since government isn't terribly good at it. Um and we'd expect that the technology is evolving so quickly, we won't be good at it. Uh but on the other hand, I think there's a widespread embrace of the notion that we need to be mitigating the worst of the harms.
▶ 1:33:32And so the question would be if if we were to create perhaps even more than a task force but an independent body um of folks in industry uh academics experts, financial regulators and others that were to establish what the best practices are in the industry. The best practices around uh for example curated data sets I know you mentioned Dr.
▶ 1:33:54that IBM utilizes it or testing and reporting or or fraud detection, watermarking, data security, a whole host of best practices in the industry and then hold that up as the standard uh and establish essentially that that is going to be the negligent standard uh or essentially the standard if everyone everyone complies with that standard then you are exempt from liability.
▶ 1:34:19uh if you're not meeting that best practices standard then good luck with the lawyers uh and the What about an approach that would combine those two? And I ask either Dr. Lee or Dr. Cox like to jump in.
▶ 1:34:34So I'll jump in first. Yeah. So I think that that is a tremendous idea that we should consider because we seem to be at the same stalemate every time we talk about these issues. Um there's a technical side of it and then there's this consumer output. And I do think that government regulators are good at the latter, the the consumer protection side of it because guess what? That's the side where our constituents are coming to us and saying something has harmed me. With regards to what you're saying, I would just like to offer to you that we did start that process under the previous administration.
▶ 1:35:03The AI blueprint for a bill of rights was actually a nice uh glide path for getting to the same protections that you're speaking of as well as collaboration. The executive order which was um soon after appended had a lot of that conversation on how do you actually bring different bodies together from various disciplines, industry sectors, government and civil society so that we actually solve this together.
▶ 1:35:25To date, we've seen a lot of that sort of eroded in the new um administration as well as in the AI action plan where we're primarily competing against ourselves when we actually just see China as our only uh uh force of nature. So I would suggest I agree with you. I think there has to be more conversation, more collaboration, more disclosure among the various entities to get to the space that you're talking about and I think a task force would not be a bad idea. That was also recommended under the prior
▶ 1:35:53Thank you, Dr. Dr. Cox.
▶ 1:35:55Um I I I think that there are some interesting things emerging already. So the the ISO 4201 standard for just the entire creation process of a of a large language model is, you know, one example of something emerging. And I think you know that that gives you a little bit of a sense of how these things are are playing out. It's a voluntary standard. You get audited against it and it's a it's some sort of mark of your you know whether you have the proper hygiene there. Now that the that that's different than regulating the algorithm though, right?
▶ 1:36:24Like it's more about regulating processes and controls right
▶ 1:36:28and documentation and auditing
▶ 1:36:31as opposed to saying this this technology this algorithm is worrisome somehow. And I I think that's one thing that I think all many of the panelists have raised is that you know it's um it's nothing intrinsic about the technology. It's about how are you using it always has to be in the context of how it's being used and we have use and risk based regulation. That's that's a normal thing we already have. So combining that with things like cyber security hygiene standards which are now then being transported to the the world of AI.
▶ 1:37:00I think that's I think that's an evolution of what we already have to a good outcome.
▶ 1:37:04Right. Thank you. Thank you both. I
▶ 1:37:07Gentleman yields back. The gentleman from Montana, Mr. Downing, is recognized for five minutes.
▶ 1:37:11Thank you, Mr. Chair, and uh thank you for the witnesses. Uh this is a really exciting topic for me. Um you know, the the thoughts of AI. I came out of technology. Uh the opportunities are are incredibly exciting, but it's also, you know, a little bit scary, you know, on where it goes. And it's uh it's interesting. I was um uh talking to uh Eric Schmidt last year.
▶ 1:37:35uh he wrote a book on AI with the late uh Henry Kissinger and uh he made a comment that I thought was really interesting about being very uh light on how you regulate it so that we're not blown out of the water by our competition. And then he said that something really caught my attention, but you need to be ready to unplug it. And I wasn't exactly sure what that meant, but it was an interesting uh interesting comment that I'm still dwelling on.
▶ 1:38:01But uh as a former regulator, you know, we had to deal with a lot of the issues with artificial intelligence as a tool for industry. Um and uh really what the implications were and and some of the um interesting things we did as an insurance regulator is is you know, we were of the opinion that once you wrote down a rule or a law or something prescriptive, it was probably already stale because this is just evolving so quickly.
▶ 1:38:24And so what we tried to do to give uh to inform industry on how we were looking at it is all the work we did on that committee is was you know was concepts based not non-prescriptive because you you didn't you you wanted to kind of guide how we as regulators were looking at it without saying this is what you have to do and a lot of the you know questions that came up there is you know is the machine biased? Is the machine bad?
▶ 1:38:50is the machine a lot a lot of this you know kind of you know concern about what was coming in it and you know me you know I I I think about the data you know as a as a former researcher I think you know garbage in garbage out you know what's the data you're training it with and and what are the you know the the thought processes on how broad and how deep that data set is and what kind of results you you have but the results coming out of it you know a lot of folks were saying well at least in terms of a regulator uh we need to have causation and not correlation And uh you
▶ 1:39:20know I don't agree. I I think if you have a high statistical probability of getting the same results that you know from from this system you know a lot of our life is based on correlation and not causation. I think you can use that you know as reality or close to reality but then if it's giving you a result you don't like like it's you know uh affecting a protected class if it's something that you don't like that's the public policy decision not machine bad. And then you have the real, you know, honest conversation about that public policy.
▶ 1:39:50And uh, you know, another thing that folks would say on the regulator side very often is, well, we need to be able to look into that black box. And I think about, well, that black box once you have an N equals close to infinity m uh, access neural network. No human can understand that. So what do you what tools do you need to understand that? And you might, this is my opinion, the tool you need to understand that is probably generated through AI. And so you have AI doing that.
▶ 1:40:18And it comes back to the, you know, ancient question of Queens custodian, I dips custody. Who's watching the watchers? You know, and it's it's an interesting problem to come up with. Um, but I'm going to start sorry about uh thank you for indulging me on that. But you know, one of the things that I I think about a lot is uh as a state regulator, states rights are very important to me. And there's been a big um conversation about that. And I strongly favor allowing states to lead on regulation where feasible.
▶ 1:40:44But at the same time, it's essential that this technology has the legal and regulatory flexibility to continue innovating and develop in the United States. So, uh, I'm going to start with Mr. Reeseman here. What role do you think the states should play in regulating AI or are you of the opinion that this is something that, uh, Congress needs to tackle?
▶ 1:41:06Sorry, thank you for the question. I think the first question that we should be asking is do we need more regulation on AI at all? And we've had an interesting discussion about looking at what powers we already have we have whether that's under federal law or under state law that allow us to address a lot of the questions that we may have about AI.
▶ 1:41:25I think that there is one thing that we have seen is for both consumers and for businesses it can be hard when there's a kaleidoscope of different state regulations and so there is a value in having interoperable federal standards. I think there is an interesting question to be explored further about whether there may be particular elements that states need to address as a complement but not
▶ 1:41:46Yeah, thank you on that and in the interest of time going to move on. Uh the Biden administration seemed determined to stifle AI uh in any way it could, focusing more on potential threats than its clear benefits. And I I I do believe the benefits are strong. So I'm going to move to uh Mr. Gorfine, please. Could you describe some of the most concerning aspects of the Biden administration's approach to AI from executive orders to his so-called AI bill of rights, and discuss what the impact would have been had Congress passed what the Biden administration proposed.
▶ 1:42:17The the gentleman's time is expired, but we will ask the witness to provide that for the record. We thank you. The gentleman
▶ 1:42:22And on that, I yield. Thank you, Mr.
▶ 1:42:23The gentleoman from Massachusetts, Miss Presley, is recognized now for five
▶ 1:42:28Thank you. Thank you to our witnesses for joining us uh today for what is really a timely hearing. When I'm in my community in the Massachusetts 7th, people are very animated about the future of artificial intelligence. I hear it all, ranging from fears of AI bias to excitement for innovative opportunities to concerns about implications on the future of work to confusion about what's next. And I'm proud that we are confronting these issues headon today.
▶ 1:42:58September 26 to October 3rd marks Boston AI week. Startups, investors, researchers, and students will all convene to explore the role they play in the evolving AI landscape. Leaders in AI like uh the Mass Technology Leadership Council will host educational and networking events to build on state investments in the next generation of leaders, educators, and workers in this rapidly growing field. I'm proud to represent a district that is trailblazing the development of the AI industry.
▶ 1:43:26In Massachusetts, jobs in the technology sector make up 14% of our labor force compared to 10% nationally. And we know diversity in AI jobs is essential to confront bias. uh to maximize opportunity and to make good decisions that help everyone benefit from these cutting edge technologies. Dr. Turner Lee, I want to focus in on your research about the role diverse teams play in AI development and deployment.
▶ 1:43:53Uh what are some ways diverse teams may be helpful to promote ethical and innovative uses of AI?
▶ 1:44:00Thank you so much for that and I congratulate you on the AI week um in Boston. So I would just say there that it's important to have representative groups for a couple of reasons that we've spoken about today. I think on both sides of this conversation there is this um tremendous excitement for the opportunities and then a a tidbit of fear and depending on who you are more or less right based on what AI can do.
▶ 1:44:23It's important that we have people who have the lived experiences of the variety of impacts that AI can have and why that is important with any other internet technology that we've had is beca because AI particularly in the financial sector is dispersed to be very individual to that person.
▶ 1:44:38So when we talk about data, data that is traumatized or discriminatory that actually takes the account of uh the wealth gap that is experienced by for example black populations where they've been denied credit loans um and other eligibility shows up in the data and when you do not have people who understand that lived experience that experience passes on generationally and the AI tends to then affect their quality of life. Loan denials continue etc.
▶ 1:45:07So I think it's really important to have not only background diversity but diversity of various people from various disciplines, various sectors sitting at the table. How are we building financial sector AI without people who actually understand how community banking works or sit in the uh roles of experienced financial literacy counselors for example. You have to have everybody
▶ 1:45:28Thank you. And Dr. Internally, how um you know, do you believe the federal government has a role to play in this? And what is that to ensure that people from all walks of life, women, people of color, low-income folks, uh can have careers in AI, especially in the financial services sector? Well, I think you pointed out really well when you talked about trying to empower local entrepreneurs. It's important for the federal government to create low benchmarks to entry.
▶ 1:45:53We see on this panel a young man who has his own company that is doing incredible things when it comes to data architecture, AI infrastructure, etc. Giving opportunities for variety of people to participate in this ecosystem is really important and I think the government can actually find ways to incentivize the private sector to be more inclusive of uh diverse founders and entrepreneurs and small businesses that want to participate in this space. I also think the federal government could do more on AI literacy. What does a national AI literacy initiative look like?
▶ 1:46:23So that people know that this is not about being an engineer. This is about actually experiencing this behaviorally and the way that AI is dispersing. This is not going to be just about experiencing on your computer. It's going to actually show up in your refrigerator in other places. And so people need to know that. And I would say the other thing that Congress should do is close the digital divide. We keep talking about AI but we actually have not closed the basic infrastructure issue. We cannot build the compute facilities and data centers without that. Thank you, doctor, for being so uh prescriptive there.
▶ 1:46:52Whether it's a hearing in Congress or a meet up at the Museum of Science, AI is the topic of conversation, and we have the responsibility to ensure that all voices are included because AI works best when it works for all. I yield back.
▶ 1:47:06The gentleoman yields. The gentleman from Iowa, Mr. Nun, is now recognized for five minutes.
▶ 1:47:12Well, thank you, Mr. Chair. I want to thank the panel for being here and chair as a fellow Air Force fellow. It's good to see you in that seat. Well, there uh as we look at what Congress has been trying to grapple with this idea of artificial intelligence, I've only been here two terms. We've talked about it every single year. In fact, we were on the AI task force, something this committee helped lead to be able to address um real solutions.
▶ 1:47:36But I think Washington does a lot of talking and we should be doing a lot more listening particular to practitioners in the field who have been addressing this, combating it, addressing it, and finding new solutions for it. and incorporating your experience into the way that we do policy. As opposed to DC being the one who writes the rules of the road and then expects you to execute them, we should be doing this in collaboration. Mr.
▶ 1:48:00Chair, it's one of the reasons I'm leading a piece of policy called the AI plan act or the artificial intelligence plan act to ensure that we here in Washington are incorporating and getting our own house in order before we start telling the innovators, the pioneers and candidly the defenders how best to do it under a government op. I look back at where we were with cyber security when I was director of cyber security at the NYSE. Many of the solutions were already being provided on the public sector and the private sector when we worked together.
▶ 1:48:27So with that, I want to be able to dive into a couple of issues that have been addressed when it comes to artificial intelligence. First and foremost, the idea of misinformation and how it happens not only in the financial space, but across our government. It's lowered the barrier to entry. It's allowed misinformation to expand at an accelerated rate. We saw it even this week tragically with the death of Charlie Kirk.
▶ 1:48:50Russian bots providing inception for everything from conspiracy theories to misaligning our own communications right here in our own country. So Dr. Christian Laauo, you have been a leader in Dynamo AI. You have made it an effort to really bring this technology on board. I want to ask you what should Congress be doing right now to do those things like combating state sponsored terrorism including information warfare.
▶ 1:49:18Yeah, that's a great question and particularly you mentioned the plan act which you proposed uh which I think is a really great step forward in looking at these different risks how AI can be used for misinformation but also used for different types of financial crimes right uh and threaten uh systemic risk to our system. Uh, one thing that I'd emphasize is that uh, uh, keeping up with the pace of the technology is key and talking to practitioners about the latest risks is absolutely essential to making effective legislation, right?
▶ 1:49:49Uh, one thing that I had an opportunity to talk to your team about is the advent of AI agents and particularly you brought up cyber security risks, right? So AI agents is where you give these systems more autonomy to carry out end-to-end workflows and tasks. A lot of times they can go in many different directions.
▶ 1:50:04But this also uh opens up new risks for agents to potentially exfiltrate data or for uh state actors to leverage agents and actually send what we call prompt injection or jailbreak attacks to actually exfiltrate uh information from protected uh systems in the financial services or elsewhere. So an AI agent can pull an email. That email could have an embedded hidden instruction or prompt injection and that agent then can then be overridden to actually perform malicious tasks like crash a system or excfiltrate sensitive information.
▶ 1:50:33So I think uh the work that you've been done doing on this act as well as talking to leading practitioners is absolutely key to keep up with these risks.
▶ 1:50:41I could not agree with you more that you know the challenge here is the technology is going to move much faster than the policy is able to keep up. So being able to provide a framework versus prescription something AI plan act intends to do but also brings in best practitioners in this. I want to talk uh with you Matt Reeseman you're at the center for information policy leadership. You know you've seen some of these innovations come on board.
▶ 1:51:01Talk to us a little bit about how not only we incorporate them to have a better safeguard for our government and our citizenry but where the innovation is really being pioneered a lot of times by our private sector partners. How should we be incorporating that into how we work here in Washington DC? Thank you for the question. I think uh number one one thing that's been encouraging to us to see is that on the one hand you you have said the technolog is moving quickly.
▶ 1:51:28Responsible actors in the space are continuing to innovate but they're not leaving their values and they're concerned about making trustworthy AI behind. A lot smart businesses in the sector including a lot of the leaders recognize that having trustworthy technology isn't just good, it's good for business because customers, government, other stakeholders have more trust and faith in the technology that we're building that they are building.
▶ 1:51:50There's a lot to be said for uh government setting certain standards around transparency and asking for folks to show how does the technology work at a high level. How are some of those specifications made? I would like to continue to see this level of collaboration in the same way we've done with cyber defense and bringing an AI consortium together of business leaders like yourself of innovators like yourself to be able to do this. I think that starts with the AI plan act. With that, I yield my time.
▶ 1:52:16The gentleman yields. The gentleoman from Colorado, Mrs. Patterson, is now recognized for five minutes.
▶ 1:52:21Thank you, Mr. Chairman, and thank you all for being here today. I'm grateful that you all are hosting this opportunity to talk about such an important issue and want to address Mr. Riceman. you talk about how we need to have it is helpful if we have a regulatory framework at the federal level and we could have some complimentary state laws as well.
▶ 1:52:45But I'm really worried because we worked in a bipartisan way to provide a framework to try to continue to move forward here in Congress. uh we aren't known for being the most uh efficient body uh in in government and our inability to come together to produce a regulatory framework on AI uh is concerning when I look at the risks that are involved. There's great opportunity for efficiency.
▶ 1:53:10I that and a lot of promise in all of the different sectors uh but I'm especially worried in the financial sector and the vulnerabilities that we have. We know that the risks that that we're facing are going to get exponentially worse with AI. Already 50% of fraud has been identified as using AI. 92% of the financial institutions surveyed indicate that fraudsters are using generative AI.
▶ 1:53:37And 44% of financial professors report that deep fakes are used in fraudulent schemes. And so when I think about the impact to our financial system and I what the gaps that we're leaving here, what are the most important steps and I this this is opening up for all of you uh how do we provide uh to to ensure that we don't have the gaps in supporting our smaller financial uh systems.
▶ 1:54:03So our our smaller banks uh making sure that they have access to the technology and how what would you recommend on providing that framework and those guard rails at the federal level. That's a lot to ask so don't feel the pressure but I if anyone wants to add on just some of the key things that we need to be thinking about right now.
▶ 1:54:25I'll start.
▶ 1:54:26Okay. So I definitely think the conversation has centered around really advancing national data privacy standards and I think that conversation as someone has already mentioned will sort of um ring in the beast of data that is actually fueling these systems. So if we can come to some partisan support on that I think that would be helpful.
▶ 1:54:45I also think that approaching um congressman as you said some of these very nuanced areas like deep fakes um data is out there that seniors are thinking that their grandchildren are calling in for money and when a senior's economic viability is compromised that's worse for the whole family. So I think going into those verticals where there is bipartisan agreement is another area. We've seen that with um the take it down act in terms of anything that has to do with any manipulated content I think is a step forward.
▶ 1:55:12But I do also think that we have to re-imagine what it looks like in terms of regulatory framework that has some guidance and guardrails which I'd be happy to share with your office um in more detail.
▶ 1:55:22That's great. I look forward to working with you on this and I you know we have a bill as well that I've worked with Representative Flood on. Um it is the preventing deep fake scams act and this is important because you know our agencies are only able to take on what we're directing them to do and I so when we look at trying to be to change with the times to make sure that they have the support they need to address the the challenges that we're facing now
▶ 1:55:53I can you speak to the greater coordination needed between the regulators the industry and the subject matter experts to develop strategies strategies to protect financial institutions and consumers from fraud using AI.
▶ 1:56:10I sure can if no one else will. I think I think on the deep fake side I think the challenge that we have is that because we have uh multiple uh agencies with jurisdiction over this particular area, we haven't come up with some shared values and goals on one how are we defining this? Um we've had a lot of conversation to your point where we've not necessarily agreed on for on the copyright area um this digital providence.
▶ 1:56:34The same thing I think is actually going to go into the deep fake space where we're looking at extracted video text and audio and trying to determine where is this coming from that's very hard for us to track at this moment but also coming up with some shared uh values and goals across the various agencies that are responsible for uh enforcement over that. I think if we're able to actually get away from some of that fragmentation, we can actually push forward with really good legislation that protects people from
▶ 1:57:01Great. And Mr. Iceman,
▶ 1:57:02yes. I just wanted to say uh to praise in the bill, you know, you've called for specifically bringing together industry and expert stakeholders to talk with regulators and government to try and figure out how to solve these problems together. It's never been more important as the technology moves quickly to have all of these voices in the room to problem solve together.
▶ 1:57:22It seems like this should just we should bring bills to do this for every agency across. Sorry. Thank you. I yield my
▶ 1:57:31The gentleoman yields. The gentleman from South Carolina, Mr. Timonss, is now recognized for five minutes.
▶ 1:57:37Thank you, Mr. Chairman, and thank you to the witnesses for being here today. I'm pleased that the subcommittee is turning its attention to the use of artificial intelligence in financial markets. Like stable coins and market structure, this is a technology that will have a significant impact on how American companies and consumers engage with our evolving financial systems in order for the United States to remain a global leader in innovation. We must establish clear rules of the road that allow the free market to thrive, and we must do so thoughtfully and effectively.
▶ 1:58:04As I continue to meet with industry leaders, I'm con I'm increasingly concerned about the growing number of conflicting state laws related to AI. These laws often vary in scope, definition, and enforcement, creating a complex regulatory environment for businesses that operate across multiple jurisdictions. This patchwork of regulation makes it more difficult for companies to scale, innovate, and remain compliant, while also increasing the risk of inconsistent protections and outcomes for consumers.
▶ 1:58:29Without a unified approach, we risk slowing progress and creating barriers that disadvantage both American businesses and the people they serve. Uh, Mr. Gorfine, states have introduced more than a thousand laws related to the use of artificial intelligence.
▶ 1:58:43From your perspective, what are the most pressing risks of a state-by-state regulatory patchwork for both consumers and So, it's a good question and I want to start by being very clear that when I look at this issue, I'm talking about it within the context of the regulated financial services sector.
▶ 1:59:01Um, there may be issues outside of financial services where states are engaging on AI, but I think that what's really important is to recognize there are existing federal, state laws, there's existing regulation, and there's existing guidance that financial institutions adhere to where a patchwork of state laws can absolutely interfere, conflict, or create ambiguity for financial services firms operating on a national level. Um, I think that's especially clear in the data context.
▶ 1:59:29We have existing GBA data privacy uh laws in place for financial services firms. If you start introducing patchwork of state approaches that can be really problematic. Um the same goes for you know when it comes to risk management there is a very careful riskmanagement framework in place for financial services and again I do worry about interference of state laws there. Um so so that that's how I'm thinking about the interplay of state federal especially in the context of the financial services space.
▶ 1:59:56I'm going to turn that a little bit around and say while this is going to be difficult for financial services companies to comply and to uh work across state lines, let's talk about the AI companies that are trying to create these products. I mean the the Chinese are ahead of us or close to being ahead of us and the businesses that they have that are focused on this do not have this burden.
▶ 2:00:17Would you say that that's an equally difficult Yeah, I mean the broader question for AI is that I do think having a proper federal framework uh that allows us to operate on a national scale makes really good policy market and competitiveness sense. Um, so that is something that I would absolutely encourage because as we're describing, it's not even just a national competition. It's global, right? So these activities transcend borders. Um, and having a coherent kind of federal framework there makes good sense in the AI context.
▶ 2:00:48Thank you for that. While there are many challenges involved in legislating and establishing clear rules of the road for artificial intelligence, there are also powerful opportunities. Artificial intelligence has the potential to equip financial institutions with advanced tools to better serve their clients, improve efficiency, and reduce risk. As I mentioned earlier, one of the most common concerns I hear from both large financial institutions and smaller community banks and credit unions is the burden that regulatory compliance places on their operations. These requirements strain both their financial resources and staffing capacity.
▶ 2:01:17Artificial intelligence offers a promising solution to help automate and streamline compliance processes, allowing firms to redirect time and resources toward innovation and customer service. Dr. Lao, given the significant burden that regulatory compliance places on financial institutions of all sizes, how is artificial intelligence helping firms streamline these processes and manage risk more effectively? Thank you for the question.
▶ 2:01:41Uh definitely we see applying AI to compliance workflows is one of the highest ROI activities that we're seeing banks implement today, including smaller community banks and regional banks, right? And the reason is not just are they able to automate uh more workflows where you have low staffing, but also they open up new opportunities for greater compliance. Think about continuous monitoring, 24-hour audits, etc. Right? But on the flip side, you also need to be able to look at as you're introducing AI to these regulated workflows.
▶ 2:02:09What type of guard rails or controls are put around those AI so that they are able to follow the right policies and procedures when executing those workflows. that actually for many in many cases is the more challenging problem is how do you rein in this this autonomous AI to actually carry out the task effectively and in compliance.
▶ 2:02:27Thank you for that. Emerging technology has the ability to make sure that the US is the center of the global economy for decades to come and we got to get this right. Uh with that, Mr. Chairman, I yield back.
▶ 2:02:37The gentleman yields. Uh the gentleman from New York, Mr. Torres, is now recognized for five minutes.
▶ 2:02:42Thank you, Mr. Chair. AI is the most transformative technology of our time. The rise of generative AI could prove to be as revolutionary as the advent of writing or the advent of the printing press. Whether AI will create a better world or a worse world, no one knows for sure. What we do know for sure is that the world will be radically different from anything we've seen before. There is nothing new about the use of AI in finance. What is new is the use of generative AI, large language models in finance.
▶ 2:03:12So my first question, what are the capabilities in finance that an large language model can perform that legacy AI has been historically unable to perform? Anyone who can answer that question is free to do so.
▶ 2:03:27I can go ahead and jump in. I think uh Dr. Cox actually mentioned at the beginning the reason why there's so much investment in the space is because it's general purpose AI, meaning that it could actually do many different things. So you can prompt it in infinitely different ways uh to carry out certain types of workflows that are very specific to your uh day-to-day all the way to doing things like continuous monitoring etc. Right? So that's the power of the technology is these general purpose AI models.
▶ 2:03:51What's the best new use what's the best new use case in finance?
▶ 2:03:55Yeah, I would say uh I mentioned compliance as one of the very high ROI activities because the manual effort involved in that. I'd say the most mature one that's delivering the highest ROI that we see is around developer productivity. The ability for these AI agents to actually go and build applications from scratch, empowering even we're seeing small community banks leverage these to actually really accelerate their own software development and integrating their infrastructure stack.
▶ 2:04:18Can a can an AI banker outperform a human banker? an AI. So I think one of the biggest challenges to actually having an AI outperform a human banker is to make sure that it complies with the common sense of a human banker, right? And that is actually been challenging when you look at AI agents to give them that type of common sense out of the box even given the vast amounts of data that they've been trained on.
▶ 2:04:42And so building guard rails that a human is trained to follow is something that's still an open challenge, but something we're helping a lot of these banks with today. I know members of Congress are irreplaceable, but I'm wondering if bankers and traders are replaceable by AI. Uh what what is the impact of AI on concrete applications like credit scoring, loan fraud prevention and detection? Like does AI lead to more loan approvals or fewer? Does it lead to more inclusive credit scoring or less inclusive credit scoring? What's the what's the outcome so far?
▶ 2:05:12I you want to Well, I can point to that. I mean I think we thought because of the objective nature of AI that it would be easier to see more loan approvals, more credit uh applications actually
▶ 2:05:23Where did we get this notion that AI is objective? Data comes from the internet.
▶ 2:05:26Well, that's what I was going to say. We
▶ 2:05:27it's a reflection of human nature.
▶ 2:05:29That's right. And because the data that is fueling the AI systems basically comes from us. Uh and particularly in generative AI, it's curated data. It's not necessarily predictive data. It's data that exists on people on the public internet. We're actually seeing in some cases the same results and in other instances worse.
▶ 2:05:44We've seen that also on the housing appraisal side when it comes to home ownership that even if you scrub your entire home of any type of remnant or artifact of you uh the AI will still generate the same results because AI uses other proxies your address your net worth in your community etc. congressman.
▶ 2:06:02So I think when we say that AI is better than humans or a banker can outperform a human probably in time but expeditious uh calculation of how we make these decisions comes at foreclosing on opportunities economically for various consumers.
▶ 2:06:16I mean can AI I guess I was going to ask can AI be harnessed to expand access to capital to expand access to credit so one of my I have real frustrations with traditional credit scoring methodology. Can AI detect new patterns of evaluating
▶ 2:06:34So I I I think that's right. I mean traditional credit scores are highly correlated with protected class characteristics. You know, one thing I would suggest is to consider second look applications of AI. And what that means is if you take an existing decline pool and you run the decline pool through kind of cutting edge genai related um uh underwriting models, you will at worst case result in another decline, but you may actually start pulling some approvals through that process and you mitigate some of your initial
▶ 2:07:04uh concerns around the impact of such models. So I think there are ways to smartly start testing these models in a way that uh upholds fairness. The question for me because inevitably AI is going to have some measure of algorithmic bias.
▶ 2:07:18That's right.
▶ 2:07:19Right. The question for me is not whether it's completely free of bias but whether we can make it less biased than the human alternative. Is that an achievable mission?
▶ 2:07:26Well, I think it's achievable if there's transparency first and foremost that an AI model is making the decision in ter on behalf of a financial uh uh uh on a financial application. Many people do not know that AI is actually being used to make those credit decisions. So you have to start with public's disclosure. I think the second thing to your point is we do need a stack that is able to actually evaluate what the disproportionate impact disparate impact
▶ 2:07:50I'm about to be replaced by AI, but thank you.
▶ 2:07:53The gentleman's uh time is expired. I would like to thank all the witnesses for your testimony today. Uh without objection, all members will have five legislative days to submit additional written questions for the witnesses to the chair. The questions will be forwarded to the witnesses for their response. Witnesses, please respond no later than October 23rd, 2025. And with that, this hearing is adjourned.