Full Committee

both sides leah

yeah he's cool no oh this is the one jesse's no it's different just that one's for reinforcement fix to the yeah no no no i i that's

i i noticed that you were doing a lot of this oh oh glad yes i just did yeah i was dancing too sometimes sometimes oh ok i guess you're yeah i was also doing this alright tom i said tom keep down driving buddy important days Right. The mike's back. Good morning. The Committee on Science, Space and Technology will now come to order. Without objection, the chairs authorized to declare recesses of the committee at any time, pursuant to Committee Rule two F, the House Rule eleven. Recorded votes may be postponed. Today we meet to consider H R ninety-three forty-one, the AI eighty Federal Data Guidelines Act, HR ninety-three sixty-three, the AI Security and Innovation Act, HR twenty-three eighty-five, the Create AI Act, HR fifty-three fifty-one, the NSF AI Education Act of twenty-twenty-five, HR fifty-five eighty-four, the Lift AI Act, HR sixty-four sixty-one, the Read AI Models Act, HR eighty-eight ninety-three, the Protect Protecting consumers from deceptive AI Act. HR ninety-three thirty-three, the AI flaw reporting and security enhancement act. HR ninety-three thirty-four, the workforce for AI trust act and HR ninety-three seventy-two, the data infrastructure energy measurement standards act. Now I now recognize myself for an opening statement. I would like to welcome everyone to this morning's full committee mark-up. Today we will mark up ten bills on a topic that seems to dominate every newspaper, news channel and podcast these days, and that is artificial intelligence or AI. Most of us are already integrating AI into our lives at the present, whether we realize it or not. For example, auto-correct and facial recognition systems on our phones, as well as filtering systems for emails and messages depend on this technology. If you shop online, chances are. products have been recommended using AI based on your past behavior. Prices may be adjusted based on demand, and chat bots can help answer customer service questions. In banking, AI helps detect fraud and provides consumers with spending insights. The United States is a global leader in AI innovation. Nearly sixty percent of the world's most cited AI researchers work at American institutions. And our nation remains the destination of choice for all top ai talent. In addition to talent, the United States benefits from significant investment and infrastructure in this space. In twenty twenty-five alone, venture capital investment in U and the United States AI companies reached approximately a hundred and ninety-four billion, roughly three quarters of all global IAI investment. That does not even account for the enormous investments being made in cloud computing, advanced infrastructure, and data center construction across the country, which dwarfs the Chinese market. Last July, the White House released an AI action plan along with an executive order focused on accelerating innovation, strengthening national security, and establishing federal AI policy priorities. In January, we held a hearing in this room with the White House Office of Sta- of Science and Technology Policy Director, to explore specific actions Congress could take to build on President Trump's AI action plan. In particular, we discussed the role of American innovation in AI and the national security challenges posed by foreign competitors, particularly as the Chinese Communist Party continues to invest heavily in AI capabilities. A key component of the AI action plan in the accompanying executive order was bolstering cyber security. protections for our nation's critical infrastructure. One of the bills we have on the docket today, the AI Security and Innovation Act, supports that goal while maintaining continued US leadership in AI research, development, and evaluation by authorizing the Center for AI Security and Innovation, or CASEI. It's important that CASEI is adequately funded. And I look forward to our committee continuing to have a conversation about how we can achieve that. I see that we have tremendous, excuse me, I see that we have amendments to increase authorization levels for Casey, which I fully support. In fact, I think funding for Casey should be perhaps higher than the proposed amendments given its importance at this time. We should, however, ensure that we authorize Casey in a manner that does not undermine future progress on the bill. And I commit to working with the sponsors to ensure that we get that number right going forward. In total, we will mark up ten bipartisan bills this morning. Members of this committee have worked diligently to address important challenges in the AI space while preserving America's competitive advantage. These bills address a wide range of ac- of issues, from improving access to federal data resources for the next generation of AI systems to equipping Americans with the tools and training necessary to succeed in an AI driven economy. Through a variety of efforts, these bills reinforce partnerships among government, industry and academia, recognizing that innovation does not happen in silos. They also invest in the workforce needed to sustain long-term leadership in this critical field. Taken together, these bills will help advance America's AI ecosystem, expand opportunity, and reinforce our leadership in increasingly competitive world. They provide access to federal computing, data, and resources necessary to propel future innovation and discovery. They also focus on voluntary consensus-based standards to meet complex challenges something this committee is very familiar with. I am very pleased to take up this legislation today and welcome constructive amendments as we work together to make these bills as strong and as a- effective as possible. I now would like to recognize the ranking member for five minutes for an opening Ms. Lofgren, you're recognized.

Uh, thank you, Chairman Babin. We have many bills to get through uh today, so I will be brief. The pieces of lev- legislation before us are focused on the challenges and opportunities presented by the wide-spread adoption of artificial intelligence technology. This committee, as well as last Congress's bipartisan AI task force, have explored in depth both the challenges and opportunities associated with AI. These bills, thoughtfully and address the nuance and complexity around these issues. The first bill on the roster is the AI Ready Data Guidance Act, introduced by Chairman Babin, who I joined as a co-sponsor. I'll speak on that bill when it is called up. The next bill on the roster is the AI Security and Innovation Act, introduced by Representatives Olbermildt, Foushee and Chairman Babin. This critical legislation will statutorily establish the Center for AI Security and Innovation. a bipartisan initiative to measure risks of advanced AI systems, support information exchange, and support AI research and evaluation. The next bill on the roster is the, uh, Create AI Act, which was originally introduced by my friend and former colleague from California, Miss Anna Eshoo, but is now sponsored by another Californian, Mister Olbernoti. The bill was ready for the finish line in the last Congress, and I'm looking forward to finally moving initiative beyond its pilot phase and into full programmatic implementation with the passage of this bill. Then we will consider the NSF Artificial Intelligence Education Act, sponsored by Mister Fong and Miss Salinas. This bill directs NSF to create and maintain AI learning initiatives and workforce training programs including opportunities for students pursuing a focus in AI education. It will also serve to fund the national AI initiative. Moving on to the Literacy and Future Technologies Act, or the LIFT Act, brought to us by Mister Cain and Mister Eimo. This bill authorizes NSF to create AI educational tools, curricula, and teacher development opportunities for grades K through twelve. Miss McBride and Mister Aber- Obernalte introduced our next bill, the Read AI Models Act. There's wide variation in how AI models are documented, which can present challenges to consumers when they seek to make This bill will direct a NIST to develop guidance that will help businesses overcome these challenges. The next bill on the roster is the Protecting Consumers from Deceptive AI Act. This bill will create task forces that will lead to the development of standards that counter malicious uses of generative AI that have been used to defraud people. I thank Representative Foushee and Delegate Morland for their thoughtful legislation. We will then consider the AI Law Incident Reporting and Security Enhancement Act, sponsored by Miss Ross and Mister Herd. NIST currently addresses cyber security vulnerabilities through its operation of the National Vulnerability Database. This bill requires NIST to develop a similar database for AI flaws that present both security and safety risks. Next is my bill, the Workforce for AI Trust Act, which is cosponsored by former Chairman uh Lucas. I will speak more on this bill Lastly, we have Mister Subramanian's Data Infrastructure Energy Measurement and Standards Act. The growth of AI has accelerated the demand for data centers to handle computing, and we know these data centers consume vast quantities of water and energy. But there's a need for new metrics to understand how data centers are using these resources and therefore how to design and build them most efficiently. We're in the middle of an affordability crisis, so this legislation, NIST and DOE to develop best practices and technical standards for measuring data center energy and water use. These are all smart and essential bills that reflect the bipartisan future-looking nature of this committee. I thank all the sponsors of these bills. I appreciate the hard work and collaboration, Mister Chairman, we have a number of amendments. Uh, I would uh admonish uh all my colleagues that you can be just as persuasive while being Uh, and we will uh hear for from uh all of those who wish to be heard. I'd also like to know just as a point of personal uh privilege, uh it is Alan McQuinn's birthday and Albert Hinman's birthday was yesterday. So happy birthday to them. And they worked very hard on this mark-up. They didn't get to celebrate their birthdays. And I also w- want to note that uh tomorrow is Carlos Medina's last day on the staff of the committee he's been in, invaluable resource to us, and we thank him for his service. And I noted also that Mister Gallagher has walked in uh to join our committee, and we welcomed him. We welcomed him on the floor when he was sworn in. Uh, the delegation uh is happy to see him here and uh look forward to working with him. We miss our beloved colleague, Doug LaMalfa, but we know you will be a voice for his part of the country. And with that, Mister Chairman, I I yield back.

Yes, I appreciate it. Don't let it yields back. Thank you very much. Other members are invited to submit opening statements and to the record. Pursuant to notice, I now call up H R ninety-three forty-one, the AI ready federal data guidelines act. The clerk will report the bill.

H R ninety-three forty-one, offered by Mister Babin of Texas, a bill to require the director of the National Institute of

Without objection,

Mm.

the bill is considered as read and open for amendment at any point. Now I recognize myself for five minutes, uh, speaking in support of the bill. I'd like to take a moment to talk about my bill, HR ninety-three forty-one, the AI ready federal data guidelines act. I'd also like to thank the ranking member for agreeing to cosponsor this bill with me. As artificial intelligence becomes increasingly important not only for accessing information but also for analyzing it, the quality of that information matters. Think of it this way, as a dentist, which I was for thirty-seven years, I could often learn a great deal about a person's diet and overall health simply by examining their teeth and gums and oral cavity. That is what I was trained to do. With AI models we also need to understand what we are looking at, whether the information is reliable or unreliable. Federal agencies maintain vast amounts of data and HRI ninety-three forty-one directs NIST to develop voluntary guidelines to help agencies prepare data sets uh for use in training AI models. This legislation will help improve the quality, accessibility and usability of federal data while supporting our national security in preserving America's competitive edge. H R ninety-three forty-one builds on the goals of the AI action plan and will help create a more secure, efficient and innovative,

Uh, Uh, uh thank you, thank you, Chairman, Chairman, for your leadership and uh working with me on developing this important legislation. for your leadership and uh, working with me on developing this important legislation. Training increasingly complex, larger AI models will require more high quality data. But as new streams of publicly available human-generated data sets are becoming scarcer, it important to US competitiveness in AI that we facilitate access to nonsensitive data that already sits uh but is unused. Moreover, as frontier models have evolved, so have their data requirements. Today, training a frontier AI model like GPT-four, Gemini or Claude requires an extremely curated multi-stage pipeline of structured data access processing and compliance. If we're to leverage our resources, it's not enough to make data available, we must make it AI-ready. Now the federal government holds a lot of nonsensitive data and is already required to make it publicly available. However, due to funding constraints, lack of expertise, uh, coordination challenges, federal data can be difficult to unlock. The bill would seek to address part of this challenge by directing the development of voluntary guidance to help agencies make their data assets AI-ready. Would also create pilot programs to facilitate federal AI ready data in specific industries or scientific disciplines. This legislation addresses recommendations that was made from the bipartisan AI task force as well as the National Security Commission on emerging biotechnology. I'd like to thank you, Mister Chairman, for your partnership and I urge uh members of this committee to support this common sense legislation. I know there is one friendly amendment uh that I look forward to support. would yield um the balance of my time to Miss Rivas, uh who I think also wishes to speak on the bill.

You're recognized, Miss Rivas.

Thank you, Mr. Chair. Uh, while I am pleased that this committee is taking up uh bipartisan bills related to AI, I'm disappointed and frankly concerned at the majority's approach to the amendment process. I had two amendments on other bills. The amendments would have helped improve prove AI literacy and accessibility, Accessibility. as well as better understand the impacts data centers have on energy and water use. Accessibility. Not only did the majority refuse to engage on these two essential topics, Did the majority? The majority. But the majority went a step further and threatened to pull the data center's bill, the majority went a step further and threatened to pull the data center's bill if I even offered my amendment on water use. if I even offered my amendment on water use. Mister Chair, I'm not sure if this approach to not even let amendments be offered was your

Yes.

Regardless, I think it's a terrible way to run this committee. Um, I may be new to Congress, but I've served in the California State Assembly and in California we took votes. Some good, some bad, some we win and some we lose. But there was no hiding, no threats to pull bills and no silencing our constituents' concerns. We have a duty here to serve our constituents and represent their voices. I heard. That the chairman will work with me and my colleagues, That's true. allowing an open process for amendments and run this committee

Yeah.

the way it's historically been run in a fair and bipartisan manner. And for the record, here is what my amendments would do. My first amendment would have allowed the funds for the development of multilingual materials and resources that promote AI literacy and education. Like it or not, our constituents speak multiple languages,

Mm-hmm.

yet this triggered the my the majority as DEI. The majority. Uh, my second amendment would allow NIST to better understand the impacts that data centers' energy and water use are having on surrounding communities and their resource demands this would be for all communities not just environmental justice communities like mine the majority threatened

Mm-hmm.

to pull the bill if I even offered this amendment and I think that's absurd. As AI continues to rapidly develop and big tech companies are building bigger and new data centers, we must understand the impact that these centers, be it positive or negative. I will not be offering my two amendments this morning and I yield back.

And I yield back, Mister Chairman.

No woman yield.

I'd like to recognize gentleman from uh Florida.

Chairman. Yes.

Yes, sir, Mister Webster.

I'd yield my time to you. That's true.

Thank you, sir. We encourage robust debate at our markups. And as always, both house and committee rules allow members to offer amendments. With that being said, it should never be assumed that by offering an amendment, this idea will automatically be considered or adopted. I have directed my staff to work very closely with member offices and your staffs, on both sides of the aisle. to ensure that we get these bills right from the get-go. We encourage constant communication around here, and I know that my staff works very hard at making this happen. We listen, we want to hear from you, but frankly not every idea is always going to be - to stick. I can't speak for how other offices work, but I know around here that we're very big on communication and open doors. I wanna talk these things through with courteous dialogue. Thank you.

Build back.

Build back. Any other discussion?

No back.

Thank you. OK. If there is no further discussion on the underlying bill, we will proceed with amendments in the order listed on the roster. First amendment on the roster is the amendment offered by Representative McClain Delaney of Maryland. Uh, the clerk shall report McClain Delaney amendment one hundred.

Amendment two, HR ninety three, forty one.

Without objection, the amendment is considered as read. I recognize

Thank you, Mr. Chairman. I want again by thanking you and the ranking member for their leadership on this issue, which is incredibly important, and by tasking NIST, which is headquartered in my district in Gaithersburg, Maryland, with helping our agencies prepare their data for AI use and their models. It's something I worked on in the last administration while Deputy Student's N- Secretarian Deputy Administrator at uh NTIA Commerce. You see, lacks standards about how we deploy, employ, and evaluate new technologies, our country risks creating real harm. I spent fifteen years as a lawyer and a nonprofit advocating for bumpers and safeguards around our kids and social media. We've all seen how the absence of these bumpers around kids and strong guardrails can allow and facilitate some serious harms, before anyone steps in to stop it. This lesson doesn't stop at social media. AI is no different. and in fact, it's much more amplified in its impact,

In general.

and it is our responsibility to get this right. This bill, as introduced, requires our federal agencies for AI use be de- developed consistent with OMB policies. However, it doesn't require these guidelines to align with the n- long-standing privacy laws that are already governed how federal agencies must handle the private information or PII of Americans. My amendment fixes that.

I I see what you're saying.

Unfortunately, I had a front row seat this Congress to what happens when the government moves fast with federal data and ignores existing laws. I serve on the Ag Committee and I saw it firsthand when some of our farmers, when they applied for farm service loans, had some of their PII and personal financial information breached. Doge operatives gained access to treasury payment records, social security files, IRS tax data, the sensitive information of our farmers, and OPM data covering hundreds of millions of Americans and ways that have been found in court to have violated the Privacy Act of nineteen seventy four. As we all know, data is power, but there must be some safeguards. OPM employees themselves sued their own agency for failing to conduct privacy impact assessments under the E government act of two thousand and two, before collecting confidential information from every civilian federal employee. The Social Security Administration has admitted the Doge and third-party entities access far more data than previously disclosed and that sensitive records have been transferred to a server outside SSA's control. These laws, the Privacy Act, and the E-government Act are not bureaucratic formalities. They represent decades of hard-won consensus about what the government owes the public when it handles private information.

For open.

It impacts trust

Yeah. OK.

in our government and impacts our youth. Open. US competitiveness and personal security. The privacy limits how agencies can collect, maintain and share records about individuals. And the E government act requires agencies to assess privacy risks before launching new information systems. These guardrails exist for exactly that kind of large scale AI data preparation. My amendment simply makes that explicit with the law it requires, that NIST guidelines be consistent with the privacy act and the E government act. It's not a heavy lift, it's a ash floor, and ensures that we help agencies.

Representative yields back. Uh, any other f- any further discussion on this?

Uh, i- if I may, Mister Chairman, Yes, please. I just wanna thank Miss MacLaine Delaney for her work on this. You may, Phil. Um, I think the uh the guidance developed by NIST will be very helpful, and it's a common sense amendment I

I wanna thank Representative McClain-Delaney for offering this amendment. It's a sensible amendment that clarifies the bill's relationship with existing federal privacy requirements. And by ensuring these voluntary guidelines are developed with current privacy laws in mind it strengthens the legislation without changing its underlying purpose. I urge the adoption of the amendment as well. Thank you. Any further discussion? The question is on the adoption of McLean Delaney amendment one. All those in favor say aye.

Aye.

Aye. Those opposed say no. And the opinion of the chair, the ayes have it. And the amendment is adopted. Next. There are no additional amendments. Aye. Aye. Aye. Aye. The ayes have it, the bill is amended, is ordered ordered favorably reported without objection to motion to reconsider is laid on the table.

Recorded.

Recognize the uh the gentleman from Florida.

Has been recorded, right?

On a recorded vote. OK. A roll call has been requested pursuant committee rule two F and rule house rule eleven. Proceedings on this vote will be postponed. So ordered. Pursuant to notice, I now call up H R ninety-three sixty-three, the AI Security and Innovation Act. The clerk will report the bill.

H R ninety-three sixty-three, offered by Mister Obernolte of California.

Without objection, the bill is considered as read and open for amendment at any point. Does any member wish to be recognized to speak on the underlying bill? Representative Obernolte.

Thank you, Mister Chair. And thank you for, uh, to all of our committee members, uh, obviously, artificial intelligence is undoubtedly the foundational, uh, transformational technology of our time, and I am delighted that we are having, uh, this hearing this morning advancing a number of bills that form the foundation of federal government oversight and governance of this technology. Uh, and, uh, it's been a long time getting here and I think the American public, uh, going to share our enthusiasm that that uh this uh we're finally getting these bills across the finish line uh i had the honor of co-chairing the house artificial intelligence task force in the last congress and at the end of that congress we issued our task force report which was uh two hundred and seventy pages and eighty nine different concrete policy recommendations that we felt needed to be uh done to establish federal uh oversight of this Unfortunately, to date we have implemented one of those eighty-nine recommendations, just one, which was the take it down act that the president signed into law last summer. Uh, other than that, all of the diff- uh, policy recommendations remain undone. Uh, and so that's why I'm so happy that we are here today, uh, finally passing, uh, these bills into law. Uh, and I, I want to point out to everyone that the task force report was unanimously approved by all twenty-four of our task force members, twelve Republicans, twelve Democrats, as well as by the speaker and the minority leader, which was required by our charter. I'm hoping that that gives everyone some confidence that AI governance and regulation can be a bipartisan issue. And it needs to be a bipartisan issue because the last thing that we want to do is to create a regulatory framework where the rules governing a major industry in America are subject to change. So that's why I'm so happy to be here uh today talking about a set of bipartisan bills that advance this framework forward. Um, several weeks ago uh I joined Congresswoman Trahan introducing what we called a discussion draft of the uh GAIA, the Great American AI Act, that implements the bulk of the task force recommendations. Uh, and I want to thank uh our bipartisan co-sponsors, uh, Congressman Ho-Ching and Franklin on the Republican side, and uh Congressman Subra-Manyan and Peters on the Democrat side. Uh, it's been a lot of work uh, to get to this point. We've gotten a lot of uh, really constructive feedback on that, and I am delighted to be here uh, today passing through the sections of that discussion draft that fall into uh Science, Space and Technologies Committee jurisdiction. So uh, thank you, Mister Chairman, for your hard work in getting us here. Uh, thank you to Ranking Member Lofgren.

It only yields back. Uh, any other further discussion on this? Mr. Franklin, Florida? Oh, we have a Democrat. I'm sorry.

Yeah, that's all right, I just I just wanted to uh take this opportunity to thank uh, Mister Olbrunelty and Miss Foushee for uh their work on this. This was not surprisingly a challenging bill for bipartisan agreement, but our two sides came together in active discussion and debate to find compromise the way legislating should work. Uh, I'll note today a few Democratic uh members will be offering amendments with sincere intentions to improve the bill and address risks of AI that are of concern to their constituents including child safety. And I just ask that everyone on the committee uh take

Yields back, and I'll recognize the gentleman from Florida.

Thank you, Mister Chairman, I appreciate your and the ranking members' leadership in bringing H uh HR ninety three sixty three the AI security and innovation act for our consideration today. Uh this legislation strengthens America's leadership in artificial intelligence by establishing CASEI the center for AI security and innovation at NIST. It will help maintain America's technological while promoting collaboration among government, industry, and researchers. Importantly, this bill does not create new regulatory authority. Instead, it builds on NIST's trusted role as a standard setting institution and provides the resources necessary to carry out this mission. I was proud to champion this effort during the FY twenty-seven house appropriations process, where I secured funding for the center and helped insure it has the resources needed to strengthen America's AI security and innovation leadership. I

Like to recognize the uh uh Miss Foushee.

Thank you, Mister Chairman, I am proud to support the AI Security and Innovation Act with Congressman Obanolti to authorize the Center for AI Security and Innovation, known as CACI. CACI will play a critical role in strengthening our national capacity to test and evaluate

Representative yields back. Any further discussion?

Thank you.

There's no further discussion on the underlying bill. We'll proceed with the amendments in order listed on

Amendment in the nature of a substitute to H R ninety-three sixty-three, offered by Mister Obernolte of California.

Thank you, Thank you, Mister Chairman. Mister Chairman, this amendment, This amendment, uh, uh, uh, codifies the compromises that we've been making on a bipartisan basis to, uh, put this bill in its final form. Uh, and I'd like to thank you, uh, and the ranking member for your work on this. Uh, this bill codifies CASEI, the Center for AI Security and Innovation, which is the key hub in the, uh, approach that the House AI Task Force is recommending for the regulation of AI. Uh, in the task force report, we recommend, uh, the, what we call, what we call a hub and sp- spoke approach to the regulation of AI at the federal level. The spokes are our sectoral regulators like, uh, the FDA, NHTSA, the FAA. And the hub is a repository for technical standards and evaluation metrics, uh, re- the development of regulatory sandboxes for the testing of potentially malicious AI, and a core hub of technical talent and resource that any of our sectoral regulators can call on, uh, i- w- uh, as they undertake the task of regulating

Representative yields back. Any further discussion? I wanna say, Yeah, I think they might be going. personally I support the amendment and the nature of a substitute for H R ninety-three sixty-three. the A I security and innovation act. The amendment makes technical, conforming and clarifying changes to the bill, and I encourage my colleagues to support the amendment. I would like to Hearing no further discussion on the amendment, we will proceed with amendments to ANS in order listed on the amendment roster. The next amendment on the roster is the amendment offered by Representative Foushee of North Carolina. The clerk shall report to Foushee amendment O eight one.

Thank you, Mister Chairman, this amendment ensures federal AI evaluation work at Casey includes child safety, including potential risks to minors from AI systems and the need for safety by design. I've been meeting with families,

Uh. Uh.

teachers, and young people who have seen these harms and real-world impacts up close. And the message is clear. Congress cannot wait until another child is harmed before we act. This amendment would help save lives by making sure our federal AI safety standards and testing efforts examine the risk children face on these systems and from AI chatbots from the start, not after tragedy occurs. So, Mr. Chairman, I hope

The amendment is withdrawn. The next amendment on the roster is the amendment offered by Representative Foster of Illinois. Clerk shall report Foster amendment O nine five.

Amendment to the amendment in the nature of a substitute to H R ninety three sixty three,

Without objection,

offered

the amendment is considered as read.

Uh, thank you, Mister Chairman, and thank you to Mister Obernolte and Mrs. Fuscheen for introducing this bill. Uh, NIST has become one of the hidden gems of the United States government and one of the crown jewels of our federal scientific enterprises. In fact, if you ask ask representatives uh from the other free democracies of the world, which of the federal agencies retain their residual respect, you know, the list often comes down to two, the Federal Reserve and NIST. Uh, the folks at NIST in the office at this legislation's renaming the Center for AI Security and Innovation are helping to set the fundamental standards for understanding of how AI systems work how they're constructed and how they interact with both the world and with each other. Uh, those paying attention to the rapid development of artificial intelligence know that we're currently experiencing the rise of what's called " agentic AI". AI agents can act on behalf of the human to perform tasks. They can write code, they can build a website, even purchase things, financial decisions on behalf of a human. Now in the near future, in fact, many expect that the most financial and commercial transactions will take place not between humans, between but between agents operating on their behalf. Uh, there are many aspects of how agents interact with each other and with the outside world, um, that need to be better understood, and like how agents communicate with each other, and in particular, um, how the identity and authorization of an agent is transmitted down Yes. Mm-hmm. Mm-hmm.

Now I'd like to recognize the uh ranking member for California.

Uh, thank you, Mister Chairman. I wanna thank Mister Foster for this amendment. Uh, advances in AI are enabling the creation of sophisticated agentic systems with increasing autonomy and their capacity, is increasing at an astonishing rate. Um, as Mister Foster's indicated, NIST is already looking at this, but I think having this amendment with giving them explicit authority to do so, is an important uh improvement in the bill. Uh, i- it im- ensures that uh the Casey and NIST lab work will uh advance voluntary standards in best practices and I think that is important. Uh, voluntary standards can be changed.

Gentlemen of yields, like to recognize the gentlemen from California.

Thank you, Mr. Chairman, I'd like to thank Doctor Foster for his thoughtful amendment. Uh, he was a very valuable part of the House AI task force in the last Congress. Uh, This amendment ensures development of agentic AI standards and uh it's the feeling of many that agentic AI will be the next big revolution in AI development and deployment. I think this is a very worthwhile addition to the bill uh and I urge its adoption. I yield back.

Gentleman yields back. I'm pleased to support this amendment as well. It makes a targeted improvement to the bill by ensuring that it keeps pace with emerging AI technologies and includes more advanced AI systems in its testing and evaluation efforts. I urge my colleagues to support the amendment.

It's all done.

Any further discussion? All the hearing none, uh the uh question is on the adoption of foster amendment O nine five. All those in favor say aye. Aye. Those opposed say no. In the opinion of the chair of the ISAB it, the amendment is adopted. The next amendment on the roster is the amendment offered by Representative Fuchsia of North

Amendment to the amendment in the nature of a substitute to H R ninety three sixty three, offered by Miss Foushee of North Carolina.

Without objection, the amendment is offered is considered as read, I recognize Representative Foushee of North Carolina for five minutes to explain the amendment.

Thank you, Mister Chairman, my amendment ensures federal AI testing and evaluation efforts include explainability and interpretability, so we can better

Thank you very much.

Um.

I'd like to recognize the ranking member of California.

Thank you, uh, Mister Chairman, I want to thank uh Representative Foushee for this amendment. Uh, it directs the Center uh for AI Security and Innovation to research methods, uh, advance explainability and interpretability um of AI systems that that would help insure a AI systems we develop are secure reliable and trustworthy, without mentioning a name. A very prominent developer of AI told me in a private conversation he couldn't fully explain how his uh AI system worked. Um so I think tasking NIST with this uh may advance our understanding and

I'd like to recognize the gentleman from California.

Thank you, Mister Chairman, I'd like to thank Representative Fusche for her amendment. She was also a valuable member of the House AI task force in the last Congress. Uh, I think this amendment, uh, is very meritorious. One of the biggest remaining research problems in AI is what computer scientists call the explainability problem. problem, which, uh, it stems from the fact that once we've trained a model, if you ask it a question, it gives you an answer. It can't tell you why it reached that conclusion. And understanding why a model is making a recommendation would, uh, really enable us to better utilize AI to know the, uh, the, uh, some of the pitfalls and the sh- and the shortcomings that, uh, model recommendations can contain. So, uh, I think it's, uh, very worthwhile to have, uh, Casey also study this issue and, uh, I urge the,

Gentleman yields back. This amendment builds on the uh builds on the build by expanding research into AI explainability and interpretabil uh interpretability. A better understanding of how AI systems operate and reach decisions will help support the development of more reliable and trustworthy AI technologies. I support the amendment and encourage its adoption. Is there any further discussion? Hearing none, the question is on adoption of Fuchsia Amendment O eight three. Those in favor say aye. Uh, those opposed say no, the opinion of the chair, the ayes have it, and the amendment is adopted. The next amendment on the roster is the amendment offered by Representative Obernolte of California. The clerk shall report Obernolte Amendment O f O seven four.

Amendment to the amendment in the nature of a substitute to H R ninety three.

Without objection, the amendment is considered as read. I recognize Representative O.

Uh, thank you, Mister Chairman, uh, the ANS authorizes Casey at a funding level of twenty million dollars a year. Uh, unfortunately, this funding level is insufficient to, uh, provide Casey the resources that are needed to fulfill the obligations that we are giving them, uh, under this bill, including some of the amendments we just took. Uh, also, uh, bills in the jurisdiction of other committees that are of the great american ai act will put even more uh requirements on casey and that will require a substantial increase in the amount of resource that we give them this amendment would increase the authorization for the center to a hundred million dollars a year which i believe is more commensurate with the responsibilities that we are giving casey uh unfortunately uh i acknowledge that it is not possible right now to know exactly how much money is required to uh allow casey to do its

The with the uh amendment is withdrawn and you have um a my word that we will work together.

Mm-hmm.

Amendment withdrawn. The next amendment on the roster is the amendment offered by our representative of white sides of California. Clerk shall report White Sides Amendment O eight two.

Amendment to the amendment in the nature of a substitute to H R ninety-three sixty-three.

Without objection, the amendment is considered as read, and I recognize Representative White Sides of California for five minutes to explain the amendment.

Thank you, Mister Chairman and Ranking Member Lofgren and the Bill's sponsors, Representatives Obernolte and Fusche. I rise to offer an amendment that expands KC's na- national security evaluations, to also look at the capabilities of AI systems conducting autonomous research and development. Right now, humans are driving AI development. We write the code, we run the tests, and we set the pace of progress. But we're rapidly approaching the horizon of recursive self-improvement, the moment where an AI system becomes capable of conducting its own research and development, writing its own code, and training the next generation of AI entirely on its own. The, this underlying bill for Casey allows the voluntary, uh, evaluations of AI systems regarding national security risks, like cyber and CBRN threats. That is extremely important. My amendment ensures we aren't blindsided by a sudden leap in autonomy by expanding our security framework framework to include autonomous AI research and development. When AI begins automating its own R and D, the pace of technological progress will no longer be linear, it will be exponential. The dual stakes, uh, the dual use stakes here are staggering. Autonomous AI R and D could discover lifesaving drugs or unlock entirely new clean energy breakthroughs. However, that same autonomous capability can lead to the AI creating self-replicating cyber-exploits whether on purpose or by accident, that cripple our electricity grid and water systems. And this is in some distant science fiction, the leading frontier labs, OpenAI, Google, Anthropic, have all made it clear that building this exact capability is uh one of their primary goals. If the private sector is openly raced racing to unleash this kind of power, the government should be partnering with those companies to have visibility into their progress. That's exactly why this amendment is so critical. It's forward-looking, common sense measure, built to ensure that we have the visibility required to protect our nation, and keep our family safe while still promoting innovation. I urge my colleagues to support it, and I yield back.

I'd like to recognize a gentleman from California.

Well, thank you, Mister Chairman, uh, I'd like to thank Representative Whitesides for his thoughtful amendment. Uh, I completely agree that recursive self-improvement is something that Casey should study more closely. Uh, when we look at the catastrophe,

Yeah, when yield's back, I'd like to recognize ranking member from California.

Yeah, just quickly, uh, thank you, Mister Whitesides.

Mm-hmm.

We have, as we've, um, learned the, uh, dramatic increase in in capability is really shocking. And uh the assessment that's required here uh is an important one, and I thank him for the amendment and yield back.

I'd also like to thank the gentlemen of California for offering this amendment. It expands the underlying bill by recognizing the role of autonomous artificial intelligence in research and development and by assessing AI's potential to drive future scientific breakthroughs. This amendment complements the bill's efforts to strengthen AI's security supporting uh continued innovation. I urge my colleagues to support this amendment. Any further discussion? Hearing none, the question is on adoption of Whiteside's Amendment O eight two. All those in favor say aye.

Aye. Aye.

Um those opposed say no, and in the opinion of the chair, the ayes have it, the amendment is adopted. The next amendment on the roster is the amendment offered by Representative Lofgren of California. The clerk shall report Lofgren amendment double O one.

Amendment to the amendment in the nature of a substitute to H R ninety three sixty three.

Without objection, the amendment is considered as read. I recognize Representative Lofgren of California for five minutes to explain the amendment.

Thank you, Mister Chairman, the amendment is simple, uh, as you have acknowledged in your opening statement and Mister Obonaty with his withdrawn amendment um Casey is a vital part of the nation's AI efforts and needs to be adequately funded the president uh proposed a, has a budget request of twenty eight million for fiscal year twenty twenty seven. Um, but um, I think that's in, I would have supported uh, Mister Obelnoty's amendment. This was intended as a back-up amendment. Uh, it would uh, uh increase the authorization to the President's budget request. Plus, apparently they had to borrow ten million dollars uh, from GSA to f-