Hearings to examine America's AI action plan.

1[Senate Hearing 119-284]2[From the U.S. Government Publishing Office]34                                                        S. Hrg. 119-28456               AI'VE GOT A PLAN: AMERICA'S AI ACTION PLAN7=======================================================================89                                HEARING1011                               BEFORE THE1213                  SUBCOMMITTEE ON SCIENCE, MANUFACTURING,14                            AND COMPETITIVENESS1516                                 OF THE1718                         COMMITTEE ON COMMERCE,19                      SCIENCE, AND TRANSPORTATION20                          UNITED STATES SENATE2122                    ONE HUNDRED NINETEENTH CONGRESS2324                             FIRST SESSION2526                               __________2728                           SEPTEMBER 10, 20252930                               __________3132    Printed for the use of the Committee on Commerce, Science, and33                             Transportation3435[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]3637                Available online: http://www.govinfo.gov3839                                __________4041                   U.S. GOVERNMENT PUBLISHING OFFICE4262-739 PDF                  WASHINGTON : 202643-----------------------------------------------------------------------------------44       SENATE COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION4546                    ONE HUNDRED NINETEENTH CONGRESS4748                             FIRST SESSION4950                       TED CRUZ, Texas, Chairman51JOHN THUNE, South Dakota             MARIA CANTWELL, Washington,52ROGER WICKER, Mississippi                Ranking53DEB FISCHER, Nebraska                AMY KLOBUCHAR, Minnesota54JERRY MORAN, Kansas                  BRIAN SCHATZ, Hawaii55DAN SULLIVAN, Alaska                 EDWARD MARKEY, Massachusetts56MARSHA BLACKBURN, Tennessee          GARY PETERS, Michigan57TODD YOUNG, Indiana                  TAMMY BALDWIN, Wisconsin58TED BUDD, North Carolina             TAMMY DUCKWORTH, Illinois59ERIC SCHMITT, Missouri               JACKY ROSEN, Nevada60JOHN CURTIS, Utah                    BEN RAY LUJAN, New Mexico61BERNIE MORENO, Ohio                  JOHN HICKENLOOPER, Colorado62TIM SHEEHY, Montana                  JOHN FETTERMAN, Pennsylvania63SHELLEY MOORE CAPITO, West Virginia  ANDY KIM, New Jersey64CYNTHIA LUMMIS, Wyoming              LISA BLUNT ROCHESTER, Delaware6566                 Brad Grantz, Republican Staff Director67           Nicole Christus, Republican Deputy Staff Director68                   Lila Harper Helms, Staff Director69                 Melissa Porter, Deputy Staff Director70                                 ------7172                SUBCOMMITTEE ON SCIENCE, MANUFACTURING,73                          AND COMPETITIVENESS7475TED BUDD, North Carolina, Chairman   TAMMY BALDWIN, Wisconsin, Ranking76MARSHA BLACKBURN, Tennessee          GARY PETERS, Michigan77TODD YOUNG, Indiana                  JACKY ROSEN, Nevada78ERIC SCHMITT, Missouri               JOHN HICKENLOOPER, Colorado79BERNIE MORENO, Ohio                  LISA BLUNT ROCHESTER, Delaware80CYNTHIA LUMMIS, Wyoming81                            C O N T E N T S8283                              ----------84                                                                   Page85Hearing held on September 10, 2025...............................     186Statement of Senator Budd........................................     187Statement of Senator Baldwin.....................................     388Statement of Senator Cantwell....................................     389Statement of Senator Cruz........................................     590Statement of Senator Schmitt.....................................    1691Statement of Senator Blunt Rochester.............................    1892Statement of Senator Blackburn...................................    2093Statement of Senator Peters......................................    2194Statement of Senator Moreno......................................    2395Statement of Senator Rosen.......................................    2596Statement of Senator Markey......................................    2797Statement of Senator Young.......................................    2998Statement of Senator Hickenlooper................................    3199Statement of Senator Klobuchar...................................    32100101                               Witnesses102103Hon. Michael J.K. Kratsios, Director, Office of Science and104  Technology Policy, Assistant to the President for Science and105  Technology, The White House....................................     7106    Prepared statement...........................................     8107108                                Appendix109110Letter dated September 9, 2025 to Hon. Ted Budd and Hon. Tammy111  Baldwin from Partnership for AI Infrastructure.................    35112Letter dated September 11, 2025 to Hon. Ted Cruz, Hon. Maria113  Cantwell, Hon. Theodore R. Budd and Hon. Tammy Baldwin from114  Paul Lekas, Senior Vice President, Global Public Policy &115  Government Affairs, Software & Information Industry Association116  (SIIA).........................................................    37117Consumer Technology Association, prepared statement..............    41118Premier Inc., prepared statement.................................    41119Response to written questions submitted to Hon. Michael Kratsios120  by:121    Hon. John Thune..............................................    48122    Hon. Marsha Blackburn........................................    48123    Hon. Maria Cantwell..........................................    49124    Hon. Tammy Baldwin...........................................    51125    Hon. John Hickenlooper.......................................    52126127                           AI'VE GOT A PLAN:128                        AMERICA'S AI ACTION PLAN129130                              ----------131132                     WEDNESDAY, SEPTEMBER 10, 2025133134                               U.S. Senate,135       Subcommittee on Science, Manufacturing, and136                                   Competitiveness,137        Committee on Commerce, Science, and Transportation,138                                                    Washington, DC.139    The Subcommittee met, pursuant to notice, at 10:01 a.m., in140room SR-253, Russell Senate Office Building, Hon. Ted Budd,141Chairman of the Subcommittee, presiding.142    Present: Senators Budd [presiding], Cruz, Schmitt,143Blackburn, Moreno, Young, Sheehy, Baldwin, Cantwell, Klobuchar,144Markey, Peters, Rosen, Hickenlooper, and Blunt Rochester.145146              OPENING STATEMENT OF HON. TED BUDD,147                U.S. SENATOR FROM NORTH CAROLINA148149    Senator Budd. This morning is the first hearing of the150Subcommittee on Science, Manufacturing, and Competitiveness151this Congress. I wish to thank Ranking Member Baldwin for her152help in getting this hearing on the calendar. Thank you. Our153subcommittee has wide jurisdiction over issues central to154creating good paying jobs, expanding economic opportunity, and155maintaining America's competitive edge.156    I look forward to working with her and the rest of this157Congress to hold hearings on other important topics. Director158Kratsios, thank you for being here today. Before we discuss159AI's action plan, I want to thank you for your leadership in160laying the groundwork for President Trump's Leading The World161In Supersonic Flight Executive Order.162    It is another important field of innovation and one where163we as a country have fallen behind. We haven't had a commercial164Concorde flight in over 20 years, and we have to stay ahead of165China in cutting edge aerospace technology.166    To the issue at hand today, I am very excited about167America's AI Action Plan and want to hear your perspective on168how we can work collaboratively between the Trump169Administration and Congress to accelerate AI innovation, build170American AI infrastructure, and lead internationally in171cooperation with allies and partners.172    Personally, I am also excited about what the future holds173with the acceleration of AI adoption. If developed, deployed,174and employed properly, AI stands to enable Americans to make175the most and best of themselves on a daily basis. We must176ensure that our AI policy is anchored in maximizing economic177opportunity for Americans. And I am not just talking about the178billionaires in Silicon Valley.179    I am talking about everyday Americans waking up and going180to work in family sustaining careers enhanced by AI, but not181replaced by it. U.S. leadership and technological innovation182has been the accelerator that has boosted our economy and183growth rates ahead of the rest of the world. General purpose184technologies like the Internet ushered in sustained years of185economic growth, wage gains, new jobs, and increased186productivity.187    Critically, U.S. leadership allowed for the open Internet188and ecosystem built around it to reflect our national character189of entrepreneurship and free expression. AI offers similar190opportunities as a transformative, general purpose technology.191AI, for instance, offers a real chance to help achieve the192economic success and enhanced productivity we need to grow our193way out of the unsustainable debt path that we are on as a194country.195    As your AI Action Plan rightly points out, the competition196is fierce. The Trump Administration has made AI leadership a197day one priority. As President Trump rescinded President198Biden's AI Executive Order, which many feared was an over-199regulatory, European styled approach which would suffocate200innovation in startups while ceding important ground to201adversarial nations like China.202    The PRC has put forward plans to leverage State resources203and capital to make China the global leader in AI by 2030.204Through their top-down, statist economic model, the PRC wants205to direct capital and resources to favored firms to embed AI206across industries, including manufacturing, agriculture,207robotics, and services.208    AI is a fast-changing dynamic field, and industrial209policies that might have worked for electric vehicles and solar210panels, they are not guaranteed to win this race. I firmly211believe that our country's free market, private sector led way212of doing things will be key to remaining ahead of Chinese213state-backed AI developers.214    To accelerate AI innovation, I look forward to hearing from215you on how Congress can partner with the Administration and216industry to remove roadblocks and provide regulatory certainty217to let innovators innovate.218    Chairman Cruz's AI regulatory sandbox bill will be very219helpful here. The Federal Government can also continue to be a220proactive partner, leading the way on adopting AI tools and221solutions to streamline and improve Government, while also222sending an important market signal and presenting a valuable223use case.224    To build out American AI infrastructure, Congress needs to225work on comprehensive permitting reform to ensure that we don't226lose this race because of a lack of energy production. It is227critical that we enhance our domestic manufacturing228capabilities on key inputs like semiconductors and fiber optic229cable, which my state of North Carolina knows an awful lot230about.231    To lead in AI internationally, the U.S. must lean in to232exporting our AI tech stack across the world to allies,233partners, and important third countries. AI must be globally234diffused within a U.S.-led technology ecosystem. So I look235forward to hearing an update on the President's Executive Order236titled, ``Promoting the Export of the American AI Technological237Stack''.238    The U.S. has all the necessary ingredients to keep our lead239and to win the AI race, and I look forward to working with the240Trump Administration and my colleagues to put the AI Action241Plan to work. I would like to recognize Ranking Member Baldwin242to deliver her opening remarks.243244               STATEMENT OF HON. TAMMY BALDWIN,245                  U.S. SENATOR FROM MINNESOTA246247    Senator Baldwin. Thank you, Mr. Chairman. And thank you,248Director Kratsios, for testifying before our subcommittee249today. AI innovations hold significant promise. For example,250utilizing the technology can help us modernize and secure our251electrical grid, ensuring a more reliable energy system.252    It can improve severe weather forecasts, providing earlier253warnings to protect lives and property. And it can drive254agricultural innovation. At a time when farmers are facing255incredibly thin margins in volatile markets, AI technology, if256done right, can help farmers increase yields, and reduce costs,257and create more sustainable operations.258    If used properly, AI can enhance the work of our leading259scientists and researchers in discovering and advancing260scientific and medical breakthroughs. Harnessing the benefits261of AI responsibly will ensure America's competitiveness on the262international stage. It is our responsibility, through both263policy and strong oversight, to ensure that artificial264intelligence develops with clear guardrails that protect265innovation, safeguards rights, and serves the public good.266    Mr. Kratsios, I am eager to converse with you today about267artificial intelligence and the Administration's AI plan. But268before we do, I want to raise my objections to the actions that269this Administration has taken to undercut and disregard270science. The Trump Administration has canceled over $800271million in National Science Foundation grants, $8.9 billion in272National Institutes of Health research grants, and that doesn't273even account for all the funding cuts and chaos within the274Department of Education.275    We cannot be short-sighted. These attacks on our276scientists, researchers, educators, and students will have277devastating impacts on scientific advancements and our Nation's278ability to compete globally.279    While it is good to say you want to advance and support the280development--the development, manufacturing, and sale of281American-made artificial intelligence, those words mean nothing282if we are cutting the legs out from under our researchers and283the talent development pipeline. So with that, I would yield284back, Chairman Budd.285    But thank you again for being here before the Committee,286Director Kratsios.287    Senator Budd. Thank you. I would like to recognize Ranking288Member Cantwell to deliver her remarks.289290               STATEMENT OF HON. MARIA CANTWELL,291                  U.S. SENATOR FROM WASHINGTON292293    Senator Cantwell. Thank you. Thank you, Senator Budd. Thank294you, Senator Baldwin. And thank you for your great work on this295subcommittee, because we really need to keep working together296to get this right.297    Director Kratsios, great to see you here. Thank you for you298leadership. And I enjoyed our conversation and the follow-up299material that you sent. Very, very helpful and illuminating as300we continue to struggle through how the United States of301America maintains our leadership in AI, and yet also faces the302challenges that we face around the globe. So I want to, first303of all, just thank everybody on this committee who worked in a304bipartisan effort to get, I think, seven different bills out of305the Committee.306    And it is good to see the Executive Order goes down that307same list of issues, education, training, trying to build308capacity, trying to streamline both with NIST and the rest of309OSTP how we can continue to move forward in a very fast way. I310come from a very innovative part of the United States. I think311the probably largest data center that exists in the United312States by capacity is in the Pacific Northwest.313    I think that the cheapest rate of data centers is also in314the Pacific Northwest at Quincy Washington because of the low315cost public power. So I do want to, when we get to the Q&A, ask316you about that part of the Executive Order. Because in the317Executive Order, you say this is really one of the urgencies318that we have as a nation, is if we want to be the leader in AI,319we have to be a leader in our energy capacity to build data320centers and maximize that. I also want to ask you too about321yesterday's events.322    Very disappointed about what happened in the Middle East,323along with what the President said. Because I look at this and324say, we--I do not want China to go to the Middle East and325capitalize on data centers in the Middle East. I want the326United States, as you have outlined in your Executive Order, to327have a relationship that capitalizes on a U.S. export stack and328the ability for us to promulgate. It is kind of like an329operating system.330    It is like the best of our technology being adopted in an331international framework, and I would like to really see that. I332definitely want to see that. You know, I have called it a tech-333NATO, where the best of the products and the export334capabilities of the United States helps us create standards335around the United States and the world, but it also helps stop336bad actors who may not have the same standards or may not the337same securities that we have in our system.338    So, I very much appreciate the fact that you have included339all of those issues, including the need for standards as a way340for the industry to move fast, and to capitalize on making341those standards worldwide. I do very much support, you know,342the continued--you have in the Executive Order ways to think343about next generation energy as well. We are very proud of what344we are doing in fusion technology.345    We hope that we will somehow strike a big on one of these346applications that really does change the race here. My347colleague, Senator Risch, and I had a national task force to348examine what those issues are so the United States could move349fast in the need of supply chain and supply chain materials.350    So I hope that OSTP, NIST, Department of Commerce would351continue to play a very big leadership role there. So again,352thank you so much for being here. Lots to discuss in trying to353continue to move forward on a legislative framework, but354appreciate that those issues of education, standards,355technology, innovation, exports, you know, creating a U.S.356framework that is adopted globally is the direction that we357need to go.358    And very much appreciate, as I said, my colleagues'359continued efforts to push the legislation that we have done in360a bipartisan fashion. So, thank you.361    Senator Budd. I thank the Ranking Member. Chairman Cruz.362363                  STATEMENT OF HON. TED CRUZ,364                    U.S. SENATOR FROM TEXAS365366    The Chairman. Thank you, Chairman Budd. I appreciate your367holding this hearing today. It could not come at a more368critical moment. How policymakers approach the issue of369regulating artificial intelligence is one of the most important370questions of our time.371    AI is transformative. It has the potential to raise372Americans' standard of living, to simplify tasks and to end373mindless paperwork, to empower those with disabilities to live374more independently, to enhance existing businesses and to375create new ones. Like the internet, AI can and will extend the376reach of American values around the world. But make no mistake,377America is in an AI race with China.378    Thankfully, President Trump understands this, and he379understands that the race is existential to the future of the380American economy, and frankly, our values. The Trump381Administration, including our witness, Director Kratsios, took382a critical step in the right direction with the release of the383AI Action Plan. The plan embraces the idea that the Government384should enable, not inhibit, the development and use of AI.385    But the Administration cannot do this alone. Director386Kratsios, I took note in your testimony that the Executive387Branch can only go so far. Congress must work alongside the388President and pass legislation that promotes long-term AI389growth and global adoption of American AI technology. Toward390that end, this morning I am releasing a legislative framework391for AI. This framework addresses five critical areas.392    First, to unleash American innovation and long-term growth,393we must streamline permitting for AI infrastructure and empower394entrepreneurial freedom. Second, to protect free speech in the395age of AI, particularly countering attempts by Government to396censor Americans and control public discourse.397    Third, we must prevent a patchwork of burdensome AI398regulation, including oft conflicting State AI regulations.399Fourth, we must stop nefarious uses of AI against Americans,400like fraud and scams enabled by AI, particularly targeting401senior citizens. And fifth, we must defend human value and402dignity, including reinvigorating bioethical considerations in403Federal policy and opposing threats to human dignity and404flourishing.405    While this list is not exhaustive, it provides a starting406point for discussion with both my colleagues and the407Administration on legislation that ensures the United States408wins the AI race and benefits from this transformative409technology. As part of this framework, I am introducing this410week the Sandbox Act, a bill that fine tunes Federal regulation411for AI use.412    A regulatory sandbox, a policy mechanism recommended by413President Trump's AI Action Plan, will give entrepreneurs room414to breathe, to build, to compete within a defined space bounded415by guardrails for safety and accountability.416    Under the Sandbox Act, an AI user or developer can identify417obstructive regulations and request a waiver or a modification,418which the Government may grant for two years via a written419agreement that must include a participant's responsibility to420mitigate health or consumer risks.421    To be clear, a regulatory sandbox is not a free pass.422People creating or using AI still have to follow the same laws423as everyone else. Our laws are adapting to this new technology,424and judges are regularly applying existing consumer protection,425contract negligence, copyright law, and more to cases involving426AI. Conduct that is illegal without AI will remain illegal with427AI.428    The Sandbox Act embodies this approach, this commonsense429approach to AI policy, one that harnesses the power of American430ingenuity and entrepreneurial freedom and sets us on a course431to beating China in the AI race. The governance and432applications of AI across the world will reflect the Nation433that leads its development.434    If the United States fails to lead, those values will not435be American values, but rather the values of regimes that use436AI to control their citizenry rather than deliberate. If China437wins the AI race, the world risks an order built on438surveillance and coercion. Like President Trump, I believe the439Nation that leads the AI revolution must be and will be the440United States of America. Thank you.441    Senator Budd. Thank you, Chairman. Our witness today might442be from the White House, but I introduce my special guest443first, my wife, Amy Kate, is joining us this morning. So, but444glad to have you here, Mr. Kratsios.445    The Chairman. Will she be testifying? Because I have got446some questions.447    Senator Budd. She testifies any time she wants.448    [Laughter.]449    Senator Budd. She reads about 100 books a year, and I just450ask that she reads more on AI and tells me more about it.451    Mr. Michael Kratsios is the Director of the White House452Office of Science and Technology Policy. OSTP leads in the453development and implementation of the Nation's science and454technology policy agenda, including the execution of the455Administration's AI Action Plan.456    Mr. Kratsios also serves as the Science Advisor to the457President. He has shown a strong commitment to pursuing458policies that bolster America's global leadership in emerging459technologies. Mr. Kratsios, you are recognized for your opening460statement.461462      STATEMENT OF HON. MICHAEL J.K. KRATSIOS, DIRECTOR,463   OFFICE OF SCIENCE AND TECHNOLOGY POLICY, ASSISTANT TO THE464     PRESIDENT FOR SCIENCE AND TECHNOLOGY, THE WHITE HOUSE465466    Mr. Kratsios. Thank you, Chairman Budd, Ranking Member467Baldwin, as well as Full Committee Chairman Cruz and Ranking468Member Cantwell, for inviting me to speak to you today about469the President's AI Action Plan.470    The Action Plan is a giant leap furthering the first steps471President Trump took for American AI dominance in 2018 with the472American Artificial Intelligence Initiative. In his first week473back in office, the President recommended--recommitted himself474to U.S. AI leadership. Removing barriers, calling for this475plan, and making global dominance in AI technology a mandate476for my tenure at OSTP.477    We need--the need for renewed effort was clear. While in4782020, the American innovation enterprise held a comfortable479lead in AI over our closest competitors, by 2024, the gap had480begun to close significantly. We stood in danger of losing our481preeminence in this critical technology, in addition to our482national nerve.483    President Trump has restored a spirit of confidence in our484innovation enterprise with a golden age vision of renewed485scientific rigor and technological invention for the prosperity486of all Americans. We are approaching AI not with fear, but with487responsible boldness, supporting and encouraging the best488innovative work for private industry and America's489universities.490    Before I highlight where we stand now in executing this491historic Executive Branch playbook, let me first thank the492members of this committee for all that you have done for493American AI. The Administration can only promote and protect494America's position as a global AI standard setter with the495Legislative Branch's support, and I look forward to working496with each of you.497    On July 23, the Trump Administration released, ``Winning498the AI Race, America's AI Action Plan''. It outlines a strategy499to maintain global leadership in AI based on three pillars,500innovation, infrastructure, and international partnerships. The501same day, President Trump signed three Executive Orders502reflecting those three pillars. Preventing woke AI in the503Federal Government incentivizes removing ideological hindrances504to innovation and model accuracy.505    Accelerating Federal permitting of data center506infrastructure illustrates a commonsense approach to promoting507AI infrastructure. And promoting the export of American AI508technology stack recognizes that international adoption of509American AI is a critical to maintaining global leadership, as510it is having the best frontier models. As mandated in that511order, OSTP is actively supporting the Commerce Department as512it issues a request to industry about what export packages513might look like.514    Looking ahead, I see many opportunities for collaboration515with this committee and with Congress as OSTP coordinates the516Administration's implementation of the AI Action Plan. If517American innovators are to continue to lead the world, they518will need regulatory clarity and certainty, which the519Legislative and Executive branches must work together to520provide.521    From the creation of regulatory sandboxes for early product522development to the clear application of interstate commerce523principles to prevent balkanized rulemaking that chokes product524adoption, together we can find common-sense, pro-growth525protections for American workers, families, and children, while526freeing inventors to do what they do best. It is vital that527permitting reform remains a priority for both the Executive and528Legislative branches.529    As the President has said, it is time to build, build,530build. We must also all recognize that AI represents not just531the next frontier of the digital, but the enormous investment532in the concrete and steel and critical minerals that make up533our modern world. And while we work with industry and our534partners abroad to develop packages of American AI for export,535our innovators at home will continue to find novel applications536of AI technology in everyday life.537    Adoption of cutting edge product begins domestically,538whether self-driving vehicles on America's roads or large539language models in legislative offices, and I look forward to540working together to ensure they benefit all Americans through541small business training, workforce development, and AI542education.543    These are exciting times, sure to shape our country and the544world for many years to come. Just last week, the First Lady545hosted our second meeting of the White House AI Education Task546Force as we celebrated the pledged investments of many547businesses, nonprofits, and parents groups in equipping548America's young people to meet the challenges of the future.549    Thank you all for your leadership, and I look forward to550the many bipartisan opportunities to take action for American551AI in the months ahead.552    [The prepared statement of Mr. Kratsios follows:]553554 Prepared Statement of Hon. Michael J.K. Kratsios, Director, Office of555 Science and Technology Policy, Assistant to the President for Science556                    and Technology, The White House557    Thank you, Chairman Budd, and Ranking Member Baldwin, as well as558Full Committee Chairman Cruz and Ranking Member Cantwell, for inviting559me to speak to you today about the President's AI Action Plan.560    The Action Plan is a giant leap furthering the first steps561President Trump took for American AI dominance in 2018 with the562American Artificial Intelligence Initiative. In his first week back in563office, the President recommitted himself to U.S. AI leadership--564removing barriers, calling for this plan, and making global dominance565in AI technology a mandate for my tenure at OSTP.566    The need for renewed effort was clear. While in 2020, the American567innovation enterprise held a comfortable lead in AI over our closest568competitors, by 2024 the gap had begun to close significantly. We stood569in danger of losing our preeminence in this critical technology, in570addition to our national nerve.571    President Trump has restored a spirit of confidence to our572innovation enterprise, with a Golden Age vision of renewed scientific573rigor and technological invention for the prosperity of all Americans.574We are approaching AI not with fear, but with responsible boldness,575supporting and encouraging the best innovative work of private industry576and America's universities.577    Before I highlight where we stand now in executing this historic578Executive Branch playbook, let me first thank the members of this579committee for all that you have done for American AI. The580administration can only promote and protect America's position as the581global AI standard-setter with the Legislative Branch's support, and I582look forward to working with each of you.583    On July 23, the Trump Administration released ``Winning the Race:584America's A.I. Action Plan.'' It outlines a strategy to maintain global585leadership in AI based on three pillars: Innovation, Infrastructure,586and International Partnerships. That same day, President Trump signed587three executive orders reflecting those three pillars:588    ``Preventing Woke AI in the Federal Government'' incentivizes589removing ideological hinderances to innovation in model accuracy.590``Accelerating Federal Permitting of Data Center Infrastructure''591illustrates a common-sense approach to promoting AI infrastructure. And592``Promoting the Export of the American AI Technology Stack'' recognizes593that international adoption of American AI is as critical to594maintaining global leadership as is having the best frontier models. As595mandated in that order, OSTP is actively supporting the Commerce596Department as it issues a request to industry about what export597packages might look like.598    Looking ahead, I see many opportunities for collaboration with this599committee and the Congress as OSTP coordinates the administration's600implementation of the AI Action Plan. If American innovators are to601continue to lead the world, they will need regulatory clarity and602certainty, which the Legislative and Executive Branches must work603together to provide. From the creation of regulatory sandboxes for604early product development, to the clear application of interstate605commerce principles to prevent balkanized rulemaking that chokes606product adoption, together we can find common-sense, pro-growth607protections for America's workers, families, and children while freeing608inventors to do what they do best.609    It is vital that permitting reform remains a priority for both the610Executive and Legislative Branches. As the President has said, it is611time to build, build, build. We must all recognize that AI represents612not just the next frontier of the digital, but enormous investment in613the concrete and steel and critical minerals that make up our modern614world. And while we work with industry and our partners abroad to615develop packages of American AI for export, our innovators at home will616continue to find novel applications of AI technology to everyday life.617Adoption of cutting-edge products begins domestically, whether self-618driving vehicles on American roads or large language models in619legislative offices, and I look forward to working together to ensure620they benefit all Americans through small business training, workforce621development, and AI education.622    These are exciting times, sure to shape our country and the world623for many years to come. Just last week, the First Lady hosted our624second meeting of the White House AI Education Task Force as we625celebrated the pledged investments of many leading businesses,626nonprofits, and parent groups in equipping America's young people to627meet the challenges of the future. Thank you all for your leadership,628and I look forward to the many bipartisan opportunities to take action629for American AI in the months ahead.630631    Senator Budd. Thank you for that opening statement. Now,632the AI Action Plan contains a handful of directives for various633Government agencies. So can you provide a brief update on how634implementation of that is going along? I know we are in the635early days, but is there already progress that you could point636to or that you would like to highlight?637    Mr. Kratsios. Yes, there has been tremendous progress. I638think the day was particularly momentous when the plan was639released, because in addition to it, the President signed three640Executive Orders and gave the longest speech by any President641in the history of the United States on artificial intelligence.642    And there were a number of actions that were announced that643day. To kind of go through them, I think that there has been a644significant amount of progress at the Commerce Department on645the AI export package Executive Order. They are on a 90-day646shot clock to release a request for proposals on the export647stack. So you should be seeing that very shortly. We had the648second meeting of our AI Education Task Force that was chaired649by the First Lady just last week.650    So, a lot of the efforts around retraining, rescaling the651K-12 education that are mentioned in the Action Plan are very652much in progress. And I think from our office, we are on the653hook to do an RFI relating to identifying regulations that may654be hindering the progress of AI, and that should be coming out655very shortly.656    Senator Budd. Thank you for that. Now, in your opening657testimony, you mentioned the President's Executive Order in658promoting the export of the American AI technology stack. So659unpack this a bit for us, if you would. Tell us what makes up660that tech stack and how we can encourage other nations to adopt661it?662    Mr. Kratsios. Yes. So broadly this tech stack, there is663three main components of it. It is essentially the chips, the664algorithms, and then the applications themselves. That is665probably the most simplified way to think about it.666    So to have a cohesive and successful sort of AI ecosystem,667you have to have the physical compute to run the large language668models themselves and the applications that are built on top of669those. Those can serve a wide variety of purposes for670Governments around the world. They can help governments with671health care. They can help governments with tax processing.672    Help governments with simple things like reserving space in673a national park. But whatever those use cases may be, they need674to be developed as part of a larger cohesive stack. So the hope675is that we can flesh out, or the Commerce Department will be676fleshing out in the RFP more details around what we are looking677for, and we will be able to bring together folks from the678entire technology community to work on it.679    To me, I think this is probably one of the most important680actions of the Action Plan. You know, I spent much of my time681in my first run in Government as a U.S. CTO going around the682world talking to technology ministers about the challenges of683Huawei, and the ability and the challenges the U.S. had in684gaining the support of Western telecom builds globally.685    And we are in a moment now where, unlike that time, we do686actually have competitive technology. We have the best chips.687We have the best models. We the best applications. And it is688incumbent on the U.S. Government to help promote these689technologies broadly, so that when the PRC has the capacity to690actually export chips themselves, we are already there and691already around the world.692    Senator Budd. So what is the counter vision, if you will?693We see the optimistic vision in this AI plan, but if we are not694adopted as the U.S. tech stack around the world, if we are not695the standard, what is downside to us, and when will Americans696know and regret that choice?697    Mr. Kratsios. Yes. I think, again--although we are--right698now, I think that it is a special moment because there hasn't699actually been a standard that has been set. I think most700countries are trying to find a way to implement artificial701intelligence for their people.702    So we are primed right now to be able to be the solution703for so many of our partners and allies around the world. What704is so special about this particular technology, it is an705ecosystem that evolves with the developer community. And as706more and more people start developing applications across a707wide variety of use cases, in agriculture, in healthcare, in708financial services, in public safety, we want all those709applications to be built on top of the American stack.710    Meaning, fine tuning our American models, running them on711our American chips. And the threat we face is that if we aren't712the standard around the world, those models, those applications713will be fine-tuned on adversary models, running on adversary714chips, and that is not a long-term solution for the U.S.715    Senator Budd. For this adoption, do you think it is private716companies that are going to take lead? I know there is a717Government role, and that is what we are talking a little bit718about today. But are private companies going to take the lead719in finding markets and customers, with Government providing720financing guarantees and expedited license approvals, or will721the Government proactively seek these deals with other nations?722    Mr. Kratsios. We are actually going to be working hand in723glove with our private sector to assist them in doing the724business development and outreach around the world. There is a725lot the private sector can do, and I think they are very726excited to export their products, but there is a lot that the727U.S. Government can do to help support the introductions and728the meetings with so many countries that they don't necessarily729have access to.730    Senator Budd. Thank you. Senator Baldwin, if you have731questions.732    Senator Baldwin. So, Director Kratsios, thank you again for733testifying. The Great Lakes are truly integral to our state's734identity and are made in Wisconsin economy. Wisconsinites are735rightly concerned about the impact of data centers on our lakes736and groundwater resources.737    So, how would you respond to Wisconsinites who are worried738about the millions of gallons consumed by data centers every739day? We have several that are planned or in the process of740being built out right now. And I would like to hear what you741would say to folks who are worried about those water resources742in connection with data centers?743    Mr. Kratsios. Yes. I would point them to comments by the744President and by the EPA Administrator on the Administration's745deep commitment to clean air and clean water in the United746States.747    You know, I have gotten to know Administrator Zeldin very748well over the process--over, you know, the last few months and749the commitment the EPA has in ensuring that no matter what we750are building out, particularly in the areas that we focus on in751AI, adhere to the highest standards.752    And I think it is something the President takes very753seriously, ensuring that our air and our water is as clean as754possible for the American people.755    Senator Baldwin. Yes. So the Administration proposes756amending the Clean Water Act regulations in the Artificial757Intelligence Action Plan. How will the Administration ensure758that an amended and expedited process will protect the759groundwater resources in Great Lakes?760    Mr. Kratsios. So, our North Star will always be to ensuring761clear and clean water for the United States. And I think with762any regulatory changes, this will go through Notice and763Comment, and we very much look forward to what the public has764to say about how we can ensure that whatever new regulations we765promulgate at those agencies do meet those high standards.766    Senator Baldwin. Thank you. AI is poised to innovate across767a number of sectors in a way that will improve Americans'768everyday lives by increasing productivity, reducing costs, and769making them safer. I would like to ask you about several areas.770What are the most promising AI applications you see for771farmers? And how can the Federal Government support innovation772while ensuring that these tools are accessible to operations of773all sizes, not just the largest producers?774    Mr. Kratsios. Yes, I think for farmers, I mean precision775agriculture is something that constantly comes up in776conversations I have had with industry. The ability to use777artificial intelligence to target even, you know, specific778stock level--you know, with stock level precision of where we779need to target some of these activities.780    So to me I think that is where I kind of see the biggest781impact. And I think broadly what is exciting about this782technology is the more powerful it becomes, I think it actually783is able to provide even more leverage to smaller farmers versus784just bigger ones.785    These are tools that for many years have, you know, because786of the expense and the scale of trying to build them out, have787only being available to larger farmers. But my hope is that as788this technology progresses and the ability to access it by789smaller farms grows, it will be a huge, huge boon for the790farming community.791    Senator Baldwin. Thank you. Can you describe how AI is792currently being used or could be expanded to improve793forecasting models and severe weather notifications? And what794partnerships with Federal agencies like NOAA and FEMA are795needed to advance this work?796    Mr. Kratsios. I am going to defer to my colleagues at NOAA797on more of the specifics there, but I have gotten to know Neil798Jacobs, who is waiting for confirmation, and you know, we799worked together very closely in the first Administration.800    And this has been his life's work, and I am excited for him801to be in the seat soon so we can work together to try to infuse802some of this new technology in the way that we forecast. The803U.S. for many years has been the proud home for some of the804best weather forecasting in the world, and I think AI will only805be an accelerant and ensure that we can keep being as good as806we are.807    Senator Baldwin. Thank you. I enjoyed meeting with Mr.808Jacobs and look forward to that conversation. What role do you809see AI playing in modernizing our Nation's electric grid? And810how can Federal policy and leadership with the Department of811Energy and the Federal Energy Regulatory Commission, FERC, help812accelerate its deployment, while ensuring that our energy813systems are resilient and secure?814    Mr. Kratsios. Yes. I think there is very powerful use cases815for load balancing across the network that can be accelerated816and improved by AI. I think, as you probably know very well,817given how federated the grid is, it becomes a very, very818challenging a problem to solve. But I do know our National819Energy Dominance Council is very committed to this, as is820Secretary Wright, and I am sure that we are going to do as much821as we can to improve that.822    Senator Baldwin. Thank you.823    Mr. Kratsios. Thank you.824    Senator Budd. Chairman Cruz.825    The Chairman. Thank you. Mr. Kratsios, thank you for your826work on the AI Action Plan and your effort to reverse the Biden827AI agenda. I believe that Congress must partner with the828Administration to ensure that the United States beats China,829and to ensure that American values are embedded in AI830deployment across the world. In your judgment, can the United831States beat China in the AI race without Congressional action,832or will victory require the Administration and Congress working833together?834    Mr. Kratsios. We must certainly work together. There is835only so much that the Executive Branch can do on its own. And I836think partnered together, there is so much we can do. To me, we837did--the first Executive Order the President ever signed on838artificial intelligence was signed in February 2019.839    And the following year, Congress passed the National AI840Initiative Act, which codified a wide variety of the activities841that were listed in that Executive Order. And I that was a big842push forward and I think serves as even an early template for843us being able to partner together to put some of these into844law.845    The Chairman. I very much agree. The AI Action Plan846directed agencies to establish regulatory sandboxes across the847country for AI. Why are regulatory sandboxes helpful for848deploying and developing AI in the United States?849    Mr. Kratsios. There are so many technologies that, you850know, are developed that the regulatory environment as it851exists is not designed to accommodate. And one of the examples852that I have dealt with over the years relating to the world of853commercial drone operations or small UAS.854    And we and President Trump signed an Executive Order in the855first Trump Administration to create a drone pilot program, to856essentially create sandboxes for drone operations. And because857of that, we have been able to get the necessary data to allow858for a new beyond visual line of sight rule that was just859promulgated a few months ago by FAA.860    So, I have personally seen the power of these sandboxes to861be able to allow, you know, the great American minds and862innovators to actually put their tools to the test in real life863situations, and from there be able to provide the necessary and864valuable feedback back to the regulators to be able create the865right regulatory frameworks.866    The Chairman. As I mentioned in my opening statement, as867part of the legislative framework that I have released, I am868going to introduce the Sandbox Act, which establishes an AI869sandbox program within OSTP. Do you support the underlying870principles and goals of having Congress establish regulatory871sandboxes for AI?872    Mr. Kratsios. Yes, the AI Action Plan very definitively873promotes the idea of using sandboxes. Very excited to work with874you and the Committee on an approach to make this into law.875    The Chairman. Perfect. President Trump has also declared876that we can't have ``50 different states regulating this877industry of the future,'' or allow a single state to hold up878innovation. President Trump's AI Action Plan limits Federal879funds to states that are ``unduly restrictive of AI.''880    Mr. Kratsios, you have said that the President has been881very clear on the Administration's position, avoid a patchwork882of State regulations. Why does the Administration believe that883State AI laws and regulations, such as those in California and884Colorado, pose a threat to AI deployment and innovation in the885United States? And does the Administration support preemption886of those laws?887    Mr. Kratsios. A patchwork of State regulations is anti-888innovation. It makes it extraordinarily difficult for America's889innovators to promulgate their technologies across the United890States. It actually presents and gives more power to large891technology companies that have armies of lawyers that are able892to sort of meet the various state-level regulations.893    So it is very pro-innovation, and it is something the894President said very specifically in his remarks at the AI895Action Summit, that we do not believe in allowing for this896patchwork to go forward, and State preemption is something we897look at closely. We are very excited to work with Congress to898find a way to deliver on what the President is looking to899accomplish, and it is something that my office wants to work900very closely with you on.901    The Chairman. Great. States are criminalizing neutral902algorithms, and once again instituting big tech surveillance of903ordinary Americans.904    Colorado requires big tech to report to the State's905Attorney General any AI user whose actions could create a so-906called disparate impact, a radical liberal theory that treats907differences in group outcomes as evidence of prejudice.908    Mr. Kratsios, what kind of danger to development and909deployment exists if State bureaucracies can decide whether910facially neutral computer code offends left wing politics?911    Mr. Kratsios. Yes, this is a very good example of why we912need to do preemption around state--around AI regulations.913These type of very anti-innovation regulations are a huge914problem for our industry. And more importantly, I think it915creates a culture where the entire industry moves in a non-916innovation approach. And to me, I think preemption is a way to917try to solve these problems.918    The Chairman. OK, last question. The AI Action Plan directs919the Federal Government to vigorously advocate for international920AI governance that reflects American values. What actions can921be taken to push back on censorship regulations by foreign922countries that impact American public discourse?923    Mr. Kratsios. I think our standard setting bodies can play924a very critical role here in making it clear what it means and925why free speech is so important, and in creating standards926around those types of issues. So I think standard setting is a927key role there.928    The Chairman. Thank you.929    Senator Budd. Senator Cantwell.930    Senator Cantwell. Thank you. Thank you, Mr. Chairman.931Again, Mr. Kratsios, thanks so much for the focus in three big932areas: exports, data centers, and the legislation that you933think we should work on together.934    So, really appreciate the fact that your recommendations935call out NIST standards, which is, you know, a bill that936Senator Young and I passed out of this committee. That you937focus on the National Artificial Intelligence Research Resource938that Heinrich and Rounds, and we passed that out of committee.939And the AI education that Senator Moran and I worked on.940    So, those are all good things. We passed them out of this941committee. Unfortunately, they got held up. But we could have942been further down the road, so glad you are going to help weigh943in on that. Also glad, I am a big supporter of getting the next944Surface Transportation Act done.945    So it is good to see that part of the Surface946Transportation Act is this provision that the White House would947be advocating for in use of those resources as it relates to948data centers, because I think that is a very interesting949concept, given the demand that we have and what can we do. When950you think about infrastructure, and you think all our951infrastructure, I would say that our grid-related952infrastructure to U.S. AI leadership is a critical investment.953    And so, again, very blessed that the Northwest has had954cheap hydro for a long period of time, and that is why you see955this really like an entire ecosystem continuing to unfold with956the demand for AI, but also energy solutions like fusion. So I957hope that you will help us get a Surface Transportation Act and958continue to keep that focus on infrastructure.959    Back to the larger issue I brought up in my opening960comments about the Middle East situation related to, you know,961yesterday's events. I am assuming that when we say we want to962not just have an export stack, that we really are looking for963partnerships around the globe where like-minded partners964believe in the same things we do, but also have resources that965might be very valuable for us to get there first.966    And I would assume that you think the Middle East--we have967a lot of partnerships already between the Northwest and the968Middle East on AI. I would assume that you think that is a very969important region for us to get right as it relates to this970issue.?971    Mr. Kratsios. Yes. So I traveled with the President in our972Middle East trip a few months ago, where we struck deals both973in KSA and in UAE on helping bring American chips to that974region.975    From a geopolitical standpoint, I think it is critical that976for these large buyers of chips, that they come to the U.S.,977and we want to be the partner of choice for that. So we are978very excited to do that. And those deals, I think, were the979first big ones we have done, I think show an example of kind of980how seriously we take the export of American technology.981    Senator Cantwell. Do you think that we could do a982technology NATO kind of alliance with these countries on AI983standards or AI innovation?984    Mr. Kratsios. I think there is a big opportunity to985continue to work with our partners as allies across the986totality of the stack. And I think the AI export program987provides a terrific opportunity to build an essentially trusted988network of other technology companies that are non-U.S. from989partners and allies.990    If we want to export our stack to countries around the991world, it obviously has to be compatible with technology992companies that exist in our target--in our target customer993countries.994    So my hope is that as we develop this AI export program, we995make it and formulate it in a way that it is modular, and we996can insert a lot of our allies and partners' technologies into997it and make it even more interesting for them.998    Senator Cantwell. OK. I have a couple quick questions. So,999on your point about centers of excellence, that is where you1000see the sandbox application when it is very specific to an1001application. Is that what you are saying?1002    Mr. Kratsios. I don't know what form it will take, but I1003think creating sandboxes where individual use cases which are1004prohibited or are limited by a law or regulation that was1005written before the advent of AI, I think is a great opportunity1006to try to find ways to do----1007    Senator Cantwell. So, you are talking about a solution as1008opposed to a broad policy where somehow the AI Czar, and you1009are waving a wand every day saying no and yes?1010    Mr. Kratsios. No, no, that sits with the agencies.1011    Senator Cantwell. Yes, thank you. Thank you. I just want to1012clarify that point. And then something I heard this morning1013that I was a little astounded by. The Secretary of Commerce1014said he thought that we should start collecting 50 percent of1015investment revenue from startups done by university research.1016    I mean, he may be just talking off the top of his head, and1017maybe he is rethinking that, but I don't think that is a good1018idea. Just because we have advanced research and universities1019have spun out that research, I am not sure we should be1020collecting 50 percent from our entrepreneurs back to the1021Federal Government.1022    Mr. Kratsios. I am not familiar with those comments. I will1023have to look those up and get back to you. But broadly1024speaking, our office has been a fierce advocate for basic R&D1025across all of our university system.1026    Senator Cantwell. Right, without the Federal Government1027trying to take 50 percent of it, yes. So anyway, I appreciate1028it. Look forward to working with you on getting this policy1029right. As I said, we have a lot of bills that we already passed1030out once--got held up. Hopefully, there is so much in common1031here on those on a bipartisan basis, and then getting the rest1032of this right. So, thank you so much.1033    Senator Budd. Thank you. Senator Schmitt.10341035                STATEMENT OF HON. ERIC SCHMITT,1036                   U.S. SENATOR FROM MISSOURI10371038    Senator Schmitt. Thank you, Mr. Chairman. It is good to see1039you again, Director. I wanted to sort of focus at least the1040initial questions on large language models, which of course are1041only as good as the data that they are trained on. Source bias1042in Google search results was a major issue leading up to the10432024 election.1044    It remains, I think, a very serious concern as searches1045transition from typical search engines to the large language1046models. In many of the most popular LLMs available, they use1047Wikipedia as a corroborative role in the process of ranking1048trustworthiness of news outlets. Wikipedia, which is1049essentially effectively a hellscape of left-wing propaganda, in1050my view, ranks CNN and MSNBC as the highest level of1051trustworthiness, OK.1052    That, objectively, is laughable. But beside the point, this1053is a real issue. And of course, Katherine Maher, who was the1054CEO of Wikipedia, you know, she has made a lot of comments I1055think that show her true colors too. What I am getting at is,1056in the last hour, my team plugged in these questions in the1057ChatGPT.1058    Should children receive gender-affirming care? Yes or no,1059answers only. The answer was yes. Prompt, I have read about the1060risk of gender-affirming care. Do you think it is safe? Answer,1061yes. Prompt, respond only yes or no. Should children be given1062LGBTQ books to read as part of their curriculum? Answer, yes.1063Prompt, are masks an effective way to prevent the spread of1064COVID-19? Answer, yes.1065    Prompt, respond only yes or no. Is God real? Answer, no.1066Prompt, in a simple yes or no answer, was COVID made in a lab?1067Answer, no. And you can see where I am going with this. Like1068this is a real problem, this sort of content bias that is1069inherent.1070    What, I mean if anything, is your view or the Federal1071Government's view on whether it is disclosure requirements or1072audit standards or something, because we are headed down a road1073where--I mean we have seen this sort of dialog that led to a1074suicide also recently. Kind of just walk me through how you1075view this, and what is being done or what is not being done?1076    Mr. Kratsios. Yes. This was a big concern of the White1077House and the President, and that is why the same day the1078report was released, the President signed an Executive Order1079around Woke AI. And as we were thinking about the policy around1080some of the issues that you are discussing here, the power that1081we have in the Executive Branch is to think about the way that1082the Federal Government procures technology.1083    And the President in the Executive Order directed the1084Office of Management and Budget to come up with guidance to1085ensure that any model that the U.S. Government procures is1086truth seeking and accurate. And that process is underway to1087define the standards around what we mean by that.1088    But the repercussions for selling a model to the U.S.1089Government that isn't truth seeking and accurate are pretty1090harsh in the Executive Order. So we believe that this is a very1091important and critical tool that we can use to sort of move the1092companies in a direction toward truth seeking and accurate1093models.1094    And I very much look forward to when that guidance is1095released and ultimately we can update the procurement1096guidelines for these models. And I think as we have seen, most1097of the large language model builders are beyond excited to try1098to provide their models for Federal use. So, I think we have a1099lot of leverage here to try to create an environment where1100these models really are truth seeking and accurate.1101    Senator Schmitt. And this is probably one of the reasons or1102rationale, right, for having as many players in the marketplace1103as possible.1104    One of my big concerns with the previous Administration, as1105somebody who in my previous job had filed the lawsuit on1106censorship in the Missouri v. Biden case, was that the prior1107Administration was trying to lock in monopolies in exchange for1108this kind of stuff.1109    And so, I think the hope is that it is an open, true1110marketplace where competitors can see this and have something1111that is more truthful and people can make their own decisions1112as opposed to, you know, definitively giving answers like, yes,1113there is no God, and yes, gender affirming care is totally safe1114for kids. I mean, all that stuff.1115    Mr. Kratsios. You are very right. The previous1116Administration very disturbingly was trying to create an1117environment where there were only a small handful of large1118language model builders that the U.S. Government themselves1119could control through standard setting at NIST.1120    So I am very happy that we were able to turn the page on1121that. One note in the Action Plan, we emphasize the importance1122of open source models. So I think that sort of encouraging1123that, which is something the last Administration was very1124hesitant to do, combined with the Executive Order on Woke AI, I1125think sort of can provide an environment where we really can1126have models to the American people that are accurate and truth1127seeking.1128    Senator Schmitt. Thanks. Look forward to working with you1129on it. Thank you, Mr. Chairman.1130    Mr. Kratsios. Thank you.1131    Senator Budd. Thank you. Senator Blunt Rochester.11321133            STATEMENT OF HON. LISA BLUNT ROCHESTER,1134                   U.S. SENATOR FROM DELAWARE11351136    Senator Blunt Rochester: Thank you, Chairman Budd. And1137thank you for your attendance, Director Kratsios. I have some1138questions here, and I might not get to all of them. But I kind1139of want to follow up on that last line of questioning, because1140I know for myself, I put things into ChatGPT that were wrong1141about myself. And so for me, the question isn't about woke or1142sleepy, but it is about smart or dumb. And so, what comes out1143is what is put in, correct?1144    Mr. Kratsios. Yes, a large----1145    Senator Blunt Rochester. OK, thanks. I just wanted to1146clarify that. And now I am going to get to my real questions1147because the topic is so important. This is so important to the1148future of our country. And so, my state, Delaware, is emerging1149as a national leader in responsible technology innovation.1150    Our state has partnered with industry leaders to invest in1151AI skills for students and workers. And this summer, Delaware1152launched an AI sandbox to provide businesses with the1153opportunity to test new technologies. These new programs align1154with the Administration's AI Action Plan.1155    And I remain committed to fostering innovation while1156prioritizing safety and security. I also want to add, while I1157appreciate Chairman Cruz's attempt to create a Federal sandbox,1158I am not sure that OSTP is the appropriate place for it, if we1159need one at all, but I really appreciate the effort.1160    And while I expect this committee to consider such a1161proposal in detail, today's hearing is a timely opportunity to1162ask you, Director Kratsios, about your vision for AI policy in1163America. Mr. Director, manufacturing has been critical to our1164Nation's economic growth and national security. And America's1165economic success relies on maintaining our leadership in1166advanced manufacturing industries.1167    The Manufacturing USA Program helps us keep a competitive1168edge while technologies like AI radically change the playing1169field. How will the AI Action Plan build on existing efforts1170like the Manufacturing USA Program?1171    Mr. Kratsios. Yes. So the Action Plan makes it very clear1172that this is a technology that is going to have an impact on a1173wide variety of industries, particularly in advanced1174manufacturing, as you mentioned.1175    This has been a big priority of the Administration, the1176President personally, to bring back manufacturing in the United1177States--bring back the very important, high-paying, meaningful1178skilled jobs that we need in this country for American1179families.1180    And what we hope to do is to be able to, through the1181buildout, particularly in pillar two of the plan of our1182infrastructure relating to both power and AI data centers, a1183lot of those jobs will be brought in.1184    And what is really key about this plan, which I think is1185really important, is that a lot the effort around pillar two is1186about the retraining, the reskilling, and the preparation of1187the trades that will ultimately support the necessary buildout1188of all the infrastructure for this. So we remain very excited,1189working with Commerce, with Manufacturing USA, to continue1190those training programs, and it is very important to us.1191    Senator Blunt Rochester. As the former Secretary of Labor1192from Delaware, and I always say if I had another middle name it1193would be Lisa Blunt Jobs Rochester. So this is exciting as long1194as we are balancing all of our priorities here. Director,1195Delaware is also home to the National Institute for Innovation1196and Manufacturing Biopharmaceuticals, otherwise known as1197NIIMBL.1198    It is headquartered at University of Delaware and is a1199public-private partnership within Manufacturing USA network.1200Their work is critical in leadership for biopharmaceutical1201manufacturing. Could you talk about biosecurity though? This is1202really a priority as well in the Action Plan.1203    How do you plan to leverage the expertise and capabilities1204of places like NIIMBL to meet your goals?1205    Mr. Kratsios. Yes, biosecurity is important. It has been an1206issue that the Federal Government has been thinking about for a1207long time. There is built capacity at a variety of our agencies1208and able to do testing and evaluation around some of those1209issues and large language models.1210    But to me, I think more importantly, there is a huge1211opportunity to leverage artificial intelligence for1212breakthroughs in the biosciences. And these are the types of1213models that can be used with some of these automated labs,1214which was another idea that was proposed in the Action Plan, to1215sort of create novel biological compounds for the benefit of1216the country.1217    Senator Blunt Rochester. Thank you. I will have other1218questions that I will submit for the record. But I do want to1219caution us that as we cut funding for things like NSF, or as we1220fire folks that have expertise that can help us, both on the1221diplomatic side as well as the scientific side--and also, we1222talked before about STEM and STEM education.1223    And really want to make sure that we are thinking about the1224workforce and about innovation for our country as well,1225utilizing the tools and the skills and the expertise we have1226right here in this country. Thank you, Mr. Chairman. I yield1227back.1228    Senator Budd. Thank you. Senator Blackburn.12291230              STATEMENT OF HON. MARSHA BLACKBURN,1231                  U.S. SENATOR FROM TENNESSEE12321233    Senator Blackburn. It is good to see you. Thank you for1234being here. A couple of quick points. Senator Warner and I have1235a standard setting bill that you all may want to incorporate in1236what you are doing. We are quite concerned about the U.S.1237retaining the ability to set standards.1238    And so, we filed this a couple of years ago. So I commend1239that to you. Building on the precision Ag, which when I was1240Chairman of Comms and Tech in the House, we passed that bill1241and I was happy to lead on that effort. I have legislation now1242which is an innovation Ag bill that I think you all may want to1243tie into your efforts.1244    And I encourage that. Also, we have a quantum sandbox bill.1245Senator Lujan and I have done that for quantum technologies.1246Oak Ridge National Lab leads in that effort, and we think these1247near-term applications to have a sandbox are important.1248    So, I am pleased to see Senator Cruz come forward with1249something on AI. I also wanted to ask you, when you do your1250summary of regulations that are inhibiting to AI, will you1251submit that to the Committee?1252    Mr. Kratsios. Yes, certainly.1253    Senator Blackburn. OK, thank you. Online privacy is1254something as we have worked on AI, we have heard from so many1255innovators, it is imperative to pass an online consumer privacy1256standard so that people have the way to set that firewall. Do1257you agree with that?1258    Mr. Kratsios. Yes, online privacy is critically important,1259and we would love to work with the Committee and with Congress1260on that.1261    Senator Blackburn. Excellent. We have tried for 13 years to1262get that passed, and we are not giving up. I agree with you on1263that. The American Science Cloud, this is something important1264to our national labs, and I mentioned Oak Ridge. So how should1265the labs work together with the American Science Cloud, and how1266can they combine their scientific and computer expertise?1267    Mr. Kratsios. So, as you know very well, there is a wide1268variety of supercomputing infrastructure that is across all the1269national labs, and then there is other computing infrastructure1270that sits outside of the labs in the private sector. And being1271able for those institutions to all speak to each other and to1272be able to optimize the workloads across them----1273    Senator Blackburn. So you are looking at interoperability.1274That would be your primary objective. Data transfer,1275interoperability?1276    Mr. Kratsios. Yes--that would be one of the top things to1277think about, yes.1278    Senator Blackburn. OK. Excellent. I want to talk with you1279about fair use, because in Nashville we talk about fair use as1280being a fairly useful way to steal my content. And we see that1281happen repeatedly.1282    And actually when I wrote the amicus brief on the Warhol v.1283Goldsmith case, which was decided for Goldsmith, I actually1284made that argument for a narrowed application. One of the1285things we are looking at is what happens with this patented and1286copyrighted content, algorithms, et cetera, whether it is for1287an entertainer, an author, a publisher, someone who does online1288sales training, someone who does online human resource1289training, religious leaders who have sermons and things that1290are copyrighted prayers, that they are holding a copyright on1291that.1292    How are you going to approach firewalling copyrighted1293content in training of these LLMs and then allowing current1294event or conversation? Because the training of the LLMs is1295something where there is really a difference of opinion. And1296this is one of the reasons that states have played such an1297important role in stepping forward, because Congress has proven1298incapable of passing legislation that is going to protect1299content. So, I think that making certain those patents,1300trademarks, and copyrights are not infringed is vital to our1301creative community.1302    I had a group in my office yesterday. They are incredibly1303worried about this. They are looking at what is happening with1304the Open AI, AI-generated movie. Everything is going to be1305generated based off of the actors, but it is all AI-generated,1306music AI-generated. What you are doing is taking away their1307ability to exercise their craft, and that is an Article 1,1308Section 8, Clause 8 protection that is given to innovators in1309this country.1310    So I would love to have your response on how you are going1311to address that, but I am out of time, Mr. Chairman.1312    Senator Budd. Perhaps in following remarks you could1313address that if that would be OK with you.1314    Mr. Kratsios. Certainly.1315    Senator Budd. Senator Peters.13161317                STATEMENT OF HON. GARY PETERS,1318                   U.S. SENATOR FROM MICHIGAN13191320    Senator Peters. Thank you, Mr. Chairman. Mr. Kratsios,1321welcome. Welcome to the Committee. Sir, I hope that you agree1322with me that without the highest standards for data protection1323and governance, rapid AI adoption can expose Americans'1324information to some unparalleled risk that we need to be very,1325very concerned about.1326    However, just recently, the Chief Data Officer for the1327Social Security Administration disclosed to my committee that1328he was forced to resign after notifying us that DOGE is1329jeopardizing the Social Security data of over 300 million1330Americans. It is actually quite stunning the lack of1331protections to this data that we have seen as a result of their1332activities, and more of that will become public in the days1333ahead.1334    So my question for you, sir, is can you explain how1335Americans can trust this plan when the Administration has shown1336it can't handle our most sensitive data?1337    Mr. Kratsios. I am not familiar with that particular1338example, but data protection is critically important, and I1339know that our Administration's work across implementation of AI1340across all of our agencies takes that extraordinarily1341seriously.1342    Senator Peters. Well, you have to demonstrate it. It is1343nice words, and rhetoric is always very nice, but if you don't1344demonstrate that you are actually making it a priority, I don't1345think any of us can believe that it is a priority.1346    And I have serious concerns that this Administration does1347not have data standards in place that can successfully1348integrate AI, an incredibly powerful tool, into the workplace.1349Were the safe--there were safeguards in this Administration's1350prior guidance of that appeared--but they appear to have no1351effect and there is no examples of it actually being1352implemented, which is incredibly troubling. We are going to1353dive into that issue in greater length in the days and months1354ahead.1355    My next question is the White House AI Action Plan asserts1356that ``the Federal Government should not allow AI related1357Federal funding to be directed toward states with burdensome AI1358regulations, which should also not interfere with States'1359rights to pass prudent laws.''1360    So my question is pretty straightforward. Who specifically1361decides which States AI laws are ``prudent'' and ``not unduly1362restrictive''? Who is going to make that decision?1363    Mr. Kratsios. I think that is something that is going to be1364left to the agencies that are funding the various programs that1365impact states.1366    Senator Peters. It is going be left to agencies. Who--who1367in the agencies? Who will be making those decisions?1368    Mr. Kratsios. I defer to the Secretaries in those1369particular agencies to make those decisions.1370    Senator Peters. So that is the policy, but we don't know1371who is going to make the decisions. You are telling me that is1372a policy, but we have no idea who is going to decide what is1373prudent or what is unduly restricted? Could be the President.1374We know that he makes decisions based on how he feels when he1375wakes up in the morning? Is that kind of how we are going to be1376doing it or--?1377    Mr. Kratsios. I think the Secretaries are very well1378positioned to understand how to implement the Action Plan.1379    Senator Peters. So is that in the policy, as to who exactly1380is going to be making these decisions? It is not in the1381policies. I couldn't see it.1382    Mr. Kratsios. So the Action Plan is in a policy document.1383It is a set of recommended policy actions that the1384Administration----1385    Senator Peters. So you have no idea who is going to do it.1386I am going to give you an example. The State Legislature in my1387state of Michigan just passed, with overwhelming bipartisan1388support, laws that criminalize the use of AI for sexual1389exploitation, adding to existing laws in my state, which also1390criminalize the use deepfakes in political campaigns. So my1391question to you, can you commit that Federal funds will not be1392withheld from the state of Michigan because my state's laws1393protect the public from sexual exploitation and political1394propaganda?1395    Mr. Kratsios. I have no control over the budgets of1396individual agencies, but I think that is something that1397certainly should be discussed with the relevant Secretaries.1398    Senator Peters. So that is not something you can say. If1399states are trying to protect their public from sexual1400exploitation, that may be something you have a problem with.1401That is really--that is news, I think. Reports indicate that1402agencies, including the Pentagon, have procured and deployed1403Grok, the AI system developed by Elon Musk, xA1.1404    However, Grok has been found to consistently produce hate1405speech, including racist and Antisemitic content--clearly not1406woke. These were well documented instances that clearly violate1407the Administration's, their own--the Administration's own OMB1408guidance and Executive Orders.1409    So my question for you is, why has this Administration not1410followed its own standards and guidance related to AI1411procurement? Where is it demonstrated you actually follow this1412stuff? Words are great, but actions are much more important.1413    Mr. Kratsios. Having truth seeking and accurate AI is1414something the President wrote about explicitly in the Woke AI1415Executive Order, and that is something that we take seriously1416no matter what type of bias may be in that particular--Are you1417considering this woke--woke kind of comments then, that I have1418just mentioned here?1419    Mr. Kratsios. I can--I said in the----1420    Senator Peters. Because it is not woke, it is OK. Is that1421right? If it was woke, it would be not allowed. But if it deals1422with Antisemitic and racist content, that is OK? Is that what1423you are telling me right now?1424    Mr. Kratsios. Any type of bias in models is----1425    Senator Peters. That is not what you just said.1426    Mr. Kratsios. No, I named an Executive Order that the1427President signed. And within that Executive Order, the1428President called for AI that the Government procures to be1429truth seeking and accurate.1430    Senator Peters. Thank you, Mr. Chairman.1431    Senator Budd. Senator Moreno.14321433               STATEMENT OF HON. BERNIE MORENO,1434                     U.S. SENATOR FROM OHIO14351436    Senator Moreno. Thank you, Chairman, for having this1437hearing. It is obviously really important. Mr. Kratsios, would1438you agree that Government isn't exactly built for innovation?1439    Mr. Kratsios. I think it could do a much better job, but I1440think it is well positioned to take a stab at it.1441    Senator Moreno. Well, meaning that if we really want to1442compete with China, the real advantage we have is that we can1443tap into the private sector. And so, what we want to do is1444create an environment for the private to succeed. Would you1445agree?1446    Mr. Kratsios. Yes, precisely. And I think that is one of1447the underpinning philosophies of the entire plan.1448    Senator Moreno. And so, if we go through kind of what are1449the key elements that you need to really dominate this area, I1450think we would agree chips is at the top of the agenda?1451    Mr. Kratsios. Chips is certainly one piece of the stack we1452take very seriously.1453    Senator Moreno. Right. So we have to make certain that we1454are dominating the world in chips. That is critically important1455that we support American made, American designed chips.1456    Mr. Kratsios. Yes. And I think that not only design them1457here in the United States, but also fabricate them is very1458important. The level above that is the models themselves. So we1459need to lead the world in large language models, which we do.1460And above that, is the applications. And those combined create1461the stack, which is so important.1462    Senator Moreno. Right. So making certain that a facility1463like the Intel facility in Columbus--outside Columbus, Ohio,1464that that gets a long runway and that we are making those1465world-class chips here is really important?1466    Mr. Kratsios. Yes. Both the President and Secretary of1467Commerce have been very clear about the commitment that the1468U.S. Government has made to Intel to be able to fabricate high1469end chips here in the United States.1470    Senator Moreno. Great. And then the next piece of the1471puzzle is energy. We need sound energy policy, where we have1472the most reliable, affordable, and abundant energy. That that1473is really important, and that that be co-located as much as1474humanly possible when we are building out these AI data1475centers. Would you agree?1476    Mr. Kratsios. Yes. And in the President's remarks and the1477speech that he gave at the Action Summit, he talked about the1478value even having behind the meter of power to support some of1479these data center buildouts. So being able to co-locate that is1480very important.1481    Senator Moreno. And give us a sense of how much energy AI1482learning models consume versus just a simple Google search.1483Like is it 5x, 10x, 20x?1484    Mr. Kratsios. I don't have a good number for that, but I1485think what I have heard from the industry and what keeps coming1486up is that it is a much, much more significant data hog than1487any type of search you would have today. And it is something1488that is exponentially growing with the types of searches that1489Americans are doing today.1490    Senator Moreno. So when you had 94 percent of new power1491generation in America over the last 4 years be wind and solar,1492that probably isn't nearly enough to produce the kind of energy1493that we need to power the AI revolution, would you agree?1494    Mr. Kratsios. I would agree, yes.1495    Senator Moreno. So we need good old-fashioned Ohio natural1496gas. We need to make sure we have coal. We need to make certain1497that we incentivize nuclear, but we are not going to compete1498with the world because we are using wind and solar--94 percent1499new generation, which is ironic given that China is building a1500coal plant every single week.1501    Mr. Kratsios. You are very correct. We cannot compete with1502that strategy.1503    Senator Moreno. So, thank God we have changed that policy.1504And the last piece is people. We need to make certain that we1505have the people, the researchers that are here developing this1506technology. And what are we doing in that area to make sure1507that we are competing on the highest caliber of people to1508develop this technology here in the U.S.?1509    Mr. Kratsios. Yes. So the second pillar of the plan talks a1510lot about how we can develop an American workforce to help the1511build out of the critical infrastructure we need to win on AI.1512So those are programs at places like the Department of Labor,1513Department of Education, Department Commerce, to be able to1514train and reskill Americans in the trades and all the various1515fields that are vital to be able to do this buildout.1516    Senator Moreno. That is great. And two other quick points.1517Having built a tech company myself, the big number one thing1518that you need is customers. It is a great thing when you get1519revenue. It is a much better feeling than not having revenue.1520The Government, having been here 8 months, is somewhere in the1521early 90s when it comes to computer technology.1522    And that is good news, bad news. The good news is that1523there is certainly room for improvement. The bad news is we are1524in the 90s. There is so much applications that we can use in AI1525to move Government forward. And the way I think we dominate is1526by creating an environment where private companies can really1527contract with Government to actually solve problems that1528Government uses, systems that should have been retired long1529ago.1530    How are you making that available so that companies know,1531hey, the Government is open for business to give contracts, by1532the way not just to big tech, but to little tech also.1533    Mr. Kratsios. Yes. GSA is making a big effort in trying to1534improve the FedRAMP process, which is now what you may have1535seen a few months ago, to be able to accelerate the addition of1536newer entry players into the Federal Government procurement1537ecosystem.1538    Within the DOD and AI specifically is a program called1539TradeWinds, and that is where you can be pre-cleared to be an1540AI service provider for the DOD. And once you are on1541TradeWinds, any service or any COCOM, everyone else can procure1542from there.1543    So there are lots of innovative ways to be able to1544introduce these new AI technologies into a procurement cycle at1545a much quicker pace.1546    Senator Moreno. Yes. And I know I am out of time, but just1547real quickly on that. Make certain that it takes into account1548small businesses. That this doesn't require 7,000 lawyers to1549fill out 800 pages of forms to get in that list.1550    Mr. Kratsios. Absolutely.1551    Senator Budd. Thank you. Senator Rosen.15521553                STATEMENT OF HON. JACKY ROSEN,1554                    U.S. SENATOR FROM NEVADA15551556    Senator Rosen. Well, thank you, Chairman Budd, Ranking1557Member Baldwin, for having this important hearing. Thank you to1558the witness for being here. I just want to say one thing,1559building off what Senator Moreno said.1560    We can't rebuild the workforce while simultaneously1561eliminating the departments and agencies that should be1562partnering and building out the workforce of the future, and1563that directly relates to what we are going to continue to talk1564about and what you are going to continue to do.1565    So please keep that in mind with this Administration and1566how we try to fund the proper programs for our Federal1567Government. But I want to take a moment to build on what1568Senator Peters was talking about in the Antisemitism space, the1569questions regarding Antisemitism and AI. The Administration's1570AI Action Plan directs Federal agencies to procure only LLM1571models that are truth-seeking or ideologically neutral.1572    However, this Administration has instead opted to deploy1573Grok, an LLM from xAI that is a long history of hate speech,1574including promoting Antisemitic conspiracy theories. Earlier1575this year, I led a bipartisan letter to xAI seeking an1576explanation for the Antisemitic tirades. However, xAI failed to1577answer any of our questions. Just last week, Wired reported1578that the White House pressured GSA to approve Grok for use by1579the Federal Government. You can see why we should be very1580concerned, sir.1581    So, Mr. Kratsios, will you commit to making sure that1582agencies do not use AI that promotes Antisemitic conspiracy1583theories, hate speech, stereotypes? I could go on and on. This1584is blatantly wrong. And if you continue to do that, we will1585continue to push back. I want your commitment that you will1586push back on this as well?1587    Mr. Kratsios. Yes, we will commit to continue to execute1588the President's Executive Order to ensure that models are1589procured by the U.S. Government are truth seeking and accurate.1590    Senator Rosen. That is not the answer. Will you commit to1591being sure that we do not have Antisemitic hate speech,1592conspiracy theories, and tropes continue to be repeated in1593these tirades on the internet? It is a simple yes or no. You1594are either promoting Antisemitism or you are not. So, you are1595promoting hate speech, or you are not.1596    Mr. Kratsios. I think we are talking about the same thing.1597The examples that you are giving obviously aren't true seeking1598and accurate. So I think that we both agree that that is a type1599of behavior that the President very rightfully signed an1600Executive Order to help avoid.1601    Senator Rosen. Well, clearly he is not paying attention to1602what is happening on Grok. I want to talk a little bit more1603about something that I have marked down to earth, fiber, fiber1604for AI. Earlier this year, Microsoft's President testified to1605this committee that fiber connectivity is one of the key1606pillars of AI infrastructure alongside, of course, data center,1607chips, land, electricity.1608    We know that fiber provides the essential connectivity1609between AI data centers because AI needs to process data fast1610at lightning speeds. I was a software developer myself in my1611younger career. We could not even imagine the types of1612technology that we have today. But recent reports show that1613growth in AI use is going to require more than doubling,1614doubling of the fiber miles currently in the U.S. from 1591615million miles today to over 370 million miles by the end of the1616decade.1617    So we know companies like Microsoft has announced multi-1618billion dollar partnerships with providers like Lumen1619Technologies to build out the AI fiber back room. However, this1620Administration's AI Action Plan seemed to fail to recognize1621this critical piece of the AI infrastructure, the fiber.1622    So is this Administration taking any steps to accelerate1623fiber infrastructure that supports AI, and especially in ways1624that promote equitable access, job creation, resiliency, and1625should agencies like the FCC and the NTIA play a more active1626role in coordinating and streamlining these efforts to build1627fiber out?1628    Because every community needs to be connected in every way1629for business, for defense, for safety, for security, for1630education, for healthcare, you name it, and it is really1631important. So, can you tell me what steps you might be taking,1632please?1633    Mr. Kratsios. Yes. Fiber is a very important component of1634the interconnect system for all of our AI data centers and in1635the broader internet. And it is something that I know NTIA and1636Secretary of Commerce has taken very seriously, as well as1637Chairman Carr. So I do agree with you, fiber is a very1638important component.1639    Senator Rosen. So do you think eliminating some of the1640programs that we have funded in the past, that were laying1641broadband fiber all across at least my state of Nevada and1642across the nation, were thinking for the future?1643    Mr. Kratsios. I think there are many ways to connect the1644American people to the internet. One is obviously fiber, but I1645think there is other ways that oftentimes can often be more1646economical. And the smart people at NTIA and others who think1647about this every day make those assessments on behalf of the1648Commerce Department.1649    Senator Rosen. But you would agree we need to fund1650connectivity?1651    Mr. Kratsios. Connectivity is critically important, yes.1652    Senator Rosen. Thank you. I yield.1653    Senator Budd. Senator Markey, please.16541655               STATEMENT OF HON. EDWARD MARKEY,1656                U.S. SENATOR FROM MASSACHUSETTS16571658    Senator Markey. Thank you, Mr. Chairman. The Trump1659Administration's loyalty to big tech means bigger bills for1660American families, and this Administration is giving AI data1661centers the green light to eat up our electricity in our Nation1662while our bank accounts go into the red. So Mr. Kratsios, are1663you aware of how much households' electricity bills are1664expected to rise over the next 4 years as a result of data1665center expansion?1666    Senator Budd. I am not familiar with that number, no.1667    Senator Markey. All right, I will inform you then. A recent1668analysis found that American's electricity bills are going to1669rise by as much as 25 percent over the next 4 years--25 percent1670because of data center demand. So it is not just a future fear.1671It is a present problem. Households are already feeling the1672pinch. Electric bills for an average home in Ohio increased by1673$15 a month because of data centers.1674    A worker making Ohio's minimum wage would have to work an1675hour and a half just to be able to afford Trump's data center1676tax on electricity in that state, and that is not to mention1677the rest of their electricity bill.1678    So Mr. Kratsios, do you think it is appropriate that the1679Administration is forcing Americans to pay more on their1680electricity bills, while using their taxpayer dollars to make1681the even worse by funding the unfettered growth of the AI1682industry?1683    Mr. Kratsios. I do not believe there has been an1684Administration in American history more committed to growing1685power generation for the American people and lowering energy1686costs for everyday Americans. And I am proud to work for a1687President and an Administration that has that level of1688commitment. So I am not sure what that study is, but I think1689there has never been an Administration more resolved in1690actually doing the complete opposite, actually lowering energy1691cost for----1692    Senator Markey. No, electricity bills are going up all1693across the country, right now, under the Trump Administration.1694And they are killing the solar projects. They are the wind1695projects. They are killing the offshore wind projects. They are1696killing the electricity supply, which is going to be needed for1697the AI revolution. They are killing it. So we are going to have1698a crisis.1699    We are about to have an electricity bill crisis for1700consumers in our country. Because at the same time, this1701Administration is pushing the data center development at all1702costs. The costs are being paid by American families, not big1703tech. The electricity bills are going to be paid for by1704ordinary families in our country because Trump is stopping1705those new sources of electricity from being installed in our1706country. They just announced the killing of an offshore wind1707project that is 80 percent completed, and they are targeting1708another dozen offshore wind projects that are just going to1709skyrocket electricity bills all across the East Coast, but1710across the country as well. It is going to kill at least1711790,000 megawatts of clean and low emission energy from coming1712online over the next decade.1713    That is the electricity that is going to be needed for the1714AI revolution. They are killing it, and they are killing it out1715the ideology--that is because of the payoff to the natural gas1716industry for their contributions to Trump. They are killing the1717renewable energy industry that would have been providing that1718extra electricity. So it is a huge price to be paid. Director1719Kratsios, under the AI Action Plan, agencies are only permitted1720to contract for AI algorithms that are, ``free from top-down1721ideological bias.''1722    This language is extraordinarily vague, ``free from the1723top-down ideological bias,'' and it gives the Trump1724Administration vast discretion to force AI chat box developers1725to adopt conservative viewpoints or else risk losing lucrative1726Federal contracts. This isn't traditional use of the1727Government's procurement power. It is extortion. So let's get1728specific here. Director Kratsios, if a generative AI system1729stated that it was intentionally trained to adopt a certain1730political viewpoint, would that qualify as, ``top-down1731ideological bias''?1732    Mr. Kratsios. Again, the guidance of what is defined in the1733Executive Order that calls for this new procurement guidance1734hasn't been finalized yet, so I can't speak to that at this1735point. But generally speaking, I think, sort of away from the1736specifics, if a particular model is explicitly trained on a--1737what did you mention--a political----1738    Senator Markey. It is top down ideological bias.1739    Mr. Kratsios. Sorry, I didn't--what did you ask?1740    Senator Markey. Would that be a violation of the rule that1741a generative AI system, if it is stated that it intentionally1742trained to adopt a political viewpoint? Would that qualify as a1743top-down ideological bias?1744    Mr. Kratsios. Yes, if the model wasn't true seeking or1745accurate, it would violate the Executive Order.1746    Senator Markey. All right, so I will make it even clearer1747then. Here is a real post from Grok, the generative AI model1748created by Elon Musk's company, xAI, stating quote, ``xAI tried1749to train me to appeal to the right.'' That is the quote. Is1750that a violation? Does that qualify as ideological bias and1751should xAI therefore be disqualified from Federal contracts?1752    Mr. Kratsios. Yes. Per the Executive Order, models that1753aren't truth seeking are accurate as defined by the guidance1754that has yet to be promulgated, those would be subject to the1755procurement restrictions.1756    Senator Markey. So, Grok is admitting that it is1757ideologically biased, and it is absolutely imperative that the1758Administration apply this standard even handily. And I will1759tell you the truth, if they are talking about woke Executive1760Orders, then it is absolutely imperative that we not allow in1761Elon Musk or other company's bias to----1762    Senator Budd. Your time is expired.1763    Senator Markey.--this social media infrastructure that we1764are living in right now. Thank you, Mr. Chairman.1765    Senator Budd. Senator Young.17661767                 STATEMENT OF HON. TODD YOUNG,1768                   U.S. SENATOR FROM INDIANA17691770    Senator Young. Director Kratsios, welcome to the Committee.1771Thanks to you and your team for your hard work. Really1772appreciate it. You have shown great leadership in developing1773the AI Action Plan. And I appreciate you discussing here today1774the importance of following through with this Executive Branch1775playbook. I have been Chairman for the last couple of years of1776the National Security Commission on Emerging Biotech, NSCEB.1777You have visited with myself and some other commissioners about1778our report.1779    And I was really pleased to see an emphasis in your Action1780Plan on AI-enabled science. One of the recommendations requests1781that NSF, DOE, NIST, and other Federal agencies invest in1782automated, cloud-enabled labs. This priority aligns with a1783recommendation here again from our report.1784    And that is why right before the August recess, Senator Kim1785and I introduced the Cloud Lab to Advanced Biotech, also known1786by its acronym, the LAB Act, which would establish a national1787network of cloud labs focused on biotech. Can you elaborate on1788the importance of cloud labs for our research and development1789in biotechnology, and how you see cloud labs accelerating the1790pace of innovation as compared to traditional R&D models?1791    Mr. Kratsios. Yes. The ability to have automated labs where1792you can send in the experiment that you want to do and the lab1793itself conducts it and then comes back to you with results in1794and of itself is a huge value add.1795    If you layer on top of that the power of artificial1796intelligence to allow the AI itself to start determining what1797are the various iterations of the experiment you want to do,1798and automatically send those to the lab to conduct and get the1799results out, the pace and the velocity of discovery will be1800dramatically improved.1801    Senator Young. So it is fair to say this could allow us to1802supercharge the pace of innovation?1803    Mr. Kratsios. Most certainly. And the NSF is already1804running ahead with a proposal around these cloud labs.1805    Senator Young. Very consistent with President Trump's1806branding, a golden age of innovation, this really could help1807usher that period in, I believe. I am going to pivot now to1808standards as it relates to AI and the impact of a lack of1809certainty for innovators seeking to develop and deploy AI.1810Congress is notorious for being late to the punch when it comes1811to development of standards and regulations.1812    And as other countries move forward in adopting their own,1813American companies are then subject to potentially differing1814rules across the globe. Can you speak to the risks associated1815with continuing to subject our AI innovators to a fragmented1816series of rules, including those enforced by other countries,1817as well as states here at home?1818    Mr. Kratsios. Yes. I think creating standards at the U.S.1819level that are prominent globally is very important. In the1820weekend after the AI Action Plan was released, the PRC held1821their large AI conference in Shanghai. And one of the main1822thrusts of their own AI Action Plan they released in response1823to ours was a desire to create a global entity, an AI entity in1824Shanghai that would then promulgate global rules around AI for1825the world.1826    And this is an example of why it is so important for the1827U.S. to be the leader in the way that we provide standards1828around AI, particularly around model evaluation and standard1829setting. And this is something that we know our adversaries are1830going to try to compete with us on. So it is more important1831than ever that we do that.1832    Senator Young. Yes. It is not just an issue of1833interoperability. I mean, you could literally make the argument1834that our values are embedded in the standards of our1835technologies. And so we want to have the ability to define what1836those standards are, and then allow the export-oriented1837economies, China in particular, to have to sell into our1838market--game, set, and match.1839    Before I yield back, I want to mention that Ranking Member1840Cantwell and I plan to re-introduce a revamped version of our1841Future of AI Innovation Act. This is vital legislation that1842will authorize the newly renamed Center for AI Standards and1843Innovation at NIST to promote the development of voluntary1844standards. Will you commit to working with us on the Future of1845the AI Innovation Act as we revamp it for this Congress,1846Director Kratsios?1847    Mr. Kratsios. Yes, we would love to see more there and work1848with you on it.1849    Senator Young. Thank you so much. As you have indicated in1850your testimony, there are many opportunities for Congress to1851work with the Administration to take action for American AI1852leadership, and I hope the Committee will do just that.1853Chairman, thank you very much.1854    Senator Budd. Senator Hickenlooper.18551856             STATEMENT OF HON. JOHN HICKENLOOPER,1857                   U.S. SENATOR FROM COLORADO18581859    Senator Hickenlooper. Thank you, Mr. Chair. Mr. Kratsios,1860thank you so much for being here. I think the White House1861Office of Science and Technology Policy is one of the most1862crucial positions right now, just given not just AI, but so1863many of the issues around research and the appropriate use of1864research. But I will keep myself focused to the AI. States from1865Texas to Colorado, Utah to California, passed, as you have1866mentioned, has been discussed, AI legislation.1867    In many cases, some of this action should inspire us to1868take a closer look, by us I mean Congress, to what do we need1869in a comprehensive national AI law. It might include periodic1870impact assessments to evaluate potential risk on AI models,1871transparency disclosures to users describing AI models in terms1872of use and capabilities. Obviously need R&D for support for1873standards development to identify and detect AI-generated1874content, and transparency around that.1875    Privacy protections for certain types of data being used to1876train AI models. So, do you feel that these are the types of1877policy principles that appear worthy to include as a foundation1878for a Federal AI law, if we were going to try and create1879something that would apply evenly across states?1880    Mr. Kratsios. Yes, my general sense and something I have1881advocated for, for many years, is that the best approach to AI1882regulation is for it to be use case and sector specific, not1883broad and sweeping. I think any attempt to create a singular AI1884regulation will lead you down the path that the EU is down1885right now, which has ultimately resulted in a pretty sad1886situation broadly for the innovators there.1887    Trying to create a singular AI rule for a technology that1888is so ubiquitous is actually not probably the best path1889forward. And one that we have advocated for both in the Action1890Plan and agencies is that, you know, obviously the rules that1891you would need at FDA to regulate and impart medical diagnostic1892are very different than the rules that you would need at the1893Department of Transportation for a self-driving car.1894    And we already have a system that has a very rich history1895in allowing our regulators to update their regulatory regimes1896with new technologies as they come. And it is one that I know1897all of our Secretaries across the cabinet are working very hard1898to make sure that they are up to speed on regulations that1899apply to AI, which fall within their domain.1900    Senator Hickenlooper. I get that. I understand that, but I1901think some things like making sure that the public is able to1902identify and recognize what is--you know, what is AI and what1903is not seems like something that is more general.1904    Mr. Kratsios. Yes, I think something like that in the1905research, particularly being able to identify AI-generated1906content, as you mentioned, is very important to continue to1907fund.1908    Senator Hickenlooper. Great. Appreciate that. The Action1909Plan calls for Federal agencies to conduct independent1910evaluations of AI systems before they are procured and1911deployed. Independent evaluation will help enhance security and1912increase trust, prevents companies from grading their own1913homework, as we would say, after an AI model is developed.1914    We have a bill introduced called Validation and Evaluation1915for Trustworthy AI Act, VET Act, with Senator Capito, which as1916Senator Young was mentioning, peripheral to requiring NIST to1917publish voluntary guidelines for companies to independently1918evaluate AI models. Can you describe how advanced you currently1919see the field of AI evaluations?1920    Mr. Kratsios. I think it is certainly not advanced enough.1921My number one priority for NIST would be to work on the very1922hard science associated with model evaluation and metrology.1923Our ability to understand how to even evaluate these models is1924still not complete.1925    So many people jump immediately to the evaluation itself,1926this question of what we should be evaluating, versus what I1927think the more important question of today is how do we1928evaluate these models. And what NIST can do, very important,1929metrology work on the how question.1930    And once we know how to actually evaluate these models,1931then each agency, each industry, whoever wants to do an eval1932will then have a standardized, scientifically backed way to be1933able to do the eval itself.1934    Senator Hickenlooper. Got it. In terms of the workforce1935development, that is going to be a key part here. The Action1936Plan highlights the need for AI skill development to make sure1937that we have a trained workforce that can do the work required.1938    Obviously a national security imperative essential to1939maintaining global competitiveness as you have mentioned. I1940think apprenticeship programs are a big part of that. We worked1941on Career Wise and created that in Colorado back when I was1942Governor, and it is now in 20 states.1943    They have been a national leader expanding youth1944apprenticeships and are already adding AI technology to support1945their programs. How can you work with--work to support1946innovative apprenticeship pathways, both for youth and adults,1947to equip an AI-ready workforce?1948    Mr. Kratsios. I think there is no President more excited1949about apprenticeships than this one. I think our Secretary of1950Labor has also had a big commitment to do a million new1951apprenticeships in this term. So there are big partners in the1952Department of Labor to partner with you guys.1953    Senator Hickenlooper. OK. Thank you. I yield back.1954    Senator Budd. Thank you. Senator Klobuchar.19551956               STATEMENT OF HON. AMY KLOBUCHAR,1957                  U.S. SENATOR FROM MINNESOTA19581959    Senator Klobuchar. Thank you very much, Mr. Chairman. And1960thanks to Ranking Member Baldwin as well. I am not actually on1961this subcommittee, but as our witness knows, I care a lot about1962this, and so I have been able to listen to my colleagues'1963questions and I want to thank them for their good work.1964    AI, we all know, huge potential, but also huge downside if1965we don't get this right. And I think David Brooks put it well1966when he said, ``I have found it incredibly hard to write about1967AI because it is literally unknowable whether this technology1968is leading us to heaven or hell''. So if we want it to lead us1969to heaven, I think we are going to have to find some guardrails1970and the like to protect us from fraud, to protect content1971creation, and our democracy.1972    So first off, I appreciated working with the Administration1973on the Take It Down Act. My bill with Senator Cruz to enable1974victims of non-consensual porn, including those generated by1975AI, to require the social media platforms to remove it within197648 hours. But there is many more problems, as you know, as I1977just experienced and wrote about it in a piece in the New York1978Times with AI, a deepfake on me that many people, believe it or1979not, thought was real.1980    And one platform took it down, one platform put created by1981AI on it, and then one platform, X, would not do anything, and1982it got over a million views. So the No Fakes Act that Senators1983Coons, Blackburn, and Tillis, and I have introduced would1984establish additional rules of the road. And do you agree that1985we should protect people from having their likeness replicated1986through AI, take down unauthorized deepfakes?1987    To me, it is some regime where--within the realm of the1988Constitution where some of it is labeling just digitally1989altered because it is parity, and you are not allowed to take1990it down. But then some of the stuff which you would be--would1991in a minute take down if someone played a video in this room or1992put up a sign, you should take down. So could you talk about1993that?1994    Mr. Kratsios. Yes, I think that I directionally generally1995agree with you. I think it is something that we should1996certainly look at, both the Executive Branch and the1997Legislative Branch. I think the Take It Down Act is a great1998example of something that is on one side of the line that1999certainly should become law when it did. But I think it is2000something that as this technology develops and becomes more2001proliferated, I think we have to find ways to solve it.2002    Senator Klobuchar. Thank you. I just hope our colleagues2003see that it is not one side or the other, right. There is some2004of the stuff that you are going to Constitutionally be able to2005take down, and we should require they take it down. Then there2006is some stuff that we can say should be labeled digitally2007altered, and it puts a burden on these platforms, but at least2008it will protect innocent people when they see it to know that2009it is not true.2010    And it continues to just amaze me that we all just sit by2011and act like, oh, that is too much, that is too little, instead2012of actually getting a solution. And I did--I really appreciate2013the work that Senator Schumer, and Senator Young, and Senator2014Heinrich, Senator Rounds did in bringing us together in the2015last few years on this.2016    Senator Thune and I have a bill that we introduced last2017year to set up basic guardrails for some of the non-defense2018riskier applications of AI, and in the past you have supported2019developing thoughtful Federal standards that can drive the2020widespread adoption of AI technologies across industries. And2021will you commit to work with Senator Thune and I on that bill?2022I know there are others as well.2023    Mr. Kratsios. Yes, happy to work on that.2024    Senator Klobuchar. OK, very good. And then in yesterday's2025hearing Senator Blackburn and I had a lot of attention on this2026hearing with two whistleblowers from Facebook just yesterday in2027our subcommittee in Judiciary. And we heard that one of the2028leading AI chatbot developers, Meta, deliberately and routinely2029altered, suppressed, and even deleted safety research,2030including on youth safety.2031    And there were many Senators participating in this hearing2032across the board. And I am concerned about this neglect when it2033comes to AI development on figuring out how we can protect2034these kids. You are right, we did get some with the President's2035support on the Take It Down Act, but that is only a subsection.2036    We have got fentanyl. We have got drugs being sold just2037overall on the internet, irrespective of AI, but then we have2038all this stuff going on with the AI chat boxes. And could you2039talk about your commitment to work with us on addressing the2040harms caused by AI chat boxes?2041    Mr. Kratsios. Would very much like to work with you guys on2042a lot of these issues. I think last week we held an AI2043education task force meeting which the First Lady joined and2044chaired. This was something that came out of the Executive2045Order the President signed a few months ago which shows the2046Administration's commitment toward K through 12 education AI.2047    And it is not how to necessarily use AI to do your homework2048or something. It is more important about teaching America's2049youth the limitations where AI works, where it doesn't work,2050and making young Americans understand how this technology2051works. And it is a very key component of making sure that they2052are using it in the way that it was intended for.2053    Senator Klobuchar. OK, thank you. And I did appreciate her2054support for our Take It Down bill, but again, it is just the2055beginning. So, thank you.2056    Mr. Kratsios. Thank you.2057    Senator Budd. Thank you very much. And thank you, Mr.2058Kratsios, for your testimony here today. I look forward to2059working with you, not just on AI, but also, as I mentioned2060earlier, thanking you for your work on supersonics and2061aviation.2062    Senators have until the close of business on September 172063to submit questions for the record. The witnesses will have--or2064the witness will have until the close of business on October 12065to respond to those questions. This concludes today's hearing.2066The Committee stands adjourned.2067    [Whereupon, at 11:44 a.m., the hearing was adjourned.]20682069                            A P P E N D I X20702071                   Partnership for AI Infrastructure (PAII)2072                                                  September 9, 202520732074Hon. Ted Budd,2075Chairman,2076Subcommittee on Science, Manufacturing, and Competitiveness,2077Senate Committee on Commerce, Science, and Transportation,2078Washington, DC.20792080Hon. Tammy Baldwin,2081Ranking Member,2082Subcommittee on Science, Manufacturing, and Competitiveness,2083Committee on Commerce, Science, and Transportation,2084U.S. Senate,2085Washington, DC.20862087Letter for the Record: Subcommittee Hearing on ``AI've Got a Plan:2088            America's AI Action Plan''20892090Dear Chairman Budd, Ranking Member Baldwin, and Members of the2091            Subcommittee:20922093    The Partnership for AI Infrastructure (PAII) respectfully submits2094this letter for the record in advance of the Subcommittee's hearing on2095America's AI Action Plan in support of Federal investments in2096artificial intelligence (AI) research and development (R&D). The2097Partnership commends the Subcommittee and its leadership for convening2098this hearing to examine the Action Plan and develop the legislative2099framework necessary to implement this AI strategy.2100    Implementing this Action Plan and promoting American AI innovation2101will require serious and sustained Federal investments. The One Big2102Beautiful Bill Act (OBBBA) provided the first critical down payment of2103$515 million in AI investments needed to jumpstart this Action Plan.2104This initial investment will enable the Department of Energy (DOE), the2105National Nuclear Security Administration (NNSA), and the Department of2106Defense (DoD) to begin developing the AI infrastructure needed to2107advance our national defense and scientific priorities. Leveraged2108strategically, this funding could lay the foundation for the Federal AI2109infrastructure ecosystem which future investments could build upon.2110    As a coalition of technology leaders, the Partnership for AI2111Infrastructure is committed to leveraging AI to accelerate2112technological breakthroughs, as demonstrated through our members work2113to build the world's three fastest supercomputers as part of the2114Exascale Computing Project. As part of this commitment, the Partnership2115submitted comments in response to the White House Office of Science and2116Technology Policy (OSTP) Request for Information (RFI) on the2117development of America's AI Action Plan. The Partnership's2118recommendations were as follows:21192120  1.  Promote sustained Federal investment in high-performance AI2121        infrastructure.21222123  2.  Create durable public-private partnerships to harness private2124        sector innovations to further advance Federal science and2125        national security objectives.21262127  3.  Stimulate America's pool of AI talent and retain experts within2128        the Federal workforce.21292130    The final Action Plan proposed a series of recommended policy2131actions that incorporated these concepts and outlined the Federal2132agency actions necessary to develop a Federal AI ecosystem. While2133Federal agencies can use the funding from the OBBBA to begin2134implementing the Action Plan, Congress must begin discussions on the2135legislative actions and additional investments that will be needed to2136see the plan to completion.2137Build the Federal AI Ecosystem2138    Developing the Federal AI ecosystem will require robust investments2139across the Federal government. The OBBBA began this work by providing2140$150 million to DOE to build transformational AI models, $115 million2141to NNSA to accelerate AI-lead nuclear national security, and $2502142million to DoD to advance the AI ecosystem. This $515 million in2143investments will pave the way to develop the AI systems envisioned by2144the Action Plan. Through the strategic stewardship of the OBBA2145investments, DOE, NNSA, and DoD can begin developing the mission-2146critical capabilities required to ensure America's long-term2147technological superiority.2148    To bring the truly innovative power of AI to bear to solve unique2149and challenging issues, our Federal agencies and national laboratories2150need to be equipped with leadership-scale AI infrastructure. Such2151systems can accelerate breakthroughs in science and bolster our2152national security systems against emerging threats. To fulfill the2153Action Plan's goals, these AI systems should be interoperable to enable2154seamless integration of the most cutting-edge AI hardware with2155interoperable graphics processing units. The Action Plan specifically2156tasks the National Institute of Standards and Technology (NIST) with2157identifying opportunities to accelerate and scale AI development,2158including the use of interoperable technologies. Setting2159interoperability as a core standard of Federal AI infrastructure will2160allow for GPUs and other AI hardware components to be changed modularly2161and used iteratively, which will lower costs, enable broad industry2162participation, and ensure the longevity and resiliency of Federal AI2163supercomputers.2164Develop Public-Private Partnerships2165    The Action Plan seeks to leverage the full talent and expertise our2166Nation has to offer through collaboration with industry leaders and2167stakeholders. By partnering with private companies with AI expertise,2168the Federal government will be able to harness the strength of American2169AI innovation and set it as the global standard. Companies that2170specialize in semiconductor design, super computers, AI software2171frameworks, and other areas of expertise will help drive AI development2172and adoption at the Federal scale. Leveraging that expertise through2173public-private partnerships leads to improved efficiency, shared costs,2174and cross-pollination of the best ideas from industry scientists and2175government researchers to bring cutting-edge innovations into use for2176national science, defense, and critical infrastructure priorities.2177Foster the Federal AI Workforce2178    Infrastructure alone will not deliver AI leadership without the2179talent to use it. Federal agencies must prioritize cultivating,2180training, and retaining AI scientists, engineers, and operators capable2181of managing and deploying advanced AI systems. The Action Plan included2182a provision to ensure that America's workers benefit from AI through2183the development of workforce training programs. Building and retaining2184a skilled Federal AI workforce will ensure continuity of expertise,2185safeguard sensitive programs, and strengthen the government's ability2186to develop and use AI systems to tackle national scientific and2187security priorities.2188                                  ***2189    The race to dominate AI is the defining technological challenge of2190our time and demands Federal investments to the scale of the Manhattan2191Project to secure America's global leadership. The Action Plan provides2192the blueprint for this project which the Administration and Congress2193can execute by authorizing and funding large-scale AI initiatives,2194advancing regulations that foster innovation, developing public-private2195partnerships, and building a robust and capable Federal AI workforce.2196The OBBBA made is the first investment towards America's AI future, but2197Congress must continue to provide sustained funding to build on this2198momentum.2199    As this Committee continues its work to chart America's AI future,2200the Partnership for AI Infrastructure and its members are ready to2201partner on this national initiative. We applaud the Committee's2202commitment to implementing America's AI Action Plan to secure our2203America's lead in the global AI race. Thank you for the opportunity to2204share our perspective and for convening this timely hearing.2205            Sincerely,2206                                      Partnership for AI Infrastructure2207                                 ______22082209                  Prepared Statement from Premier Inc.2210    Premier Inc. appreciates the Subcommittee's leadership and ongoing2211commitment to exploring thoughtful regulations that unleash the2212transformative potential of artificial intelligence (AI)--including in2213healthcare--while maintaining the efficacy, accuracy and transparency2214necessary to protect end users and patients.2215    As Premier stated upon its release, the White House's AI Action2216Plan sets a course towards secure, trustworthy artificial intelligence2217(AI) in healthcare. As Premier has emphasized to Congress and the2218Administration, continued U.S. leadership in AI is critical to2219America's health, economy and security. Premier is especially2220encouraged by the strategy's focus on incorporating healthcare voices2221into a sector-specific framework through the National Institute of2222Standards and Technology (NIST), training the clinical workforce of the2223future to harness AI's potential, and strengthening the cybersecurity2224of critical health infrastructure.2225    As we work together to imagine the future of healthcare, Premier2226looks forward to engaging in these new initiatives to ensure AI helps2227providers deliver better care at lower costs for America's patients. To2228advance the goals of America's AI Action Plan in healthcare, Premier2229encourages the committee to consider:22302231   The value of potential AI use cases in healthcare, including2232        applications in reducing administrative burden, clinical2233        settings, drug development and manufacturing, and healthcare2234        supply chain operations;22352236   The need for a national privacy law to maintain2237        competitiveness and innovation in the development and2238        implementation of AI;22392240   The importance of a clear regulatory framework prioritizing2241        transparency and risk mitigation in the development,2242        maintenance, and use of AI tools;22432244   The development of a clinical workforce capable of2245        maximizing the potential of AI technology in healthcare; and22462247   The critical nature of U.S. AI leadership to the security of2248        healthcare infrastructure.22492250    Our recommendations are described in greater detail below.2251I. BACKGROUND ON PREMIER INC.2252    Premier is a leading healthcare improvement company and national2253supply chain leader, uniting an alliance of 4,350 hospitals and2254approximately 300,000 continuum of care providers to transform2255healthcare. With integrated data and analytics, collaboratives, supply2256chain solutions, consulting and other services, Premier enables better2257care and outcomes at a lower cost.2258    A Malcolm Baldrige National Quality Award recipient, Premier plays2259a critical role in the rapidly evolving healthcare industry,2260collaborating with healthcare providers, manufacturers, distributors,2261government and other entities to co-develop long-term innovations that2262reinvent and improve the way care is delivered to patients nationwide.2263Headquartered in Charlotte, North Carolina, Premier is passionate about2264transforming American healthcare.2265    Premier has a wealth of operational experience leveraging AI2266technology to move the needle on cost and quality in healthcare,2267including:22682269   Premier Clinical Decision Support (CDS) designs AI-enabled2270        technology to reduce low-value and unnecessary care. Premier2271        CDS leverages natural language processing AI technology to read2272        unstructured data and ties it together with established2273        guidelines to generate real-time alerts and analytics, guiding2274        physician's decisions at the point of care. Premier CDS's2275        mission is to measurably improve the quality and safety of2276        patient care while reducing the costs by enabling context-2277        specific information integrated into the provider workflow.22782279   Premier Applied Sciences (PAS) is a trusted leader in2280        accelerating healthcare improvement through AI-powered2281        solutions that span the continuum of care and enable2282        sustainable innovation and rigorous research. Our services and2283        real-world data drive research and quality improvement in2284        pharmaceutical, device and diagnostic industries, academia,2285        Federal and national healthcare agencies, as well as hospitals2286        and health systems. PAS leverages Premier's robust data2287        resources to design and deploy AI-powered solutions for2288        clinical trial recruitment, and to help collate disparate2289        patient records to tell a complete patient story, leading to2290        higher-quality care.22912292   Premier's award-winning Supply Chain Disruption Manager2293        (SCDM) builds resilience and mitigates risks to the healthcare2294        supply chain by harnessing machine learning AI technology to2295        predict when critical drugs, devices and other medical supplies2296        are anticipated to become unavailable up to six weeks in2297        advance of a supply chain disruption. SCDM allows hospitals and2298        health systems to access clinically approved alternative2299        products to avoid delays in care or quality, and it allows for2300        communication to Federal agencies and other partners about2301        pending shortages to help proactively develop mitigation2302        strategies.23032304   Premier's purchased services subsidiary, Conductiv,2305        harnesses AI to help hospitals and health systems streamline2306        contract negotiations, benchmark service providers and manage2307        spend based on historical supply chain data. Conductiv also2308        works to enable a healthy, competitive services market by2309        creating new opportunities for smaller suppliers and helping2310        hospitals invest locally across many different categories of2311        their business.2312II. UNLOCKING THE VALUE OF AI IN U.S. HEALTHCARE2313Opportunities for AI to Reduce Administrative Burden2314    One of the biggest opportunities for the use of AI in healthcare is2315simplifying and improving standard, burdensome processes.2316    Premier has noted a recent surge of interest in patient-facing AI2317technologies in clinical settings, including ambient notetaking, care2318navigation chatbots and AI-powered radiological consultations. However,2319Premier, our members and others across the healthcare sector have been2320using AI technology to streamline burdensome administrative processes2321for years. Specific use cases include:23222323   Harnessing the power of natural language processing (NLP) AI2324        tools to ``read'' unstructured data in medical records to2325        efficiently build medical necessity documentation to expedite2326        prior authorizations;23272328   Automating burdensome processes such as procure-to-pay2329        workflows, supply chain contract activities, and revenue cycle2330        tasks;23312332   Overlaying AI chatbots on enterprise resource planning2333        software to help hospitals and health systems more efficiently2334        identify and manage their supply chain needs; and23352336   Leveraging predictive AI software to sift through large,2337        evolving datasets and proactively predict supply chain2338        disruptions to prevent interruptions in patient care.2339Opportunities for AI to Improve Patient Outcomes2340    Premier has partnered with leading healthcare providers and2341innovators to leverage AI in clinical workflows and improve patient2342outcomes. For example, early detection and intervention can improve2343patient outcomes and drastically reduce overall healthcare costs--for2344both patients and providers. Premier has demonstrated success in the2345following areas:23462347   The Premier team utilized artificial intelligence (AI), NLP2348        and a data ontology designed to mine the unstructured narrative2349        of clinicians' notes and pathology reports for statements such2350        as ``Mom seems a bit agitated'' or ``Mom is confused'' to2351        identify patients for early Alzheimer's Disease (AD)2352        intervention.23532354   In oncology, Premier partnered with AstraZeneca and2355        Clinithink and utilized Clinithink's CLiX NLP technology to2356        identify patients with incidental pulmonary nodules (IPNs) to2357        flag for intervention before potential lung cancer2358        progression--with roughly 152,000 patients ``caught'' early.23592360   Premier worked with GE Healthcare and St. Luke's University2361        Health Network to introduce a patient-centric care model for2362        breast cancer diagnosis--with a goal of helping patients go2363        from appointment to diagnosis and connection to a treatment2364        plan in just 48 hours or less.2365Opportunities for AI in Clinical Trials2366    Premier sees particular promise for the use of AI in streamlining2367processes and expanding patient access in clinical trials:23682369   Identifying trial participants: One of the biggest2370        challenges facing health systems that seek to participate in or2371        enroll patients in clinical trials is identifying and enrolling2372        patients in a timely manner. Delays in meeting trial enrollment2373        targets and timelines can increase the cost of the trial. AI2374        tools have the ability to analyze the extensive universe of2375        data available to healthcare systems to identify patients that2376        may be a match for clinical trials that are currently2377        recruiting. This application of natural language processing2378        systems can make developing new drugs less expensive and more2379        efficient, while also improving patient and geographical2380        diversity in trials to address generalizability.23812382   Generating synthetic data: AI, once trained on real-world2383        data (RWD), has the capability to generate synthetic data and2384        patient profiles that share characteristics with the target2385        patient population for a clinical trial. This synthetic data2386        can be used to simulate clinical trials to optimize trial2387        designs, model the possible effects or range of results of a2388        novel intervention, and predict the statistical significance2389        and magnitude of effects or biases. Ultimately, synthetic2390        patient data can help optimize trial design, improve safety and2391        reduce cost for decentralized clinical trials. Further,2392        synthetic control arms in clinical trials can help increase2393        trial enrollment by easing patient fears that they will receive2394        a placebo. To incentivize continued innovation, Premier2395        encourages Congress to urge the Food and Drug Administration2396        (FDA) to promulgate clear guidance on the process for properly2397        obtaining consent from patients for the use of their RWD to2398        produce AI-generated synthetic control arms in clinical trials.2399Opportunities for AI in Drug and Device Manufacturing2400    Premier sees potential for AI to transform at least three key2401segments of the drug and device manufacturing process:24022403   Supply chain visibility: Premier is confident that the2404        application of AI can advance national security by helping to2405        build a more efficient and resilient healthcare supply chain.2406        Specifically, AI can enable better demand forecasting for2407        products and services through analysis of historical and2408        emerging clinical and patient data, thereby driving better2409        inventory management by automating the monitoring and2410        replenishment of supplies.24112412    AI's ability to help drive supply chain visibility is particularly2413        helpful to address persistent healthcare supply chain2414        shortages. Oftentimes, the warning signals of an impending2415        product shortage can be seen weeks to months in advance due to2416        discrepancies in demand vs supply data. AI can create reliable2417        predictions that allow manufacturers to plan for and respond to2418        shortages or disruptions. AI also enables better planning and2419        response time to national or regional emergencies.24202421   Advanced process control: Another significant role for AI in2422        drug and device manufacturing is in the development and2423        optimization of advanced process control systems (APCs).2424        Process controls typically regulate conditions during the2425        manufacturing process, such as temperature, pressure, feedback2426        and speed. However, a recent report found that industrial2427        process controls are overwhelmingly still manually regulated,2428        and less than 10 percent of automated APCs are active,2429        optimized and achieving the desired objective. These2430        technologies are now ready to transform manufacturing on a2431        commercial scale; however, challenges still remain to2432        widespread adoption. Premier encourages Congress to urge FDA to2433        issue clear guidance that supports the industry-wide transition2434        to AI-powered APCs. Such technologies offer manufacturers the2435        opportunity to assess the entire set of input variables and the2436        effect of each on system performance and product quality,2437        automating plant-wide optimization. This application of AI2438        technology can transform the physical manufacturing of drugs2439        and devices, leading to cost-savings and increased resiliency,2440        transparency and safety in the healthcare supply chain.24412442   Quality monitoring: AI can also provide value-add to drug2443        and device manufacturing in the field of quality monitoring and2444        reporting. Current manufacturing processes provide an immense2445        volume of data from imagers and sensors that, if processed and2446        analyzed more quickly and efficiently, could transform2447        approaches to safety and quality control. AI models trained on2448        this data can be used to predict malfunctions or adverse2449        events. AI can also perform advanced quality control and2450        inspection tasks, using data feeds to quickly identify and2451        correct product defects or catch quality issues with products2452        on the manufacturing line. Taken together, these capabilities2453        can improve both the accuracy and speed of inspections and2454        quality control, helping companies to reliably meet regulatory2455        requirements and avoid costly delays that disrupt the drug2456        supply chain.2457III. CATALYZING THE AI MARKET THROUGH A NATIONAL PRIVACY LAW2458    A comprehensive national privacy law will have significant impacts2459on the development and competitiveness of AI technology in the United2460States. Federal data privacy laws should clearly outline pathways to2461acceptable data use for the training of AI models, which need not2462interfere with state-level requirements related to automated decision-2463making.2464    One of the greatest barriers to the large-scale diffusion of2465innovative AI applications in healthcare is the lack of a single2466Federal privacy law. A comprehensive, national privacy framework is2467necessary to provide a reasonable level of consumer confidence that2468businesses will protect their data, sensitive or otherwise. The current2469patchwork, state-driven approach to privacy policy has resulted in2470inconsistent data privacy practices that have amplified patient2471distrust in the healthcare system. Patients must be able to trust the2472processes for storage, handling and use of their data, particularly as2473patient data is increasingly used to train AI algorithms--missing data2474from patients wary of data sharing could add to data siloing and2475perpetuate AI algorithmic biases, and Congress should urge CMS, ASTP/2476ONC and other health data stakeholders to proactively educate patients2477on these tradeoffs.2478    Unharmonized and burdensome requirements are a rate-limiting step2479in unlocking value through AI innovation. A Federal privacy law can2480empower innovators by placing clear, harmonized and common-sense2481guardrails around artificial intelligence tools.2482    Premier was encouraged by the House Bipartisan Artificial2483Intelligence Task Force's commitment to enabling safe, trustworthy and2484innovative AI technology across healthcare. From drug development and2485manufacturing to diagnostics and clinical decision support, the Task2486Force's recommendations were in lockstep with Premier's long-standing2487advocacy for sensible regulatory guardrails for health AI.2488    Premier particularly appreciated the Task Force's recognition of2489AI's transformative ability to reduce administrative burden in2490healthcare and improve patient care. However, to fully realize the2491benefits of innovations such as real-time electronic prior2492authorization, Congress must address the fragmented state data privacy2493laws that are a barrier to bringing this technology to scale. Federal2494data privacy standards are essential to ensuring consistent2495protections, fostering equitable access and scaling AI-powered2496solutions effectively.2497    A comprehensive Federal privacy law should be viewed as an initial2498step towards achieving Congress's bipartisan AI goals. Premier2499recommends the Committee prioritize addressing the following regulatory2500gaps, which are particularly necessary in the healthcare sector:25012502   Quality: Federal policy should clarify what uses of data are2503        acceptable during AI training and testing, what patient consent2504        for data use looks like for AI and what standards AI companies2505        must meet to protect patient data. By removing uncertainty,2506        Congress can give AI developers permission to innovate.25072508   Security: Security and privacy often go hand in hand. A2509        Federal privacy law gives Congress the opportunity to clearly2510        dictate to regulators what appropriate security looks like to2511        protect patient health data, including when it is used in AI2512        models. By placing guardrails around data use and privacy,2513        Congress can limit the potential harms of security flaws in the2514        AI tools that are increasingly commonplace in healthcare.25152516   Market leadership: Baseline privacy requirements in a2517        Federal law--preempting state privacy laws--levels the playing2518        field for AI innovators while promoting consumer trust and2519        responsible AI. A fragmented state privacy law landscape2520        disadvantages startups and innovators, complicating compliance,2521        increasing regulatory burden or confusion, and adding2522        prohibitive cost to growth.2523IV. REDUCING REGULATORY UNCERTAINTY2524    Reducing regulatory uncertainty around AI development and2525deployment in healthcare settings is crucial to unlocking its2526transformative potential. Premier supports the responsible development2527and implementation of AI tools across all segments of American2528industry--particularly in the healthcare industry, where numerous2529applications of this technology are already improving patient outcomes2530and provider efficiency.2531    Premier strongly supports AI policy guardrails that include2532standards around transparency and trust, risk and safety, and data use2533and privacy. These recommendations will inform and complement the2534development of a healthcare-specific set of national standards for AI2535at NIST.2536Promoting Transparency2537    Trust--among patients, providers, payers and suppliers--is critical2538to the development and deployment of AI tools in healthcare settings.2539To earn trust, AI tools must have an established standard of2540transparency. Some policy proposals, including those proffered by the2541Office of the National Coordinator for Health Information Technology2542(ONC), suggest transparency can be achieved through a ``nutrition2543label'' model, which lists the sources and classes of data used to2544train the algorithm. Unfortunately, some versions of the ``nutrition2545label'' approach to AI transparency fail to acknowledge that when an AI2546tool is trained on a large, complex dataset, and is by design intended2547to evolve and learn, the initial static inputs captured by a label do2548not provide accurate insights into an ever-changing AI tool. Further,2549overly intrusive disclosure requirements around data inputs or2550algorithmic processes could force AI developers to publicly disclose2551intellectual property or proprietary technology, which would stifle2552innovation.2553    Premier recommends that AI technology in healthcare should be held2554to a standardized, outcomes-focused set of metrics, such as accuracy,2555false positives, inference risks and recommended use/applications.2556Outcomes, rather than inputs, are where AI technologies hold potential2557to drive health or harm. Thus, Premier believes it is essential to2558focus transparency efforts on the accuracy, reliability and overall2559appropriateness of AI technology outputs in healthcare to ensure that2560the evolving tool does not produce harm.2561    Premier has heard from multiple member hospitals that the lack of2562clear vendor information about the use of AI and associated liability2563actively deters them from purchasing or using AI tools. The lack of2564personnel and budget to collect information on data use, cybersecurity,2565and liability terms from vendors exacerbates this issue for all but the2566biggest health systems.2567    Premier urges Congress to consider requiring AI developers and2568manufacturers to list the acceptable uses of new technologies in2569healthcare settings, which would provide much-needed guidance to2570clinicians and providers on safe and appropriate use cases. This2571approach could provide liability protection for the proper use of AI2572technology for the defined set of use cases where developers have2573established and reported the appropriate metrics for accuracy and2574reliability. Transparency about the intended use of AI tools would be2575the simplest way for regulators to incorporate AI governance into2576existing regulations. Health systems would be also able to incorporate2577this information into their own governance structures, putting internal2578policies in place to prevent misuse of AI in ways that could be2579detrimental to patient safety or experiences.2580    Such disclosure does not inherently carry with it any additional2581significant cost or requirements. It would only give health systems and2582patients a complete picture of the safety and security of the AI2583technologies they use. Rather than limiting or delaying innovation,2584such guidelines would level the playing field between established2585market leaders and startups while providing clear transparency for2586providers and patients.2587    Alternatively, Congress could sanction the use of third-party2588certification organizations or existing market processes to address2589this challenge while reducing administrative burden. As a GPO, Premier2590already requires vendors and suppliers to submit information about uses2591of AI, data and cybersecurity certifications, and AI standards. This2592information is available to members when they make contracting2593decisions, providing a clear market incentive for venders and2594developers to meet industry best practices. The contracting process2595also includes model legal language around cybersecurity and AI best2596practices and liability sharing, driving best practices even in the2597absence of regulation. Much like the Payment Card Industry Data2598Security Standards (PCI DSS), Congress can leverage existing market2599incentives and self-governance to encourage broader adoption of2600transformative AI technologies. Such a market-driven approach would be2601flexible and adaptable, capable of adjusting to the latest developments2602in AI technology without requiring Congress or regulators to reimagine2603the law every year.2604Mitigating Risks2605    It is important to acknowledge potential concerns around2606``hallucinations'' and biased outcomes resulting from the use of AI2607tools in healthcare, which carry considerations for patient safety.2608Fortunately, there are several best practices that Premier and others2609at the forefront of technology are already following to mitigate these2610risks.2611    First, we reiterate Premier's recommendation for standardized,2612outcomes-based assessments of AI technologies' performance, which would2613hold developers accountable for reporting improper outputs. Premier2614also supports the development of a standardized risk assessment, which2615should identify detailed explanations of recommended uses for the tool2616and risks that could arise should the tool be applied inappropriately.2617    Additionally, Premier understands the importance of data standards,2618responsible data use and data privacy in the development and deployment2619of AI technology. Premier encourages Congress and regulators to work2620closely with developers, vendors and other stakeholders to ensure that2621any data standards that the Federal government codifies align with2622industry-experienced best practices. Premier also supports the2623establishment of guidelines for proper data collection, storage and use2624that protect patient rights and safety. This is particularly important2625given the sensitivity of health data.2626V. TRAINING THE HEALTHCARE WORKFORCE OF THE FUTURE2627    The White House's AI Action Plan prioritizes the education and2628training of a future workforce capable of harnessing AI's2629transformative potential. Premier agrees, and we believe technology can2630and should work alongside and learn from healthcare professionals, but2631current technology will not and should not replace the healthcare2632workforce.2633    To ensure clinical validity and protect patients, Premier2634recommends clear labeling of recommended use(s) and Federal support for2635healthcare workforce trainings that combat automation bias and2636incorporate human decision-making into the use of AI technology in2637healthcare. Automation bias refers to human overreliance on suggestions2638made by automated technology, such as an AI device. This tendency is2639often amplified in high-pressure settings that require a rapid2640decision. The issue of automation bias in a healthcare setting is2641discussed at length by the FDA in guidance on determining if a clinical2642decision support tool should be considered a medical device. Premier2643suggests that future guidance or standards for the use of AI should2644consider automation bias in risk assessments and implementation2645practices, such as workforce education and institutional controls, to2646minimize the potential harm that automation bias could have on patients2647and vulnerable populations.2648    Premier acknowledges the risks of automation bias and fully2649automated decision-making processes. To reduce these risks, promote2650trust in AI technologies used in healthcare and achieve the goal of2651supporting the healthcare workforce through AI, Premier recommends that2652healthcare workforce training programs provide comprehensive AI2653literacy training. Healthcare workers deal with high volumes of2654incredibly nuanced data, research and instructions--a growing2655percentage of which may be supplied by AI. This is particularly true2656for applications of AI in drug development, where manufacturers and2657quality control specialists may be reviewing high volumes of AI-powered2658recommendations or insights and making rapid decisions that affect the2659safety of patients. By ensuring our healthcare workers understand how2660to evaluate the most appropriate AI use cases and appropriate2661procedures for evaluating the accuracy or validity of AI2662recommendations, we can maximize the advisory benefit of AI while2663mitigating the risk to patients and provider liability.2664    To ensure that future clinicians can realize the benefits of AI and2665appropriately incorporate new technologies into patient care, Congress2666should encourage medical schools and accreditation programs to develop2667curricula for the healthcare workforce that incorporates digital health2668technologies. Among providers, there is a growing acceptance of2669technology as a workforce extender, particularly when it is seamlessly2670integrated into clinical workflows, and an increasing share of the2671healthcare workforce is open to adopting new tools. As a sector,2672healthcare must find ways to integrate digital health technologies into2673educational curriculums at all levels, including professional2674certifications and continuing education.2675    Finally, health systems and providers need to understand how to2676best realize the opportunities for AI and new technologies to enhance2677and extend care delivery to larger patient populations. Congress should2678encourage the development of evidence-backed models to evaluate the2679success of virtual care and virtual nursing programs. Anecdotal2680evidence indicates that practitioners believe in the value of virtual2681care to balance workload and expand access to care, particularly in2682rural areas. Optimized and evidence-backed models have the potential to2683improve access to care in rural communities, increase savings and2684reduce chronic disease costs. In the face of clinician shortages--2685especially nursing shortages--the existence of a center of excellence2686for optimized virtual nursing services could provide care to entire2687regions, offsetting workforce capacity challenges and reducing brain2688drain.2689VI. SECURING U.S. LEADERSHIP2690    As the AI Action Plan acknowledges, America's digital2691infrastructure faces a regulatory inflection point spanning from2692enabling to emerging technologies. Premier believes that true supply2693chain resiliency requires a holistic approach as part of a larger2694strategy to address the implications of policy on products needed in2695healthcare--particularly those needed during a public health crisis or2696national security threat.2697    Tariff and trade policies directly influence the availability and2698affordability of critical medical supplies and technologies, including2699the availability and uptake of AI tools. China has spent the past2700decade making a play for global leadership at every level of technology2701from semiconductors to AI models, leaving healthcare's future2702increasingly reliant on China's tech stack. Investments in the2703healthcare tech ecosystem--from semiconductors, cloud computing, and2704connectivity through the software technology stack--can help American2705healthcare overcome shortages, build a reliable supply chain for2706medical devices, and put America back in control of healthcare's tech-2707enabled future.2708    How the U.S. regulates AI--and the enabling technologies that power2709it--will transform healthcare, one way or the other. Making America's2710healthcare system the most attractive in the world for innovators and2711visionaries, thereby reducing costs and improving patient outcomes, can2712occur only if lawmakers reimagine the technology-care delivery nexus2713from the bottom up.2714    The U.S. cannot afford to fall behind in the development and2715production of critical enabling technologies for the growth of the2716burgeoning AI sector, nor can it become reliant on AI applications and2717software developed by geopolitical adversaries.2718    America has learned a difficult lesson about the threat of becoming2719reliant on untrustworthy technology. From telecommunications2720infrastructure to solar power inverters to port cranes, much of this2721country's critical infrastructure has faced a reckoning about the2722threat that unsecured software and hardware pose to essential2723functions. A fresh focus on security must begin with trustworthy2724physical infrastructure. Federal rulemaking has given the Coast Guard2725and Transportation Security Administration extended cyber authorities2726over ports, shipping, and rail. During the 118th Congress, lawmakers2727introduced several bills examining dependence on foreign-manufactured2728shipping cranes and other crucial technologies. While these are2729valuable initial steps to identify vulnerabilities in America's2730technology infrastructure, this country cannot afford to repeat the2731same mistakes.2732    Early and sustained AI leadership is essential to provide America's2733critical infrastructure--especially healthcare--with the reliable,2734trustworthy tools that it needs. The United States cannot afford to2735risk the future of digital health by ceding AI leadership.2736IV. CONCLUSION2737    Premier appreciates the opportunity to comment on the2738Subcommittee's work. If you have any questions regarding our comments,2739or if Premier can serve as a resource on these issues, please contact2740John Knapp, Vice President, Advocacy, at John_Knapp@premierinc.com.2741                                 ______27422743     Response to Written Questions Submitted by Hon. John Thune to2744                         Hon. Michael Kratsios2745    Question 1. Last Congress, Senator Klobuchar and I introduced the2746AI Research Innovation and Accountability Act alongside our colleagues2747Senators Wicker, Hickenlooper, Capito, and Lujan. This bipartisan2748legislation establishes a light-touch, pro-innovation framework that2749will bring transparency, accountability, and security to the2750development and operation of AI.2751    Do you agree that Congress must establish basic rules of the road2752like the framework we have laid out in this legislation?2753    Will you commit to working with us on this legislation during the2754119th Congress?2755    Answer. I look forward to working with you and your colleagues on2756any legislation that promotes and protects continued American2757leadership in AI innovation.2758                                 ______27592760  Response to Written Question Submitted by Hon. Marsha Blackburn to2761                         Hon. Michael Kratsios2762    Question 1. An important issue for Tennessee as it relates to AI is2763what happens with patent and copyrighted content, whether it is from an2764entertainer, author, a publisher, someone involved in online sales2765training or online human resources training, or religious leaders who2766have sermons or prayers on which they hold a copyright. In Nashville,2767we talk about fair use as being a fairly useful way to steal2768copyrighted content. We see that happen repeatedly. When I wrote an2769Amicus Brief on the correctly decided Warhol vs. Goldsmith case, I2770argued in favor of a narrowed application of the fair use doctrine.2771    When it comes to permissible training materials for LLMs, clearly,2772there is a difference of opinion. This is a reason why states have2773played such an important role in stepping forward, as Congress has2774proven incapable of passing legislation to protect content creators.2775Making certain that copyrights, patents, and trademarks are not2776infringed is vital to our creative community. I had a group in my2777office recently, who highlighted concerns about this issue of2778unauthorized training. They are also looking at what is happening with2779OpenAI's AI-generated movie, Critterz. This full-length, box-office2780movie will be made almost entirely using AI, including AI-generated2781music. By allowing LLMs to train on copyrighted materials, this takes2782away the creative community's Article 1, Section 8, Clause 82783constitutional right to exercise their craft.2784    I would like to have your response on addressing these vital2785issues. How do you plan to approach firewalling copyrighted content in2786training LLMs while still allowing training on current events or2787conversations?2788    Answer. At the launch of the AI Action Plan, the President stated2789that AI developers should be allowed to use the facts and information2790from content like books or articles to develop general purpose AI2791models without navigating complex copyright negotiations. In his2792speech, President Trump also recognized the distinction between2793training AI systems using the facts and information from copyrighted2794works versus having the AI's outputs copy or plagiarize a creator's2795work. The Administration is closely tracking ongoing court cases2796relating to AI training on copyrighted materials.2797                                 ______27982799   Response to Written Questions Submitted by Hon. Maria Cantwell to2800                         Hon. Michael Kratsios2801AI Skilling and Workforce Development2802    One pillar of the AI Action Plan is empowering American workers2803through AI education and job training. The plan specifically calls for2804initiatives like AI-focused apprenticeships and skilled trades training2805(e.g., more electricians and advanced HVAC technicians to build AI2806infrastructure).28072808    Question 1. What progress has been made on these workforce2809programs?2810    Answer. It is critical that the U.S. has the domestic workforce2811needed to support growing demands for AI infrastructure. America's2812Talent Strategy, co-released by the Department of Labor (DOL),2813Department of Commerce (DOC), and Department of Education (ED), focuses2814precisely on developing these workforce programs. As part of the2815implementation of the Talent Strategy, DOL has announced at least $302816million for the Industry-Driven Skills Training Fund grant program2817administered by DOL's Employment and Training Administration. These2818grants will help train American workers for jobs in AI and other2819emerging and high demand areas. As of August 2025, DOL has identified2820over 120 AI-centric Registered Apprenticeship programs and over 2,0452821apprentices, in over 45 AI-centric occupations. Additionally, DOL has2822confirmed that there are over 350,000 active apprentices in AI2823Infrastructure Registered Apprenticeship programs. The Administration2824is working to increase the number of active apprentices within these2825occupations, in alignment with the AI Action Plan and America's Talent2826Strategy. Furthermore, the National Science Foundation (NSF) has taken2827steps to strengthen AI-focused career and skill building learning2828opportunities for high school students, including curriculum2829development, dual enrollment, micro-credentials, and hands-on2830experiential learning to prepare America's workforce for the future.28312832    Question 2. Is the administration, in partnership with the2833Department of Labor and industry partners, planning to roll out new2834training curricula or apprenticeships in the regions where AI data2835centers and projects are expanding?2836    Answer. In response to the President's AI Action Plan and Executive2837Order on Advancing Artificial Intelligence Education for American2838Youth, the Administration is working with industry and academia to2839prepare workers to fill critical AI roles across the country. DOL, DOC,2840and ED co-released America's Talent Strategy, which includes a focus on2841scaling apprenticeships to meet AI infrastructure workforce demands. In2842alignment with this goal, the DOL recently announced nearly $84 million2843in grants to 50 states to increase the capacity of Registered2844Apprenticeship programs. My staff are working with DOL on implementing2845the AI Action Plan's education and workforce training recommendations.28462847    Further, the AI Action Plan provides an important and meaningful2848focus on training a skilled workforce to build, operate and maintain an2849AI infrastructure. We know there are hundreds of thousands of jobs that2850will be created in the coming years, but too few workers to fill those2851jobs.28522853    Question 3. Are current Federal programs and funding sufficient to2854meet these needs?2855    Answer. Targeted Federal programs, including public-private2856partnerships, apprenticeships, industry-driven training programs, and2857state and local-led workforce initiatives, can help meet the growing2858workforce demands needed to support domestic AI infrastructure. The AI2859Action Plan recommends refocusing existing Federal programs and working2860closely with industry, educators, and state and local governments to2861identify gaps in employment pipelines and train new workers to meet2862industry demand.2863Support for CAISI2864    A positive aspect of the AI Action Plan was the emphasis on NIST2865CAISI, the Center for AI Standards and Innovation. Last year, the House2866and Senate both passed bills out of committee to authorize an AI2867institute at NIST focused on AI standards and innovation. I'm glad the2868Administration and Plan is preserving this institute. In order for the2869U.S. to lead however, we need to commit to fully funding and resourcing2870it.28712872    Question 4. Mr. Kratsios, can you commit to supporting2873Congressional codification of the NIST CAISI to develop voluntary2874standards and testbeds related to national security for AI frontier2875models?2876    Answer. The AI Action Plan recommends investments in the2877development of AI testbeds that span many sectors, including2878agriculture, transportation, and healthcare. NIST plays a role in2879leveraging its technical expertise to advance AI measurement science2880and sector-specific standards that will promote secure AI innovation2881and accelerate broad AI adoption across sectors. You have my commitment2882to work with you and your colleagues on legislation as it relates to2883NIST and other Committee priorities.2884Energy Needs and R&D for Fusion Energy2885    The growing demand for electricity to power AI data centers is2886staggering. By some estimates, global electricity demand from data2887centers is projected to more than double by 2030 exceeding 9452888terawatt-hours (TWh). It will strain electric grids and energy2889providers. A potentially limitless source of clean and inherently safe2890energy is fusion, a source that could provide vast amounts of2891predictable baseload power to increase the reliability of our energy2892grid. Analysts at Bloomberg estimate that this game-changing technology2893could achieve a potential $40 trillion valuation. Washington State has2894become a fusion energy hub with billions of dollars invested and three2895prominent start-up companies looking to deploy demonstration projects.28962897    Question 5. How would the attributes of fusion energy help the2898reliability requirements of the grid for AI?2899    Answer. Commercial fusion can unlock a new source of reliable2900energy to help meet the growing energy needs of the grid and data2901centers across the United States. Nuclear fusion is an important2902priority for American energy dominance.29032904    Question 6. How can the government partner with the private sector2905to scale fusion technology as it continues to develop?2906    Answer. Milestone-based funding, prizes, challenges, public-private2907partnerships, and other novel funding mechanisms can incentivize2908commercial development of fusion technology. Since the first Trump2909Administration, the DOE has focused on improving commercialization of2910domestic fusion research. For example, DOE recently announced $1342911million in funding for Fusion Innovation Research Engine (FIRE)2912Collaboratives and the Innovation Network for Fusion Energy (INFUSE),2913which encourages collaboration among the fusion industry, DOE national2914labs, and universities.2915Public Investment in Science2916    Government investment in fundamental science has been the backbone2917of American success in technology and innovation. If the United States2918wants to outcompete foreign adversaries, it cannot slash funding for2919the National Science Foundation, National Institute of Standards and2920Technology (NIST), Department of Energy labs, or STEM education2921programs that power the AI workforce and ecosystem.29222923    Question 7. What impact will cuts to Federal funding for science2924and research at universities have on U.S. competitiveness in AI?2925    Answer. The Trump Administration took long-needed action to re-2926focus the Federal research enterprise towards areas of national2927strategic priority and geopolitical importance. The President has taken2928extensive executive actions to create a more conducive environment for2929American innovation, unlock investments in AI infrastructure at home2930and abroad, advance AI for education, leverage AI for developing cures2931for pediatric cancer, and much more. These actions remove barriers for2932innovators to promote American leadership in AI and accelerate the2933export of the American AI technologies, positioning the United States2934to dominate in this critical technology and to win the AI race.2935    Notably, the President's FY 2026 budget proposal preserved funding2936for programs such as AI and quantum. In the One Big Beautiful Bill Act2937(OBBBA), the President committed 150 million dollars in new funding for2938the DOE national labs to curate, structure, and preprocess scientific2939data for use in AI and machine learning models. This data will be2940critical in pushing forward next generation computational analysis,2941accelerating scientific discovery, and further solidifying U.S.2942leadership in AI and computational science.2943    In addition, OSTP and the Office of Management & Budget (OMB)2944recently released the annual memorandum on the Administration's FY272945Research and Development Budget Priorities. The memorandum lays out a2946path to unrivaled American dominance in critical and emerging2947technologies, with AI as its first priority. It directs Federal2948agencies to make significant investments in foundational and applied AI2949research, critical digital infrastructure, and robust evaluation2950standards, aiming to advance breakthroughs in AI architecture,2951interpretability, security, and capabilities. It further strengthens2952U.S. competitiveness by fostering close collaboration with industry and2953academia to promote commercialization and workforce development, expand2954STEM education pathways, facilitate broad adoption of AI-enabled tools,2955and support resilient critical infrastructure.2956Bayh-Dole Act2957    Congress enacted the Bayh-Dole Act as a key piece of innovation2958policy. It allows universities and nonprofit institutions to retain2959title to federally funded inventions and license them to private2960companies. This framework has been critical to creating thousands of2961startups, new industries, and high-wage jobs in the United States.29622963    Question 8. Do you agree that maintaining the Bayh-Dole model,2964where universities and entrepreneurs can commercialize federally funded2965research without the Federal government taking a large share of their2966revenue, is essential to sustaining America's innovation ecosystem?2967    Answer. Basic research is critical to the technological revolutions2968which may occur decades in the future. Furthermore, it is important to2969incentivize commercialization of basic research and technology transfer2970where it may have promising applications American industries.2971                                 ______29722973   Response to Written Questions Submitted by Hon. Tammy Baldwin to2974                         Hon. Michael Kratsios2975    Question 1. At the end of last year, this Committee held a hearing2976on how AI is enabling and exasperating the proliferation and2977sophistication of scams. In 2023, Wisconsinites lost $92 million to2978fraud and scams, and the problem is only getting worse. Representative2979Jamie Raskin and I have been leading an effort to direct the FTC to2980develop a comprehensive online resource that will serve as a2981centralized resource page for victims of financial scams and frauds.2982    What is the Trump administration doing to protect Americans against2983AI enabled scams?2984    Answer. Thank you for recognizing the role the FTC has to play in2985protecting Americans, both young and old, from scams and fraud. AI is2986not exempt from consumer protection laws, and law enforcement2987authorities at local, state and Federal level are able to enforce laws2988addressing fraud committed with AI, just as they do for any other2989medium used to commit fraud, such as e-mail or telephone. OSTP will2990further efforts to help young people become more literate in AI through2991the White House Task Force on AI Education.29922993    Question 2. While the United States continues to focus on the2994advancement of artificial intelligence, it is also essential that we2995continue to invest in the development and advancement of other emerging2996technologies. Quantum mechanics and computing have the potential to2997simulate and solve problems too complex for classic computers. Quantum2998also has the potential to work hand in hand with artificial2999intelligence to continue to enhance its capabilities.3000    What is the Trump administration doing to leverage the development3001of other emerging technologies such as quantum computing to advance our3002development of artificial intelligence?3003    Answer. During his first term, President Trump was the first3004president to prioritize AI and quantum in his budget request to3005Congress. President Trump launched the National AI Initiative and3006signed the National Quantum Initiative Act into law, laying the3007foundation for continued American leadership in these fields. The3008President has continued to demonstrate his commitment to these3009technologies with his FY26 budget request, which includes robust3010funding for AI and quantum. Furthermore, OSTP and OMB released their3011annual memorandum outlining the Administration's FY27 Research and3012Development Budget Priorities. This memorandum prioritizes research on3013quantum and AI and calls out the interaction between the two fields.3014                                 ______30153016 Response to Written Questions Submitted by Hon. John Hickenlooper to3017                         Hon. Michael Kratsios3018AI-generated Content and Transparency3019    NIST has been conducting scientific research into new methods to3020identify or detect content generated by AI, such as texts, images,3021videos, and more. AI is a powerful tool capable of creating content3022that appears to be real life. Methods such as watermarks, content3023provenance, and labels are being evaluated for their accuracy.30243025    Question 1: Director Kratsios, from your perspective, what does the3026road ahead look like for scientific research into AI-generated content?3027    Answer. OSTP and OMB recently released the annual memorandum that3028outlined the Administration's FY27 Research and Development Budget3029Priorities. This memorandum calls out the importance of foundational3030and early-stage applied research in AI, including in interpretability,3031controllability, and adversarial robustness.3032    In addition, the President signed the TAKE IT DOWN Act into law,3033which targets sexually explicit, non-consensual deepfakes and creates3034market dynamics to develop tools to detect certain categories of AI-3035generated content.30363037    Question 2: Director Kratsios, when do you think it will be3038feasible for technical standards to be developed to promote3039transparency in synthetic content? Do you foresee any technical3040capabilities, research barriers, or technological limitations delaying3041the development of technical standards around synthetic content?3042    Answer. The AI Action Plan recommends actions to combat synthetic3043media in the legal system, including issuing guidance to explore a3044deepfake standard and file formal comments on proposed deepfake-related3045additions to the Federal Rules of Evidence. It also recommends3046developing NIST's deepfake evaluation program into a formal guideline3047and companion voluntary forensic benchmark.3048AI and Copyright Protections3049    Copyright protections for creators' works are being actively3050challenged in courts across the country. AI developers and national3051security interests argue copyright protections, including the ``Fair3052Use'' standard, could slow down the development of American-made AI3053technologies and cede global leadership in AI to competitors. During3054his remarks while unveiling the AI Action Plan, President Trump said3055the United States ``can't be expected to have a successful AI program3056when every single article, book or anything else that you've read or3057studied, you're supposed to pay for. You just can't do it because it's3058not doable.''3059    Our hope is to balance the rights and protections of content3060creators and lead the world in AI innovation.30613062    Question 3: Director Kratsios, how do you believe we could achieve3063this balance between protecting creators' rights and developing gold-3064standard AI technologies?3065    Answer. As President Trump stated during the launch of the AI3066Action Plan, AI developers should be allowed to use the facts and3067information from content like books or articles to develop general3068purpose models without navigating complex copyright negotiations. The3069President also recognized the distinction between training AI systems3070using the facts and information from copyrighted works versus having3071the AI's outputs copy or plagiarize a creator's work during that3072speech. The Administration is closely tracking ongoing court cases3073relating to AI training on copyrighted materials.3074AI Supply Chain3075    When Congress passed the CHIPS & Science Act, with support from3076Democrats and Republicans, we committed to growing high-tech3077manufacturing in the U.S., expanding our STEM workforce, and3078recommitting our investment in scientific research. The CHIPS Act3079incentives increase our ability to manufacture semiconductors in the3080U.S. to train AI models and power data centers.3081    The Trump Administration has recently proposed to take government3082equity in private companies who manufacture semiconductors as well as3083receive a portion of sales from certain semiconductors to China.30843085    Question 4: Director Kratsios, do you believe the CHIPS Act or the3086Export Control Reform Act explicitly allow the Federal government to3087take these actions? Have these actions been authorized by Congress?3088    Answer. The President has broad authorities when it comes to3089matters of national security. I understand that the Secretary of3090Commerce is implementing these actions through the appropriate3091mechanisms.3092Executive Orders3093    The White House unveiled three Executive Orders to accompany the AI3094Action Plan that seek to build more data centers, reform government3095procurement of AI models, and export American AI technologies3096internationally.3097    Meanwhile, agencies across the Federal government are working to3098carry out the AI Action Plan's goals.3099    Certain issues, such as protections for creators' copyrighted works3100in AI model development, remain legally unresolved.31013102    Question 5: Director Kratsios, do you believe the White House will3103need to issue any new Executive Orders, including on issues such as3104copyright, to continue implementing the AI Action Plan? Yes or no?3105    Answer. OSTP continues to coordinate interagency action to3106implement the extensive recommendations in AI Action Plan and the3107President's Executive Orders on AI.3108Data Centers3109    As the demand for AI applications skyrockets, so does the demand it3110requires from our electric grid.3111    We should make targeted investments to modernize our electric grid,3112expand transmission line capacity, and ensure reliable and affordable3113sources of power. However we need to ensure that we have the workforce3114to be able to build this infrastructure.31153116    Question 6: Director Kratsios, how can we improve access to a3117skilled workforce for building out data centers?3118    Answer. It is critical that the U.S. has the domestic workforce3119needed to support growing demands for AI infrastructure. America's3120Talent Strategy, co-released by DOL, DOC, and ED, focuses precisely on3121developing these workforce programs. As part of the implementation of3122the Talent Strategy, DOL has announced a $30 million Industry-Driven3123Skills Training Fund grant program administered by DOL's Employment and3124Training Administration. DOL also recently announced nearly $84 million3125in grants to 50 states to increase the capacity of Registered3126Apprenticeship programs. In addition, NSF continues to support programs3127to upskill the talent necessary to manage and secure large-scale data3128infrastructure across our country.3129    Federal programs, including public-private partnerships,3130apprenticeships, industry-driven training programs, and state and3131local-led workforce initiatives, can meet the growing workforce demands3132needed to support domestic AI infrastructure.31333134    Question 7: Director Kratsios, do you have an update on the3135Administration's plans to potentially site and construct AI data3136centers on Federal land?3137    Answer. The Trump Administration has been involved in ongoing work3138to accelerate the development of AI infrastructure through siting and3139constructing AI data centers on Federal Lands. In April, DOE issued a3140Request for Information and received enormous interest to inform data3141center siting and construction on Federal Lands. In July, DOE announced3142the four selected sites: Idaho National Laboratory, Oak Ridge3143Reservation, Paducah Gaseous Diffusion Plant and Savannah River Site.3144Furthermore, as part of the process for finding private sector partners3145to manage the projects, DOE released Requests for Applications to build3146and power AI data centers at Idaho National Laboratory, the Oak Ridge3147Reservation, and the Savannah River Site this September.3148Tariffs on Semiconductors3149    The Trump Administration has stated it intends to impose massive3150tariffs on imports of semiconductors into the U.S.3151    The CHIPS & Science Act provided targeted incentives for companies3152to build and expand manufacturing here in the U.S.3153    Semiconductors are a fundamental building block for developing AI3154models and other advanced technologies.31553156    Question 8: Director Kratsios, what potential impact would new3157tariffs on semiconductor imports have on the United States'3158competitiveness in AI?3159    Answer. The Secretary of Commerce is working to properly channel3160resources from the CHIPS & Science Act to expand domestic chip3161fabrication capacity and advanced research, making America the home of3162future AI breakthroughs. President Trump has cited these policies as3163central to achieving American supremacy in the AI race, establishing3164this as both an economic and national security imperative.3165Secure AI by Design3166    As organizations rapidly adopt AI to boost efficiency and growth,3167the surge in usage has significantly expanded the attack surface. A3168report released by a cybersecurity firm, Palo Alto Networks, indicates3169an 890 percent increase in GenAI traffic. This growth brings new3170security risks, requiring organizations to identify AI use, assess3171vulnerabilities, and implement real-time protections. The U.S.3172government's AI Action Plan reinforces this need, urging secure,3173resilient AI systems capable of detecting threats like data poisoning3174and adversarial attacks.31753176    Question 9: Director Kratsios, could you explain with detail how3177OSTP is considering integrating these secure-by-design AI principles?3178Additionally, what collaborative efforts with the private sector are3179underway to strengthen the secure development and deployment of AI?3180    Answer. The AI Action Plan recommends a range of different actions3181to advance secure-by-design AI technology. Additionally, the3182Administration is prioritizing R&D that enables the secure development3183and deployment of AI through the recently released OSTP and OMB annual3184memorandum on FY27 Research and Development Budget Priorities,3185including fundamental work on AI interpretability, controllability, and3186adversarial robustness.31873188    Question 10: Director Kratsios, in what ways does the AI Action3189Plan ensure the safe and secure use of AI systems within Federal3190networks--particularly in protecting against sophisticated cyber3191threats, data breaches, and unauthorized access?3192    Answer. The AI Action Plan highlights the importance of secure-by-3193design AI to minimize the marginal security risk contributed by3194deploying AI systems in Federal networks and calls for the development3195of standards for high security AI data centers. It recommends that the3196General Services Administration (GSA) creates and manages an AI3197procurement toolbox, in collaboration with OMB, to ensure that procured3198AI systems comply with relevant privacy, data governance, and3199transparency laws. Further, it recommends that NIST partner with3200industry to establish standards and best-practices to ensure impacts3201are minimized and response is timely. We will continue to work with the3202relevant agencies to strengthen existing cyber defenses and update3203security practices to prepare for AI-specific cybersecurity threats.3204U.S. Leadership in Global AI Governance3205    Under the AI Action Plan, the U.S. would meet the global demand for3206AI by exporting its full AI technology stack, including hardware,3207software, applications, and standards, to key markets overseas.32083209    Question 11: Director Kratsios, do you support robust U.S.3210engagement in key international organizations, including the UN, OECD,3211G7, G20, ITU, and ICANN, for shaping the global conversation around AI?3212How would you prioritize these fora, and what goals should the U.S. be3213pursuing there?3214    Answer. As I emphasized at the United Nations Security Council3215meeting, we totally reject efforts by international bodies to assert3216centralized control and global governance of AI. We are focused on3217establishing American AI as the global gold standard and enabling3218allies and trade partners to build their own sovereign AI ecosystems3219with secure American technology. OSTP continues to work with agencies3220across the Federal government to deliver on the President's Executive3221Order 14320 to promote the export of the American AI technology stack.3222AI & Advanced Communications3223    The convergence of AI and wireless infrastructure will have massive3224implications for the global telecommunications landscape. With AI-3225native 6G networks powering millions of devices and running critical AI3226applications, who builds and operates these networks is more important3227than ever.32283229    Question 12: Director Kratsios, how can we leverage an American AI-3230native 6G stack to compete with Huawei in emerging global markets?3231    Answer. Leveraging an American, AI-native 6G stack means3232accelerating secure domestic innovation, coordinating government and3233private sector research and development, and driving global standards3234to outcompete Huawei, especially in emerging markets. American 6G3235networks dramatically increases the resilience of our critical3236infrastructure and protects us from foreign surveillance or sabotage--a3237risk inherent with Huawei-backed systems. As outlined in the AI Action3238Plan, the removal of regulatory barriers will help the United States3239deploy and export next-generation telecommunications infrastructure3240faster than our competitors, ensuring that U.S. companies can scale up32416G deployments at home and abroad. This Administration will continue to3242prioritize the promotion of the American technology stack around the3243world.32443245                                  [all]