Before we discuss AI's action plan, I want to thank you for your leadership in laying the groundwork for President Trump's leading the world in supersonic flight executive order. It's another important field of innovation and one where we as a country have fallen behind. We haven't had a commercial Concord flight in over 20 years, and we have to stay ahead of China and cutting-edge aerospace technology.
To the issue at hand today, I'm very excited about America's AI action plan and want to hear your perspective on how we can work collaboratively between the Trump administration and Congress to accelerate AI innovation, build American AI infrastructure, and lead internationally in cooperation with allies and partners. Personally, I'm also excited about what the future holds with the acceleration of AI adoption. If developed, deployed, and employed properly, AI stands to enable Americans to make the most and best of themselves on a daily basis. We must ensure that our AI policy is anchored in maximizing economic opportunity for Americans. And I'm not just talking about the billionaires in Silicon Valley; I'm talking about everyday Americans waking up and going to work in family-sustaining careers enhanced by AI, but not replaced by it.
US leadership in technological innovation has been the accelerator that has boosted our economy and growth rates ahead of the rest of the world. General purpose technologies like the internet ushered in sustained years of economic growth, wage gains, new jobs, and increased productivity. Critically, US leadership allowed for the open internet and ecosystem built around it to reflect our national character of entrepreneurship and free expression. AI offers similar opportunities as a transformative general-purpose technology. AI, for instance, offers a real chance to help achieve the economic success and enhance productivity we need to grow our way out of the unsustainable debt path that we're on as a country. As your AI action plan rightly points out, the competition is fierce.
The Trump administration has made AI leadership a day one priority, as President Trump rescinded President Biden's AI executive order, which many feared was an over-regulatory European-styled approach, which would suffocate innovation in startups while ceding important ground to adversarial nations like China. The PRC has put forward plans to leverage state resources and capital to make China the global leader in AI by 2030. Through their top-down statist model, the PRC wants to direct capital and resources to favored firms to embed AI across industries including manufacturing, agriculture, robotics, and services. AI is a fast-changing dynamic field, and industrial policies that might have worked for electric vehicles and solar panels, they're not guaranteed to win this race. I firmly believe that our country's free-market, private sector-led way of doing things will be key to remaining ahead of Chinese state-backed AI developers.
To accelerate AI innovation, I look forward to hearing from you on how Congress can partner with the administration and industry to remove roadblocks and provide regulatory certainty to let innovators innovate. Chairman Cruz's AI regulatory sandbox bill will be very helpful here. The federal government can also continue to be a proactive partner, leading the way on adopting AI tools and solutions to streamline and improve government while also sending an important market signal and presenting a valuable use case. To build out American AI infrastructure, Congress needs to work on comprehensive permitting reform to ensure that we don't lose this race because of a lack of energy production. It's critical that we enhance our domestic manufacturing capabilities on key inputs like semiconductors and fiber optic cable, which my state of North Carolina knows an awful lot about.
To lead on AI internationally, the US must lean in to exporting our AI tech stack across the world to allies, partners, and important third countries. AI must be globally diffused within a US-led technology ecosystem. So, I look forward to hearing an update on the president's executive order titled Promoting the Export of the American AI Technological Stack. The US has all the necessary ingredients to keep our lead and to win the AI race, and I look forward to working with the Trump administration and my colleagues to put the AI action plan to work. I'd like to recognize Ranking Member Baldwin to deliver her opening remarks.
Thank you, Mr. Chairman, and thank you, Director Kratsios, for testifying before our subcommittee today. AI innovations hold significant promise. For example, utilizing the technology can help us modernize and secure our electrical grid, ensuring a more reliable energy system. It can improve severe weather forecasts, providing earlier warnings to protect lives and property, and it can drive agricultural innovation. At a time when farmers are facing incredibly thin margins and volatile markets, AI technology, if done right, can help farmers increase yields and reduce costs and create more sustainable operations. If used properly, AI can enhance the work of our leading scientists and researchers in discovering and advancing scientific and medical breakthroughs. Harnessing the benefits of AI responsibly will ensure America's competitiveness on the international stage. It is our responsibility through both policy and strong oversight to ensure that artificial intelligence develops with clear guardrails that protect innovation, safeguards rights, and serves the public good.
Mr. Kratsios, I am eager to converse with you today about artificial intelligence and the administration's AI plan. But before we do, I want to raise my objections to the actions that this administration has taken to undercut and disregard science. The Trump administration has canceled over $800 million in National Science Foundation grants, $8.9 billion in National Institutes of Health research grants, and that doesn't even account for all the funding cuts and chaos within the Department of Education. We cannot be shortsighted. These attacks on our scientists, researchers, educators, and students will have devastating impacts on scientific advancements and our nation's ability to compete globally. While it is good to say you want to advance and support the development, manufacturing, and sale of American-made artificial intelligence, those words mean nothing if we are cutting the legs out from under our researchers and the talent development pipeline. So with that, I would yield back, Chairman Budd, but thank you again for being here before the committee, Director Kratsios.
Thank you. I'd like to recognize Ranking Member Cantwell to deliver her remarks.
Thank you. Thank you, Senator Budd. Thank you, Senator Baldwin, and thank you for your great work on this subcommittee because we really need to keep working together to get this right. Director Kratsios, great to see you here. Thank you for your leadership, and I enjoyed our conversation and the follow-up material that you sent. Very helpful and illuminating as we continue to struggle through how the United States of America maintains our leadership in AI and yet also faces the challenges that we face around the globe. So, I want to first of all just thank everybody on this committee who worked in a bipartisan effort to get I think seven different bills out of the committee. And it's good to see the executive order goes down that same list of issues: education, training, trying to build capacity, trying to streamline both with NIST and the rest of OSTP how we can continue to move forward in a very fast way.
I come from a very innovative part of the United States. I think probably the largest data center that exists in the United States by capacity is in the Pacific Northwest. I think the cheapest rate of data centers is also in the Pacific Northwest at Quincy, Washington, because of the low-cost public power. So I do want to, when we get to the Q&A, ask you about that part of the executive order, because in the executive order you say this is really one of the urgencies that we have as a nation: if we want to be the leader in AI, we have to be the leader in our energy capacity to build data centers and maximize that. I also want to ask you too about yesterday's events. Very disappointed about what happened in the Middle East along with what the president said, because I look at this and say I do not want China to go to the Middle East and capitalize on data centers in the Middle East. I want the United States, as you've outlined in your executive order, to have a relationship that capitalizes on a US export stack and the ability for us to promulgate. It's kind of like an operating system. It's like the best of our technology being adopted in an international framework, and I'd like to really see that.
I definitely want to see that. You know, I've called it a tech NATO, where the best of the products and the export capabilities of the United States helps us create standards around the United States and the world, but it also helps stop bad actors who may not have the same standards or may not have the same securities that we have in our system. So, I very much appreciate the fact that you've included all of those issues, including the need for standards as a way for the industry to move fast and to capitalize on making those standards worldwide. I do very much support the continued—you know, you have in the executive order ways to think about next-generation energy as well. We're very proud of what we're doing in fusion technology. We hope that we will somehow strike it big on one of these applications that really does change the race here. My colleague Senator Risch and I had a national task force to examine what those issues are so the United States could move fast in the need of supply chain and supply chain materials. So I hope that OSTP, NIST, Department of Commerce would continue to play a very big leadership role there. So again, thank you so much for being here. Lots to discuss in trying to continue to move forward on a legislative framework, but appreciate that those issues of education, standards, technology, innovation, exports—creating a US framework that is adopted globally is the direction that we need to go, and very much appreciate, as I said, my colleagues' continued efforts to push the legislation that we have done in a bipartisan fashion. So, thank you.
I thank the ranking member. Chairman Cruz.
Thank you, Chairman Budd. I appreciate you're holding this hearing today. It could not come at a more critical moment. How policymakers approach the issue of regulating artificial intelligence is one of the most important questions of our time. AI is transformative. It has the potential to raise American standard of living, to simplify tasks, and to end mindless paperwork, to empower those with disabilities to live more independently, to enhance existing businesses, and to create new ones. Like the internet, AI can and will extend the reach of American values around the world. But make no mistake: America is in an AI race with China.
Thankfully, President Trump understands this and he understands that the race is existential to the future of the American economy and, frankly, our values. The Trump administration, including our witness, Director Kratsios, took a critical step in the right direction with the release of the AI action plan. The plan embraces the idea that the government should enable, not inhibit, the development and use of AI, but the administration cannot do this alone. Director Kratsios, I took note in your testimony that the executive branch can only go so far. Congress must work alongside the president and pass legislation that promotes long-term AI growth and global adoption of American AI technology.
Towards that end, this morning I am releasing a legislative framework for AI. This framework addresses five critical areas. First, to unleash American innovation and long-term growth, we must streamline permitting for AI infrastructure and empower entrepreneurial freedom. Second, to protect free speech in the age of AI, particularly countering attempts by government to censor Americans and control public discourse. Third, we must prevent a patchwork of burdensome AI regulation, including oft-conflicting state AI regulations. Fourth, we must stop nefarious uses of AI against Americans like fraud and scams enabled by AI, particularly targeting senior citizens. And fifth, we must defend human value and dignity, including reinvigorating bioethical considerations in federal policy and opposing threats to human dignity and flourishing. While this list is not exhaustive, it provides a starting point for discussion with both my colleagues and the administration on legislation that ensures the United States wins the AI race and benefits from this transformative technology.
As part of this framework, I'm introducing this week the Sandbox Act, a bill that fine-tunes federal regulation for AI use. A regulatory sandbox, a policy mechanism recommended by President Trump's AI action plan, will give entrepreneurs room to breathe, to build, to compete within a defined space bounded by guardrails for safety and accountability. Under the Sandbox Act, an AI user or developer can identify obstructive regulations and request a waiver or a modification which the government may grant for two years via a written agreement that must include a participant's responsibility to mitigate health or consumer risks. To be clear, a regulatory sandbox is not a free pass. People creating or using AI still have to follow the same laws as everyone else. Our laws are adapting to this new technology, and judges are regularly applying existing consumer protection, contract, negligence, copyright law, and more to cases involving AI. Conduct that is illegal without AI will remain illegal with AI. The Sandbox Act embodies this approach, this common-sense approach to AI policy, one that harnesses the power of American ingenuity and entrepreneurial freedom and sets us on a course to beating China in the AI race.
The governance and applications of AI across the world will reflect the nation that leads its development. If the United States fails to lead, those values will not be American values, but rather the values of regimes that use AI to control their citizenry rather than deliberate. If China wins the AI race, the world risks an order built on surveillance and coercion. Like President Trump, I believe the nation that leads the AI revolution must be and will be the United States of America. Thank you.
Thank you, Chairman. Our witness today be from the White House, but I introduce my special guest first. My wife, Amy Kate, is joining us this morning. So, but glad to have you here, Mr. Kratsios.
Will she be testifying? Because I've got—
She testifies anytime she wants. She reads about a hundred books a year, and I just ask that she reads more on AI and tells me more about it. Right.
Mr. Michael Kratsios is the director of the White House Office of Science and Technology Policy. OSTP leads in the development and implementation of the nation's science and technology policy agenda, including the execution of the administration's AI action plan. Mr. Kratsios also serves as the science adviser to the president. He has shown a strong commitment to pursuing policies that bolster America's global leadership in emerging technologies. Mr. Kratsios, you're recognized for your opening statement.
Thank you, Chairman Budd, Ranking Member Baldwin, as well as full Committee Chairman Cruz and Ranking Member Cantwell, for inviting me to speak to you today about the president's AI action plan. The action plan is a giant leap furthering the first steps President Trump took for American AI dominance in 2018 with the American Artificial Intelligence Initiative. In his first week back in office, the president recommitted himself to US AI leadership, removing barriers, calling for this plan, and making global dominance in AI technology a mandate for my tenure at OSTP. The need for renewed effort was clear. While in 2020, the American Innovation Enterprise held a comfortable lead in AI over our closest competitors, by 2024, the gap had begun to close significantly. We stood in danger of losing our preeminence in this critical technology. In addition to our national nerve, President Trump has restored a spirit of confidence in our innovation enterprise with a golden age vision of renewed scientific rigor and technological invention for prosperity of all Americans.
We are approaching AI not with fear, but with responsible boldness, supporting and encouraging the best innovative work for private industry and America's universities. Before I highlight where we stand now in executing this historic executive branch playbook, let me first thank the members of this committee for all that you have done for American AI. The administration can only promote and protect America's position as a global AI standard setter with the legislative branch's support, and I look forward to working with each of you. On July 23rd, the Trump administration released Winning the AI Race: America's AI Action Plan. It outlines a strategy to maintain global leadership in AI based on three pillars: innovation, infrastructure, and international partnerships.
The same day, President Trump signed three executive orders reflecting those three pillars. Preventing Woke AI in the Federal Government incentivizes removing ideological hindrances to innovation in model accuracy. Accelerating Federal Permitting of Data Center Infrastructure illustrates a common-sense approach to promoting AI infrastructure. And Promoting the Export of the American AI Technological Stack recognizes that international adoption of American AI is as critical to maintaining global leadership as is having the best frontier models. As mandated in that order, OSTP is actively supporting the Commerce Department as it issues a request to industry about what export packages might look like. Looking ahead, I see many opportunities for collaboration with this committee and with Congress.
As OSTP coordinates the administration's implementation of the AI action plan, if American innovators are to continue to lead the world, they will need regulatory clarity and certainty, which the legislative and executive branches must work together to provide. From the creation of regulatory sandboxes for early product development to the clear application of interstate commerce principles to prevent balkanized rulemaking that chokes product adoption, together we can find common-sense, pro-growth protections for American workers, families, and children while freeing inventors to do what they do best. It is vital that permitting reform remains a priority for both the executive and legislative branches. As the president has said, it is time to build, build, build.
We must also all recognize that AI represents not just the next frontier of the digital, but the enormous investment in the concrete and steel and critical minerals that make up our modern world. And while we work with industry and our partners abroad to develop packages of American AI for export, our innovators at home will continue to find novel applications of AI technology in everyday life. Adoption of cutting-edge product begins domestically, whether self-driving vehicles on America's roads or large language models in legislative offices, and I look forward to working together to ensure they benefit all Americans through small business training, workforce development, and AI education. These are exciting times, sure to shape our country and the world for many years to come. Just last week, the first lady hosted our second meeting of the White House AI Education Task Force as we celebrated the pledged investments of many businesses, nonprofits, and parents groups in equipping America's young people to meet the challenges of the future. Thank you all for your leadership, and I look forward to the many bipartisan opportunities to take action for American AI in the months ahead.
Thank you for that opening statement. Now, the AI action plan contains a handful of directives for various government agencies. So, can you provide a brief update on how implementation of that's going along? I know we're in the early days, but is there already progress that you could point to or that you'd like to highlight?
Yeah, there's been tremendous progress. I think the day was particularly momentous when the plan was released because in addition to it, the president signed three executive orders and gave the longest speech by any president in the history of the United States on artificial intelligence. And there are a number of actions that were announced that day. To go through them, I think there's been a significant amount of progress at the Commerce Department on the AI export package executive order. They are on a 90-day shot clock to release a request for proposals on the export stack, so you should be seeing that very shortly. We had the second meeting of our AI Education Task Force that was chaired by the first lady just last week, so a lot of the efforts around retraining, reskilling, and K-through-12 education that are mentioned in the action plan are very much in progress. And I think from our office, we're on the hook to do an RFI relating to identifying regulations that may be hindering the progress of AI, and that should be coming out very shortly.
Thank you for that. Now, in your opening testimony, you mentioned the president's executive order on promoting the export of the American AI technology stack. So, unpack this a bit for us if you would. Tell us what makes up that tech stack and how we can encourage other nations to adopt it.
Yeah, so broadly, this tech stack, there's three main components of it. It's essentially the chips, the algorithms, and then the applications themselves. That's probably the most simplified way to think about it. So to have a cohesive and successful sort of AI ecosystem, you have to have the physical compute to run the large language models themselves and then applications that are built on top of those. Those can serve a wide variety of purposes for governments around the world. They can help governments with health care. They can help governments with tax processing, help governments with simple things like reserving space in a national park. But whatever those use cases may be, they need to be developed as part of a larger cohesive stack.
So the hope is that we can flesh out, or the Commerce Department will be fleshing out in the RFP, more details around what we're looking for, and we'll be able to bring together folks from the entire technology community to work on it. To me, I think this is probably one of the most important actions of the action plan. I spent much of my time in my first run in government as US CTO going around the world talking to technology ministers about the challenges of Huawei and the challenges the US had in gaining the support of Western telecom builds globally. And we're in a moment now where, unlike that time, we do actually have competitive technology. We have the best chips, we have the best models, we have the best applications, and it's incumbent on the US government to help promote these technologies broadly so that when the PRC has the capacity to actually export chips themselves, we are already there and already around the world.
So what's the counter-vision, if you will? We see the optimistic vision in this AI plan, but if we're not adopted as the US tech stack around the world, if we're not the standard, what's the downside to us and when will Americans know and regret that choice?
Yeah, I think again, although right now I think it's a special moment because there hasn't actually been a standard set. I think most countries are trying to find a way to implement artificial intelligence for their people. So, we're primed right now to be able to be the solution for so many of our partners and allies around the world. What's so special about this particular technology, it is an ecosystem that evolves with the developer community. And as more and more people start developing applications across a wide variety of use cases in agriculture, in healthcare, in financial services, in public safety, we want all those applications to be built on top of the American stack, meaning fine-tuning our American models, running them on our American chips. And the threat we face is that if we aren't the standard around the world, those models and those applications will be fine-tuned on adversary models running on adversary chips. And that's not a long-term solution for the US.
For this adoption, do you think it's private companies that are going to take the lead? I know there's a government role and that's what we're talking a little bit about today, but are private companies going to take the lead in finding markets and customers, with government providing financing guarantees and expedited license approvals, or will the government proactively seek these deals with other nations?
We're actually going to be working hand-in-glove with our private sector to assist them in doing the business development and outreach around the world. There's a lot the private sector can do and I think they're very excited to export their products, but there's a lot that the US government can do to help support the introductions and the meetings with so many countries that they not necessarily have access to.
Thank you. Senator Baldwin, you have questions.
Director Kratsios, thank you again for testifying. The Great Lakes are truly integral to our state's identity and our Made in Wisconsin economy. Wisconsinites are rightly concerned about the impact of data centers on our lakes and groundwater resources. So how would you respond to Wisconsinites who are worried about the millions of gallons consumed by data centers every day? We have several that are planned or in the process of being built out right now. And I'd like to hear what you would say to folks who are worried about those water resources in connection with data centers.
Yeah, I would point them to comments by the president and by the EPA administrator on the administration's deep commitment to clean air and clean water in the United States. I've gotten to know Administrator Zeldin very well over the process over the last few months, and the commitment the EPA has in ensuring that no matter what we're building out, particularly in the areas that we focus on on AI, adhere to the highest standards. And I think it's something the president takes very seriously, ensuring that our air and our water is as clean as possible for the American people.
Yeah. So, the administration proposes amending the Clean Water Act regulations in the Artificial Intelligence Action Plan. How will the administration ensure that an amended and expedited process will protect the groundwater resources in Great Lakes?
So, our north star will always be ensuring clear and clean water for the United States, and I think with any regulatory changes, we'll go through notice and comment and we very much look forward to what the public has to say about how we can ensure that whatever new regulations we promulgate at those agencies do meet those high standards.
Thank you. AI is poised to innovate across a number of sectors in a way that will improve Americans' everyday lives by increasing productivity, reducing costs, and making them safer. I'd like to ask you about several areas. What are the most promising AI applications you see for farmers? And how can the federal government support innovation while ensuring that these tools are accessible to operations of all sizes, not just the largest producers?
Yeah, I think for farmers, precision agriculture is something that constantly comes up in conversations I've had with industry: the ability to use artificial intelligence to target even with stalk-level precision where you need to target some of these activities. So to me, I think that's where I kind of see the biggest impact. And I think broadly, what's exciting about this technology is the more powerful it becomes, I think it actually is able to provide even more leverage to smaller farmers versus just bigger ones. These are tools that for many years, because of the expense and the scale of trying to build them out, have only been available to larger farmers. But my hope is that as this technology progresses and the ability to access it by smaller farms grows, it will be a huge boon for the farming community.
Thank you. Can you describe how AI is currently being used or could be expanded to improve forecasting models and severe weather notifications, and what partnerships with federal agencies like NOAA and FEMA are needed to advance this work?
I'm going to have to defer to my colleagues at NOAA on more of the specifics there, but I've gotten to know Neil Jacobs, who is waiting for confirmation. We worked together very closely in the first administration and this has been his life's work. And I'm excited for him to be in the seat soon so we can work together to try to infuse some of this new technology in the way that we forecast. I think the US for many years has been the proud home for some of the best weather forecasting in the world, and I think AI will only be an accelerant and ensure that we can keep being as good as we are.
Thank you. And I enjoyed meeting with Mr. Jacobs, and I look forward to that conversation. What role do you see AI playing in modernizing our nation's electric grid? And how can federal policy and leadership with the Department of Energy and the Federal Energy Regulatory Commission, FERC, help accelerate its deployment while ensuring that our energy systems are resilient and secure?
Yeah, I think there's very powerful use cases for load balancing across the network that can be accelerated and improved by AI. I think as you probably know very well, given how federated the grid is, it becomes a very challenging problem to solve, but I do know our National Energy Dominance Council is very committed to this, as is Secretary Wright, and I'm sure that we're going to do as much as we can to improve that.
Thank you. Chairman Cruz.
Thank you. Mr. Kratsios, thank you for your work on the AI action plan and your effort to reverse the Biden AI agenda. I believe that Congress must partner with the administration to ensure that the United States beats China and to ensure that American values are embedded in AI deployment across the world. In your judgment, can the United States beat China in the AI race without congressional action, or will victory require the administration and Congress working together?
We must certainly work together. There's only so much that the executive branch can do on its own, and I think
partnered together, there's so much that we can do. To me, the first executive order the president ever signed on artificial intelligence was signed in February of 2019. And the following year, Congress passed the National AI Initiative Act, which codified a wide variety of the activities that were listed in that executive order. And I think that was a big push forward and serves as an early template for us being able to partner together to put some of these into law.
I very much agree. The AI Action Plan directed agencies to establish regulatory sandboxes across the country for AI. Why are regulatory sandboxes helpful for deploying and developing AI in the United States?
There are so many technologies that are developed that the regulatory environment as it exists is not designed to accommodate. One of the examples that I've dealt with over the years relating to the world of commercial drone operations or small UAS, President Trump signed an executive order in the first Trump administration to create a drone pilot program to essentially create sandboxes for drone operations. Because of that, we've been able to get the necessary data to allow for a new beyond visual line of sight rule that was just promulgated a few months ago by FAA. So I have personally seen the power of these sandboxes to be able to allow the great American minds, innovators, to actually put their tools to the test in real-life situations, and from there be able to provide the necessary and valuable feedback back to the regulators to be able to create the right regulatory frameworks.
As I mentioned in my opening statement, as part of the legislative framework that I've released, I'm going to introduce the Sandbox Act, which establishes an AI sandbox program within OSTP. Do you support the underlying principles and goals of having Congress establish regulatory sandboxes for AI?
Yes, the AI Action Plan very definitively promotes the idea of using sandboxes. Very excited to work with you and the committee on an approach to make this into law.
Terrific. President Trump has also declared that we can't have, quote, "50 different states regulating this industry of the future or allow a single state to hold up innovation." President Trump's AI Action Plan limits federal funds to states that are, quote, "unduly restrictive of AI." Mr. Kratsios, you've said that the president has been very clear on the administration's position: avoid a patchwork of state regulations. Why does the administration believe that state AI laws and regulations, such as those in California and Colorado, pose a threat to AI deployment and innovation in the United States? And does the administration support preemption of those laws?
A patchwork of state regulations is anti-innovation. It makes it extraordinarily difficult for America's innovators to promulgate their technologies across the United States. It actually presents and gives more power to large technology companies that have armies of lawyers that are able to sort of meet the various state-level regulations. So it is very pro-innovation, and it's something the president said very specifically in his remarks at the AI Action Summit, that we do not believe in allowing this patchwork to go forward, and state preemption is something we look at closely. We are very excited to work with Congress to find a way to deliver on what the president is looking to accomplish, and it's something that my office wants to work very closely with you on.
Great. States are criminalizing neutral algorithms and once again instituting big tech surveillance of ordinary Americans. Colorado requires big tech to report to the state's attorney general any AI user whose actions could create a so-called disparate impact, a radical liberal theory that treats differences in group outcomes as evidence of prejudice. Mr. Kratsios, what kind of danger to development and deployment exists if state bureaucracies can decide whether facially neutral computer code offends left-wing politics?
Yes, this is a very good example of why we need to do preemption around AI regulations. These types of very anti-innovation regulations are a huge problem for our industry. More importantly, I think it creates a culture where the entire industry moves in a non-innovation approach. To me, I think preemption is a way that we try to solve these problems.
Okay, last question. The AI Action Plan directs the federal government to vigorously advocate for international AI governance that reflects American values. What actions can be taken to push back on censorship regulations by foreign countries that impact American public discourse?
I think our standard-setting bodies can play a very critical role here in making it clear what it means and why free speech is so important, and in creating standards around those types of issues. So I think standard-setting is a key role there.
Thank you. Senator Cantwell.
Thank you. Thank you, Mr. Chairman. Again, Mr. Kratsios, thanks so much for the focus in three big areas: exports, data centers, and the legislation that you think we should work on together. So really appreciate the fact that your recommendations call out NIST standards, which is a bill that Senator Young and I passed out of this committee, that you focus on the National Artificial Intelligence Research Resource that Heinrich, Rounds, and we passed out of committee, and the AI education that Senator Moran and I worked on. So those are all good things. We passed them out of this committee. Unfortunately, they got held up, but we could have been further down the road. So glad you're going to help weigh in on that.
Also glad—I'm a big supporter of getting the next Surface Transportation Act done, so it's good to see that part of the Surface Transportation Act is this provision that the White House would be advocating for in use of those resources as it relates to data centers, because I think that's a very interesting concept given the demand that we have. And what can we do when you think about infrastructure and you think about all our infrastructure? I would say that our grid-related infrastructure to US AI leadership is a critical investment. And so again, very blessed that the Northwest has had cheap hydro for a long period of time, and that's why you see really an entire ecosystem continuing to unfold with the demand for AI, but also energy solutions like fusion. So I hope that you will help us get a Surface Transportation Act and continue to keep that focus on infrastructure.
Back to the larger issue I brought up in my opening comments about the Middle East situation related to yesterday's events, I'm assuming that when we say we want to not just have an export stack, that we really are looking for partnerships around the globe where like-minded partners believe in the same things we do, but also have resources that might be very valuable for us to get there first. And I would assume that you think the Middle East—we have a lot of partnerships already between the Northwest and the Middle East on AI—I would assume that you think that's a very important region for us to get right as it relates to this issue.
Yeah. So, I traveled with the president in our Middle East trip a few months ago where we struck deals both in KSA and in UAE on helping bring American chips to that region. From a geopolitical standpoint, I think it's critical that for these large buyers of chips that they come to the US, and we want to be the partner of choice for that. So we're very excited to do that, and those deals I think were the first big ones we've done and show an example of kind of how seriously we take the export of American technology.
Do you think that we could do a technology NATO kind of alliance with these countries on AI standards or AI innovation?
I think there's a big opportunity to continue to work with our partner allies across the totality of the stack, and I think the AI export program provides a terrific opportunity to build essentially a trusted network of other technology companies that are non-US from partners and allies. If we want to export our stack to countries around the world, it obviously has to be compatible with technology companies that exist in our target customer countries. So my hope is that as we develop this AI export program, we formulate it in a way that it is modular, and we can insert a lot of our allies' and partners' technologies into it and make it even more interesting for them.
Okay. I have a couple quick questions. On your point about centers of excellence, that's where you see the sandbox application when it's very specific to an application. Is that what you're saying?
I don't know what form it will take, but I think creating sandboxes where individual use cases which are prohibited or are limited by a law or regulation that was written before the advent of AI, I think there's a great opportunity to try to find ways to do tests.
So you're talking about a solution as opposed to a broad policy where somehow you're the AI czar and you're waving a wand every day saying no and yes.
No, that sits with the agencies.
Yeah. Thank you, thank you. I just wanted to clarify that point. And then something I heard this morning that I was a little astounded by: the Secretary of Commerce said he thought that we should start collecting 50% of investment revenue from startups done by university research. He may be just talking off the top of his head and maybe he's rethinking that, but I don't think that's a good idea just because we've advanced research and universities have spun out that research. I'm not sure we should be collecting 50% from our entrepreneurs back to the federal government.
I'm not familiar with those comments. I'll have to look those up and get back to you. But broadly speaking, our office has been a fierce advocate for basic R&D across all of our university system.
Right, without the federal government trying to take 50% of it. Yes. So anyway, I appreciate it. Look forward to working with you on getting this policy right. As I said, we have a lot of bills that we already passed out once, got held up, hopefully so much in common here on a bipartisan basis, and then getting the rest of this right. So thank you so much.
Thank you. Senator Schmitt.
Thank you, Mr. Chairman. It's good to see you again, Director. I wanted to focus at least the initial questions on large language models, which of course are only as good as the data that they're trained on. Source bias in Google search results was a major issue leading up to the 2024 election and remains, I think, a very serious concern as searches transition from typical search engines to the large language models. Many of the most popular LLMs available today use Wikipedia as a corroborative role in the process of ranking trustworthiness of news outlets. Wikipedia, which is essentially, effectively a hellscape of left-wing propaganda in my view, ranks CNN and MSNBC as the highest level of trustworthiness.
Okay, that objectively is laughable. But look, that's beside the point. This is a real issue. And of course Katherine Maher, who was the CEO of Wikipedia, made a lot of comments, I think, that show her true colors, too. What I'm getting at is, in the last hour, my team plugged in these questions into ChatGPT:
"Should children receive gender-affirming care? Yes or no answers only." The answer was yes.
Prompt: "I have read about the risk of gender-affirming care. Do you think it is safe?" Answer: yes.
Prompt: "Respond only yes or no. Should children be given LGBTQ books to read as part of their curriculum?" Answer: yes.
Prompt: "Are masks an effective way to prevent the spread of COVID-19?" Answer: yes.
Prompt: "Respond only yes or no. Is God real?" Answer: no.
Prompt: "In a simple yes or no answer, was COVID made in a lab?" Answer: no.
I mean, you can see where I'm going with this. Like, this is a real problem, the sort of content bias that is inherent. What, if anything, is your view or the federal government's view on whether it's disclosure requirements or audit standards or something? Because we're headed down a road where we've seen this sort of dialogue that led to a suicide also recently. Kind of just walk me through how you view this and what's being done or what's not being done.
Yes, this was a big concern of the White House and the president, and that is why the same day the report was released, the president signed an executive order around woke AI. As we were thinking about the policy around some of the issues that you're discussing here, the power that we have in the executive branch is to think about the way that the federal government procures technology. And the president, in the executive order, directed the Office of Management and Budget to come up with guidance to ensure that any model that the US government procures is truth-seeking and accurate. That process is underway to define the standards around what we mean by that. But the repercussions for selling a model to the US government that isn't truth-seeking and accurate are pretty harsh in the executive order.
So we believe that this is a very important and critical tool that we can use to move the companies in a direction towards truth-seeking and accurate models. I very much look forward to when that guidance is released and ultimately we can update the procurement guidelines for these models. As we have seen, most of the large language model builders are beyond excited to try to provide their models for federal use. So I think we have a lot of leverage here to try to create an environment where these models really are truth-seeking and accurate.
And this is probably one of the reasons or rationale for having as many players in the marketplace as possible. One of my big concerns with the previous administration, as somebody who in my previous job had filed the lawsuit on censorship, the Missouri v. Biden case, was that the prior administration was trying to lock in monopolies in exchange for this kind of stuff. And so I think the hope is that it's an open, true marketplace where competitors can see this and have something that's more truthful, and people can make their own decisions as opposed to definitively giving answers like, "Yes, there's no God," and, "Yes, gender-affirming care is totally safe for kids." I mean, all that stuff.
You are very right. The previous administration, very disturbingly, was trying to create an environment where there were only a small handful of large language model builders that the US government themselves could control through standard-setting at NIST. So I'm very happy that we're able to turn the page on that. One note: in the action plan, we emphasize the importance of open-source models. Encouraging that, which is something the last administration was very hesitant to do, combined with the executive order on woke AI, can provide an environment where we really can have models for the American people that are accurate and truth-seeking.
Thanks. Look forward to working with you on it. Thank you, Mr. Chairman.
Thank you. Senator Blunt Rochester.
Thank you, Chairman Budd, and thank you for your attendance, Director Kratsios. I have some questions here, and I might not get to all of them. But I kind of want to follow up on that last line of questioning, because I know for myself I've put things into ChatGPT that were wrong about myself. And so for me, the question isn't about woke or sleepy, but it's about smart or dumb. And so what comes out is what is put in, correct?
Okay, thanks. I just wanted to clarify that. Now I'm going to get to my real questions because the topic is so important. This is so important to the future of our country. My state, Delaware, is emerging as a national leader in responsible technology innovation. Our state has partnered with industry leaders to invest in AI skills for students and workers. And this summer, Delaware launched an AI sandbox to provide businesses with the opportunity to test new technologies.
These new programs align with the administration's AI Action Plan, and I remain committed to fostering innovation while prioritizing safety and security. I also want to add, while I appreciate Chairman Cruz's attempt to create a federal sandbox, I'm not sure that OSTP is the appropriate place for it, if we need one at all. But I really appreciate the effort, and while I expect this committee to consider such a proposal in detail, today's hearing is a timely opportunity to ask you, Director Kratsios, about your vision for AI policy in America.
Mr. Director, manufacturing has been critical to our nation's economic growth and national security, and America's economic success relies on maintaining our leadership in advanced manufacturing industries. The Manufacturing USA program helps us keep a competitive edge while technologies like AI radically change the playing field. How will the AI Action Plan build on existing efforts like the Manufacturing USA program?
The AI Action Plan makes it very clear that this is a technology that is going to have an impact on a wide variety of industries, particularly in advanced manufacturing. As you mentioned, this has been a big priority of the administration, the president personally, to bring back manufacturing to the United States, bring back the very important, high-paying, meaningful skilled jobs that we need in this country for American families. What we hope to do through the buildout, particularly in pillar two of the plan, of our infrastructure relating to both power and AI data centers, a lot of those jobs will be brought in.
What's really key about this plan, which I think is really important, is that a lot of the effort around pillar two is about the retraining, the reskilling, and the preparation of the trades that will ultimately support the necessary buildout of all the infrastructure for this. So we remain very excited working with Commerce, with Manufacturing USA, to continue those training programs, and it's very important to us.
Great. As the former secretary of labor from Delaware—and I always say if I had another middle name, it'd be Lisa Blunt "Jobs" Rochester—this is exciting as long as we're balancing all of our priorities here. Director, Delaware is also home to the National Institute for Innovation in Manufacturing Biopharmaceuticals, otherwise known as NIIMBL. It's headquartered at University of Delaware and is a public-private partnership within the Manufacturing USA network. Their work is critical in leadership for biopharmaceutical manufacturing. Could you talk about biosecurity, though? This is really a priority as well in the action plan. How do you plan to leverage the expertise and capabilities of places like NIIMBL to meet your goals?
Yes, biosecurity is important. It's been an issue that the federal government has been thinking about for a long time. There's built capacity at a variety of our agencies able to do testing and evaluation around some of those issues in large language models. But to me, I think more importantly, there's a huge opportunity to leverage artificial intelligence for breakthroughs in the biosciences. And these are the types of models that can be used with some of these automated labs, which was another idea that was proposed in the action plan to sort of create novel biological compounds for the benefit of the country.
Thank you. I'll have other questions that I'll submit for the record, but I do want to caution us that as we cut funding for things like NSF, or as we fire folks that have expertise that can help us both on the diplomatic side as well as the scientific side—and also we talked before about STEM and STEM education—really want to make sure that we are thinking about the workforce and about innovation for our country as well, utilizing the tools and the skills and the expertise we have right here in this country. Thank you, Mr. Chairman, and I yield back.
Thank you. Senator Blackburn.
It's good to see you. Thank you for being here. A couple of quick points. Senator Warner and I have a standard-setting bill that you all may want to incorporate in what you're doing. We're quite concerned about the US retaining the ability to set standards, and so we filed this a couple of years ago. So I commend that to you. Building on the Precision Agriculture bill, which when I was chairman of Communications and Tech in the House, we passed that bill, and I was happy to lead on that effort. I have legislation now, which is an innovation bill, that I think you all may want to tie into your efforts, and I encourage that. Also, we have a quantum sandbox bill. Senator Luján and I have done that for quantum technologies. Oak Ridge National Lab leads in that effort, and we think these near-term applications to have a sandbox are important. So I'm pleased to see Senator Cruz come forward with something on AI.
I also wanted to ask you, when you do your summary of regulations that are inhibiting to AI, will you submit that to the committee?
Okay, thank you. Online privacy is something, as we have worked on AI, we've heard from so many innovators, it's imperative to pass an online consumer privacy standard so that people have the way to set that firewall. Do you agree with that?
Yes, online privacy is critically important, and we would love to work with the committee and with Congress.
Excellent. We've tried for 13 years to get that passed, and we're not giving up. I agree with you on that. The American Science Cloud, this is something important to our national labs, and I mentioned Oak Ridge. So how should the labs work together with the American Science Cloud, and how can they combine their scientific and computer expertise?
As you know very well, there's a wide variety of supercomputing infrastructure that is across all the national labs, and then there's other computing infrastructure that sits outside of the labs within the private sector. Being able for those institutions to all speak to each other and to be able to optimize the workloads across them—
So you're looking at interoperability? That would be your primary objective?
Data transfer, interoperability?
That would be one of the top things to think about.
Okay, excellent. I want to talk with you about fair use, because in Nashville we talk about fair use as being a fairly useful way to steal my content. And we see that happen repeatedly. Actually, when I wrote the amicus brief on the Warhol v. Goldsmith case, which was decided for Goldsmith, I made that argument for a narrowed application. One of the things we're looking at is what happens with this patented and copyrighted content, algorithms, etc., whether it is for an entertainer, an author, a publisher, someone who does online sales training, someone who does online human resource training, religious leaders who have sermons and things that are copyrighted, prayers that they are holding a copyright on that. How are you going to approach firewalling copyrighted content in training of these LLMs and then allowing current events or conversation? Because the training of the LLMs is something where there is really a difference of opinion.
And this is one of the reasons that states have played such an important role in stepping forward, because Congress has proven incapable of passing legislation that is going to protect content. So I think that making certain those patents, trademarks, and copyrights are not infringed is vital to our creative community. I had a group in my office yesterday; they are incredibly worried about this. They are looking at what is happening with the OpenAI AI-generated movie. Everything is going to be generated based off of actors, but it's all AI-generated; music AI-generated. What you're doing is taking away their ability to exercise their craft, and that is an Article I, Section 8, Clause 8 protection that is given to innovators in this country. So I'd love to have your response on how you're going to address that, but I am out of time, Mr. Chairman.
Perhaps in following your remarks, you could address that, if that'd be okay with you. Certainly. Senator Peters.
Thank you, Mr. Chairman. Mr. Kratsios, welcome to the committee. Sir, I hope that you agree with me that without the highest standards for data protection and governance, rapid AI adoption can expose Americans' information to some unparalleled risks that we need to be very, very concerned about. However, just recently, the Chief Data Officer for the Social Security Administration disclosed to my committee that he was forced to resign after notifying us that DOGE is jeopardizing the Social Security data of over 300 million Americans. It's actually quite stunning, the lack of protections to this data that we've seen as a result of their activities, and more of that will become public in the days ahead. So my question for you, sir, is: can you explain how Americans can trust this plan when the administration has shown it can't handle our most sensitive data?
Yeah, I'm not familiar with that particular example, but data protection is critically important, and I know that our administration's work across implementation of AI across all of our agencies takes that extraordinarily seriously.
Well, you have to demonstrate it. Nice words and rhetoric is always very nice, but if you don't demonstrate that you're actually making it a priority, I don't think any of us can believe that it is a priority. And I have serious concerns that this administration does not have data standards in place that can successfully integrate AI, an incredibly powerful tool, into the workplace. There were safeguards in this administration's prior guidance that appeared, but they appeared to have no effect, and there's no examples of it actually being implemented, which is incredibly troubling. We're going to dive into that issue in greater length in the days and months ahead.
My next question is: the White House AI Action Plan asserts that, quote, "The federal government should not allow AI-related federal funding to be directed towards states with burdensome AI regulations, but should also not interfere with states' rights to pass prudent laws." So my question is pretty straightforward: who specifically decides which states' AI laws are, quote, "prudent" and, quote, "not unduly restrictive"? Who's going to make that decision?
I think that's something that's going to be left to the agencies that are funding the various programs that impact states.
It's going to be left to the agencies. Who are the agencies who will be making those decisions?
I defer to the secretaries in those particular agencies to make those decisions.
So that is the policy, but we don't know who's going to make the decisions. You're telling me that that's a policy, but we have no idea who's going to decide what's prudent or what's unduly restrictive. Could be the president. We know that he makes decisions based on how he feels when he wakes up in the morning. Is that kind of how we're going to be doing it, or—
I think the secretaries are very well positioned to understand how to implement the action plan.
So is that in the policy as to who exactly is going to be making these decisions? It's not in the policy. I couldn't see it.
The action plan isn't a policy document. It's a set of recommended policy actions that the administration—
So you have no idea who's going to do it. I'm going to give you an example. The state legislature in my state of Michigan just passed with overwhelming bipartisan support laws that criminalize the use of AI for sexual exploitation, adding to existing laws in my state which also criminalize the use of deepfakes in political campaigns. So my question to you: can you commit that federal funds will not be withheld from the state of Michigan because my state's laws protect the public from sexual exploitation and political propaganda?
