DeepSeek and Unitree Robotics

The committee on Homeland Security subcommittee on cybersecurity and infrastructure will come to order. With that objection, the chair may declare the committee at recess at any point. The purpose of this hearing is to examine the national and economic security risks to the United States by artificial intelligence, robotics, and autonomous sensing technologies developed by companies affiliated with the People's Republic of China. This hearing will also evaluate whether existing federal guidance and supply chain security practices sufficiently address the risks posed by China as these classes of emerging technologies are rapidly adopted and our adversaries try to keep pace with American innovation. I now recognize myself for an opening statement. Good afternoon and thank you all for being here. Today, this subcommittee is examining a national security challenge that is currently unfolding inside the United States. The People's Republic of China is moving aggressively to dominate the technologies that are reshaping the global economy and security, including artificial intelligence, robotics, and autonomous systems. This competition is already present on American campuses, in American police departments, on American smartphones, and increasingly within systems connected to our critical infrastructure. The pattern behind this expansion should look familiar. Chinese companies replicate or acquire American innovation, benefit from large-scale state support, undercut competitors on price, and then use their growing market presence to collect data and expand Beijing's strategic reach. Washington saw this pattern before in industries such as solar panels, drones, and electric vehicles. We are now seeing it again in artificial intelligence and robotics. I want to start with DeepSeek because what happened during the past year represents something genuinely alarming that we have not fully reckoned with. In January 2025, a Chinese artificial intelligence company called DeepSeek released a model that rivaled the best systems produced by leading American AI innovators at a claim cost that was a fraction of what American companies invested. The technology world and the market were shocked. Observers called it an AI Sputnik moment. But the real story was not just what DeepSeek built, it was how. Within weeks of its release, American AI companies raised serious concerns that DeepSeek used their proprietary models without consent to train its own through a process known as model distillation. Rather than investing years of effort and billions of dollars in frontier research, DeepSeek appeared to have fed outputs from leading American AI systems into its training pipeline, illicitly reverse engineering the capabilities those companies had developed through legitimate investment. More recently, in February 2026, DeepSeek, along with Chinese AI firms MiniMax and Moonshot AI, had reportedly orchestrated an industrial-scale campaign to extract capabilities through distillation attacks. Further reporting suggested that DeepSeek employees used third-party routers and networks of unauthorized resellers to circumvent existing safeguards and may have sought to override safety protections around biological and chemical weapons research. Let me be clear about what that means. DeepSeek did not simply outcompete American companies, according to these disclosures, it stole from them. And DeepSeek is not just a competitive concern, it is also a data and influence concern sitting on millions of American devices. The application is available today in the Apple App Store. Every interaction is stored on servers located in the People's Republic of China where authorities can demand access to that data under Chinese law. Researchers have also documented that DeepSeek censors questions related to issues sensitive to Beijing and frequently produces responses that align with official Chinese Communist Party positions. The same concerns apply to Unitree Robotics. Unitree has become a dominant global supplier of robotic dogs and humanoid robots, supported by Chinese manufacturing scale and state subsidies that allow its products to be sold at a fraction of the price of American systems. As a result, these robots are already appearing in police departments, universities, and even parts of the federal government here in the United States. Researchers have discovered vulnerabilities in Unitree systems that could allow unauthorized access to live camera feeds or even remote control of the devices. Additional research has identified weaknesses that could allow one compromised robot to spread attacks to others nearby. Some of these systems have also been observed transmitting operational data back to servers located in China. American companies are innovating and competing seriously in robotics and artificial intelligence, and this subcommittee strongly supports their work. The answer is not to retreat from these technologies. It is to invest in trusted American alternatives, strengthen cybersecurity practices, and prevent federal funds from supporting platforms that put American data and infrastructure at risk and give China an economic and military advantage. The witnesses before us today have direct knowledge of these threats and the larger dynamics at play across these industries. I look forward to their testimony. Thank you. I now recognize the ranking member, the gentleman from Virginia, Mr. Walkinshaw, for his opening statement.

Thank you, Mr. Chairman, and good afternoon everyone. I appreciate the opportunity to discuss the national security risks posed by technological advances in the People's Republic of China. We're living through a period of rapid technological transformation. AI, advanced computing, and digital technologies are reshaping our economy and our national security landscape. The implications are profound and the United States must be prepared. Maintaining leadership in technological innovation is not simply an economic priority, it's a national security imperative. To succeed, we must pursue policies that both drive innovation at home and mitigate the risks posed by adversaries. That requires a clear-eyed understanding of the PRC's strategy. We've known for many years that the PRC views its industrial policy as a critical tool to reshape the world order and emerge as the economic leader and global superpower. With extensive state investment in technology companies and laws that enlist private companies to serve the interests of the government, the PRC's military-civil fusion is a serious threat to our own national security. Historically, Congress has recognized this threat. There's been bipartisan consensus around reducing supply chain vulnerabilities, restricting the use of certain PRC technologies, limiting exports of sensitive U.S. technologies, and strengthening secure supply chains with trusted partners. Unfortunately, policies of the Trump administration have undermined that progress. Upon taking office, President Trump took his DOGE chainsaw to the federal workforce, weakening agencies essential to our national security and scientific leadership. The Cybersecurity and Infrastructure Security Agency, CISA, has lost nearly a third of its workforce because the Trump administration fired people, reassigned people, or bullied people into quitting. At the very moment our adversaries are expanding their cyber arsenals, the administration is cutting our own defenses. From the PRC to Russia to Iran, our adversaries leverage cyberspace and related technologies to fight international conflicts on our homeland. We've seen Iran do this already. Just last week, a pro-Iran, Iran-connected hacker group caused global disruptions at a major medical device manufacturer, Stryker, in response to the president's war with Iran. The PRC is a far more capable cyber actor than Iran and the potential consequences of a future conflict with the PRC are far more severe. Yet instead of strengthening our technological and cybersecurity defenses, the administration has weakened the agencies responsible for securing our networks and advancing our scientific leadership. Key institutions like the National Science Foundation and the National Institute of Standards and Technology, NIST, have also seen major workforce cuts. At a time when technological competition is intensifying, dismantling our technical expertise is incomprehensible to me. The administration also launched a chaotic trade war that fails to distinguish between allies and adversaries. Rather than building trusted supply chains with our partners, these erratic policies alienate allies and inject uncertainty into the global economy. Even more troubling, in the fall last year, the Trump administration announced that it would allow for the export of advanced semiconductors, NVIDIA H200 chips, to the PRC. With access to more advanced chips, the risk of more rapid frontier AI development in China increases. Imagine you were in a race and only you had access to high-octane fuel that let you drive 10 miles per hour faster than your opponent. Would you give that high-octane fuel to your opponent? Your opponent who's determined to leave you in the dust and crush you. That's what President Trump did. It was weak. Finally, we should recognize a fundamental truth about American innovation. The U.S. leads the world in technology because we attract the best talent from around the globe. Immigrants have played an outsized role in building our technology sector. Policies and statements that target immigrants do not strengthen America. They weaken our ability to compete with our adversaries. We could spend all day naming the immigrants who have played a critical role in developing technologies that advance American economic and national security. I'll name just a few. Sal Khan, the son of a Bengali Muslim immigrant, created Khan Academy, a household name, and in 2023 launched Khanmigo, an AI-powered personal tutor and teaching assistant now offered free to every teacher and every classroom in the U.S. Rana el Kaliouby, an Egyptian-American who founded Affectiva, which built AI that understands human emotions and cognitive states by analyzing facial and voice expressions. She co-hosts a PBS Nova series on AI. And Dina Katabi, an immigrant from Syria, who's one of MIT's most decorated scientists. The winner of every science award there is to win. She developed an AI-powered device that can detect Parkinson's based on how a patient breathes at night. Mr. Chairman, these are some of the people that Chairman Ogles and other Republicans in this body recently said don't belong in America. Sal Khan's father came to this country from Bangladesh. Rana el Kaliouby from Egypt, Dina Katabi from Syria, all happen to be Muslim. And together, along with many others, they're helping America win the AI race, win the robotics race, win the technology race. Our adversaries, China, Russia, Iran, would love nothing more than for us to drive out the scientists, engineers, innovators, and entrepreneurs who help make this country the most advanced and yes, the greatest nation on earth. So I beg my colleagues, I beg you, Chairman Ogles, stop doing their work for them. With that, I look forward to hearing from our witnesses and I yield back.

Thank you, ranking member Walkinshaw. Other members of the committee are reminded that opening statements may be submitted for the record. I am pleased to have a distinguished panel of witnesses before us today on this important topic. Pursuant to committee rule 8C, I ask our witnesses please rise and raise their right hands. Do you solemnly swear that the testimony that you will give before the committee on Homeland Security of the United States House of Representatives will be the truth, the whole truth, and nothing but the truth, so help you God? Let the record reflect that the witnesses have answered in the affirmative. Thank you. Please be seated. I would now like to formally introduce our witnesses. First, Mr. Max Fenkell is the global head of policy and government relations at Scale AI, where he leads the company's global engagement on AI governance, data policy, and emerging technology strategy to shape frameworks that ensure U.S. technological leadership and national security while promoting innovation. Prior to joining Scale, he most recently held jobs and roles at Joby Aviation and the Aerospace Industries Association. Next, Mr. Matthew Malchano is the vice president of software at Boston Dynamics, where he has worked for 22 years, contributing to projects on advanced robotics technologies, AI, and autonomy. He also currently holds the role of Boston Dynamics' security officer. Mr. Malchano holds degrees from MIT in computer science and electrical engineering and prior to Boston Dynamics, worked in software development and engineering roles at Akamai Technologies and Google. Mr. Michael Robbins is the president and CEO at the Association for Uncrewed Vehicle Systems International, an association representing uncrewed systems, robotics, and autonomous companies in both the commercial and defense sectors. Mr. Robbins also currently serves as the Mitre Corporation's aviation advisory committee and the National Advanced Mobility Consortium's advisory committee and is an officer in the United States Navy Reserve. Thank you for your service. Dr. Rush Doshi is the C.V. Starr senior fellow for Asia studies and director of the China Strategy Initiative at the Council on Foreign Relations and is an assistant professor at Georgetown's School of Foreign Service. His expertise includes China's foreign policy, U.S. strategy toward China, cross-strait issues, and Indo-Pacific security. Prior to joining CFR, Dr. Doshi most recently served on the National Security Council's China Directorate. I thank each of our distinguished witnesses for being here today. I now recognize Mr. Fenkell for five minutes to summarize his opening statement.

Chairman Fong, Ranking Member Walkinshaw, members of the subcommittee, thank you for the opportunity to appear before you today. My name is Max Fenkell, and I lead global policy and government relations at Scale AI. Since 2016, we have worked with nearly every major AI lab, powered Fortune 500 companies and the most ambitious government AI programs, and are now at the frontier of physical AI and robotics. We do this across two main lines of effort: providing high-quality data to train, improve, and evaluate AI models, and building applications to help our customers bring AI into production. Let me start with a reflection of where things stand today. Last month, Chinese viewers rang in the Lunar New Year watching AI-powered humanoid robots perform coordinated martial arts on live television. The video went viral not because it was impressive, but because of what happened when people compared it to last year. Twelve months ago, those same robots could barely shuffle through a dance routine. This year, they were doing karate. That is the speed of this competition. The United States is winning on the dimensions Washington tracks: the models and the chips. But we are losing where the future will be decided: data implementation. We are seeing two different races play out, and I fear right now the United States may be winning the wrong one. Nowhere is that gap more alarming than in physical AI and robotics. Unlike large language models, which train on data scraped from the internet, robots have no such resource. A robot learns by doing in the physical world. That requires thousands of hours of real-world data collected, labeled, and ready to train on. This is why autonomous vehicles spend months driving city streets before launch. Every mile is training data. For robotics, think warehouse-scale facilities filled with machines running around the clock, not to move goods, but to generate the fuel that trains the next generation of AI. China has industrialized that process; we have not. According to Scale's internal estimates, China owns roughly 90 percent of the commercially available robotics AI data market today, and they produce that data at 60 percent lower cost than U.S. companies can match. The U.S. government invests roughly $100 million a year in AI data; China invests $1 to $1.5 billion, 10 to 15 times more. They have seven government-directed data labeling hubs and warehouses with robots running 17 hours a day solely to generate training data. By all accounts, this is a state-directed campaign to win. That structural advantage has direct national security consequences. The same platforms training on Chinese data today are candidates for deployment in American ports, hospitals, defense facilities, and homes tomorrow. The data that trains these systems shapes how they behave, what they optimize for, and where they are vulnerable. Seeding the robotics data layer to China is not a technology policy concern; it is a critical infrastructure risk, and the window to address it is closing fast. Winning requires a whole-of-government approach and four key actions. First, investigate Chinese robotics data as a national security threat and treat U.S. data as a national asset. Congress should ask the national security apparatus to study this dependency and require U.S. government data assets be made AI-ready. Second, establish one national regulatory framework for artificial intelligence. This should govern outputs, not inputs. A patchwork of state rules is not a regulatory system; it is a competitive liability. Third, mandate government AI deployment. The Department of War has transitioned one AI program to program of record since 2016. China deployed 81 large language models into their People's Liberation Army in the first half of 2024 alone. One versus 81. That is not a technology gap; it is a failure of will. Congress should direct every agency to launch at least five flagship agentic AI programs by 2028 to start to fill the gap. Fourth, export the AI technology stack to our allies before China does. China won on 5G because America didn't show up with a competitive alternative. We are at that same fork in the road on AI right now today. The Exporting the American AI Tech Stack executive order is a great first step, but Congress must codify and resource it. This is not the first time America has faced a defining technological competition, and it is not the first time the outcome looked uncertain. We have every asset required to win: the talent, the capital, and the allies. The variable is urgency, and the time to act is now. Scale is fully committed to supporting the United States in this race. We look forward to working with Congress and the Trump administration to find ways to do just that. Thank you again for the opportunity to be here, and I look forward to your questions.

Thank you very much. I now recognize Mr. Malchano for five minutes to summarize your opening statement.

Mr. Chairman and members of the subcommittee, I am the Vice President of Software at Boston Dynamics, and I serve as the company's Security Officer, responsible for overseeing our cybersecurity and data sovereignty protocols. At Boston Dynamics, our mission has always been to create a future in which humans and machines work together to improve everyone's safety, productivity, and quality of life. Our robots are in use in manufacturing facilities, semiconductor fabrication, energy plants, and by first responders. Our robots are used by the Secret Service to protect the President and by military agencies to neutralize the threat of IEDs. One of our robots has taken bullets in the line of duty, keeping police and bystanders safe from harm. We have a growing workforce of over 1,000 employees and proudly design and manufacture our robots in the United States. There are at least four reasons why robotics should be a top policy priority for Congress. First, advanced robots are the physical manifestation of AI. Robots leverage AI techniques such as computer vision and reinforcement learning to perform a wide range of useful tasks. In order to win the AI race, we must also win the robotics race. Second, robots are increasingly crucial in manufacturing. Humanoids allow for automated physical operations in workspaces designed and built for humans. The countries that build and deploy robots will be the ones that will manufacture on their own shores and address demographic challenges. Robotics will be crucial to economic security. Third, robots present new and unique classes of cybersecurity risk and beyond AI and computer systems. The combination of hardware, software, sensors, mobility, and communications makes advanced mobile robots higher risk than other IoT technologies. Attackers could sabotage those systems to interfere with manufacturing lines or halt security patrols. Robots are becoming a potent threat vector to critical infrastructure. Fourth, robotic technologies are dual-use. Robots with enhanced mobility and AI-powered capabilities will become increasingly important in military operations. We have seen how aerial robotics, namely drones intended for consumer use, can become a pivotal technology in warfare. Robots increasingly will become part of national defense strategies. The Chinese government understands this. This is why China just released a new five-year plan that identifies robotics as their top new industry tracks. The plan envisions transforming virtually every major industry in China by integrating AI-powered robots. It unlocks access to funding and implements policies that drive adoption across industries, creating a flywheel effect for robot AI. As one analyst remarked, this is less an industrial policy for robots than an industrial policy through robots. This government support isn't new. Previous government plans from the past decade outline the many mechanisms and policies that have supported Chinese robotics industry. These plans are already working. China last year adopted and installed 10 times more robots in factories than the United States. China also uses its little giant designation to provide companies with government funding and support. One of the companies designated as a little giant is Unitree, the most visible Chinese company selling legged robots in the U.S. Unitree is rapidly capturing market share in the U.S. university research and law enforcement markets. Researchers have reported crucial Unitree's cybersecurity vulnerabilities. Unitree's products have also been seen in PLA military exercises, and its ties to Chinese military have been noted by the House Select Committee on China as well as the Department of War. So it's not surprising that Unitree has captured the attention of this committee and the U.S. government. However, there are dozens of other Chinese companies working to develop advanced robotics technologies powered by AI. At the recent CES Las Vegas show, Chinese robots on display, humanoids, outnumbered American humanoids by five to one. While Unitree deserves close scrutiny, a broad whole-of-government strategic approach is what's truly needed. This is why we are asking for the development and launch of a national robotics strategy, consistent with a broad set of policies designed to support the growth and success of the industry. Time is running out for policy efforts to have an impact. The recently introduced National Robotics Commission Act, H.R. 7334, would kickstart the process of developing such a strategy. We urge Congress to pass this bill. We also respectfully ask this subcommittee to direct CISA to swiftly undertake a security evaluation of foreign robots. Independent researchers have already revealed crucial vulnerabilities, but a more comprehensive, government-directed study would provide broader guidance to users and government about the risk of foreign robotics projects. Thank you very much for the opportunity to testify. I welcome your questions.

Thank you very much. I now recognize Mr. Robbins for five minutes to summarize his opening statement.

Chairman Fong, Ranking Member Walkinshaw, distinguished members of the committee, thank you so much for the invitation to testify today. My name is Michael Robbins. I am the President and CEO of AUVSI, the trade association for the autonomous systems, drones, and robotic industry. Our members, including Boston Dynamics, are at the leading edge of the transformation where AI, robotics, and autonomy are converging to create a new frontier of cyber-physical systems. Today, we are at the precipice of a new era defined by the convergence of AI and physical machines, and it holds immense promise for American productivity and competitiveness, but also introduce unprecedented national security vulnerabilities when these systems are developed within adversary-controlled technology ecosystems. Modern robots are highly networked physical cyber systems. A compromised laptop can expose data, but a compromised robot can expose data and map, move, surveil, or physically disrupt operations in the real world. We are here today because our adversaries, specifically the People's Republic of China, are methodically executing a centrally planned playbook to hollow out the U.S. robotics industrial base and install Trojan horses within our critical infrastructure. Unfortunately, we have seen this exact threat before, and we cannot ignore the lessons learned from the commercial drone industry. Over the last decade, the PRC leveraged massive state subsidies, IP theft, and deliberate market dumping to drive many American drone manufacturers out of business by flooding the global market with artificially cheap systems. This deliberate market capture engineered a profound national security crisis here in America, exposing sensitive U.S. data to Chinese state intelligence and leaving our nation dangerously dependent on adversary-built technologies. Today, the PRC is aggressively executing the exact same playbook in the advanced robotics sector. Fueled by over $137 billion in state subsidies and initiatives like Made in China 2025, PRC robotics firms are systematically undercutting American companies, starving U.S. firms of the revenue needed to scale domestic capacity and maintain our technological leadership. If we fail to act, the U.S. robotics and autonomy industrial base will be hollowed out, leaving us dangerously dependent on adversary-linked technology to run our economy. Furthermore, we must address this vulnerability in the upstream supply chain, where PRC's dominance of rare earth minerals and the threat of export controls could effectively decapitate American robotics manufacturing before a single system is even built. This aggressive market capture is not just an economic issue; it is a profound national security threat. The PRC is actively weaponizing advanced civilian robotics to enhance its military capabilities. Recent footage released by the People's Liberation Army features a video ominously titled, "The Robot Dog's Time to Kill Has Come," depicting a Unitree Robotics quadruped equipped with an automatic rifle striking targets with precision. Because of Beijing's military-civil fusion strategy, PRC robotics firms flooding our markets are directly serving as R&D extensions of the PLA. Even more alarming, PRC companies pair their robots with subsidized LiDAR from firms like Hesai and DJI's Livox, companies our Department of Defense designates as Chinese military companies. These sensors generate precise 3D maps and persistent spatial awareness, thereby handing the CCP a persistent digital map of our nation. Furthermore, PRC AI firms like DeepSeek are accelerating this threat, illicitly extracting the capabilities of advanced U.S. AI models to rapidly and cheaply build the brains for these subsidized robotic platforms. And the PRC's national intelligence and cybersecurity laws create a homeland security crisis, compelling all Chinese commercial entities to provide state intelligence unfettered access to their systems. Allowing PRC market dominance is a national security nightmare for America, allowing the CCP to install Trojan horses within U.S. critical infrastructure, exposing our nation to severe risks of data exfiltration, remote operational disruption, and the weaponization of our supply chains. We simply cannot allow this dangerous erosion of American industrial capacity and national security to happen. Recognizing this, AUVSI recently launched our Partnership for Robotics Competitiveness. Securing our future requires both carrots and sticks. We must pair aggressive restrictions on adversary technology with bold incentives for domestic production to restore a level playing field and ensure American technological leadership. We urge Congress to pass a coordinated national robotics strategy and update trade enforcement mechanisms to counter PRC dumping. We must address risk-based restrictions on PRC technologies in sensitive environments, including bans on PRC UGVs and on the importation of 1268H companies, while also deploying robust economic incentives like production tax credits, loan guarantees, and employing advanced market commitments to stimulate domestic manufacturing and de-risk private investment. We must learn from the past to protect our future. The window to ensure American dominance in autonomy and robotics is closing, and the stakes could not be higher. By acting today, we ensure that the next generation of autonomous systems bolsters rather than threatens the security and resilience of the United States. Thank you, and I look forward to your questions.

Thank you, Mr. Robbins. I now recognize... Thank you, Mr. Robbins. I now recognize I now recognize Dr. Doshi for five minutes to summarize your opening statement.

Thank you, Chairman Fong, Ranking Member Walkinshaw, distinguished members of the subcommittee. Thank you for the opportunity to testify today. My core argument today is simple. AI and robotics are changing the global balance of power. China has moved faster and achieved more than most Americans realize, but we still have advantages. The question before this subcommittee is whether we will use them before China pulls too far ahead. I'd like to structure my testimony to answer four questions. First, what are China's objectives in AI and robotics? The Chinese Communist Party aims to catch up and surpass the West. It believes there have been four industrial revolutions that determined the fate of nations. The first was steam power, which brought British dominance. The second was electrification and the third was mass manufacturing, which brought American dominance. And now we're in the fourth: quantum, biotech, and of course, robotics and AI, which China is determined to win, not just for the sake of prosperity, but for the sake of relative power. That brings me to a second question: how is China doing? In robotics, China increasingly is winning. In 2024, Chinese factories installed roughly 300,000 new industrial robots. American factories installed 30,000. Today, there are now more than two million industrial robots working in Chinese factories. In the U.S., the number is one-fifth that. China is investing in the future. It's pulling ahead in research publications and patent filings and pouring hundreds of billions of dollars into the robotics sector. Just 10 years ago, China imported three-quarters of its industrial robots. So how did all this happen so quickly? Part of it is because China is the world's greatest manufacturer, with two times our share of global manufacturing. But a lot of it is because of sustained government focus. China acquired robotics know-how from German companies like Kuka. It protected its home market. It then launched a decade of industrial policy programs like Made in China 2025 to dominate the sector. This year alone, China may spend $400 billion on industrial policy. The U.S. CHIPS and Science Act, by comparison, provided $50 billion over multiple years. In AI, the story is more complex. China has made AI an industrial policy priority for more than a decade. It's the only country that rivals the United States. Many of its open-source models are on top lists for key benchmarks, and it's diffusing those models throughout society. But America is still ahead at the frontier. Unlike in robotics, where China's manufacturing advantages pay rich dividends, China's progress in AI relies on access to compute, that is access to advanced AI chips. China's own AI companies openly admit this, and reports show that U.S. chips are four to five times better. They may soon be 17 times better in the next two years, that the U.S. has a growing share of global compute. If we surrender this advantage, we will probably fall behind. China has more energy. It added eight times more power to the grid last year than we did, and it has more talent, but for now, it has less compute. That leads me to a third question: what are the risks of falling behind in robotics and AI? Let me offer three. First, robotics will be critical to manufacturing. If we lose robotics the way we lost solar panels and batteries and EVs, we will hollow out our American manufacturing ecosystem and become even more dependent on China. We've seen that in rare earth minerals and magnets. It's not wise to put ourselves in that position of dependency. Second, China's national intelligence law, its counter-espionage law, require Chinese companies to cooperate with Chinese intelligence and not disclose that fact to others. That creates espionage risks. These robots are equipped with LiDAR systems, microphones, and cameras that make them mobile surveillance platforms. Already, Unitree's quadruped robots have been shown to have backdoor camera access. And its humanoid robots transmit data back to China every five minutes. But espionage is not the only risk. Last year, I testified before this committee on how China was pre-positioning for destructive cyberattack on American power, gas, water, telecom, and transportation infrastructure. We simply cannot employ robots in sensitive facilities from the very country that is targeting those facilities. A third and final risk is that AI and robotics are dual-use. China is using American AI tools to launch cyberattacks. It's incorporating armed robots into military exercises. If we fall behind, it means our deterrence will be less effective. Which leads me to the question of what we should do. First, we should extend the ICTS framework that we apply to connected vehicles to cover PRC robots. That limits their importation unless China addresses key vulnerabilities. Second, we should direct CISA to audit where robots and AI models are deployed in critical infrastructure so we know the risks that we face. Third, we should prohibit federal agencies and federal contractors from procuring PRC robotic systems or AI systems. We currently treat American AI companies like Anthropic with more regulatory scrutiny than we treat Chinese ones like DeepSeek. Fourth, we should maintain and strengthen semiconductor export controls, the one advantage we retain in AI, and limit remote access and smuggling. And finally, we should build allied scale to rival China's scale in the sector. That means better terms for allies and partners than for Beijing, and it gives us a chance to bring down costs and reduce China dependence. China believes this is a pivotal moment. They're probably right. We should believe it too. And with that, thank you, and I look forward to your questions.

Thank you, Dr. Doshi. Members will be recognized by order of seniority for their five minutes of questioning. I now recognize myself for five minutes of questioning. My first question goes to you, Mr. Fenkell. Many American startups and developers are beginning to build on Chinese open models like DeepSeek because they appear highly capable and are offered at extremely low cost. We have seen this strategy before. Chinese companies like DJI use aggressive pricing and rapid deployment to dominate the global drone market, which made it extremely difficult for American companies to compete. Unlike the drone sector, however, the United States already has leading AI developers that are building highly capable frontier models. Do you see Chinese AI companies attempting to follow a similar strategy with models like DeepSeek by rapidly expanding global adoption in order to capture market share before trusted alternatives can compete?

Thank you for that question and yes, simply, I think. As we've seen play out in AI, the diffusion of the technology will ultimately win out. And I think for the last year or so, we've believed it's been a model-based adoption, and I think we're shifting into a belief of a tech stack-based adoption. And so when you look at the AI tech stack, it fundamentally is made up of compute, data, model, and applications. And I think we've seen people start begin to recognize in order to really win on AI, you have to win on the tech stack, which ultimately solves a hard problem. And we've seen China look to imitate their Belt and Road Initiative, their Digital Silk Road Initiative to ensure that their models are winning across the world. It's imperative, and we spoke about it in our testimony, that the U.S. not only counters this but wins on it. And that's why we feel really strongly that the Exporting American AI Tech Stack Executive Order is a critical path forward on this. That must be codified and resourced, and then we as industry have to do everything we can to bring our technology to the world.

So as a follow-up question, but in terms of censorship and information control by DeepSeek. Researchers and journalists have found that DeepSeek censors its own answers in real time, deleting responses related to topics like Tiananmen Square, Taiwan, or criticism of the Chinese Communist Party. In some cases, the model initially generates accurate information but then erases it and replaces it with a different answer. What risks arise when an AI system that may reflect information controls and propaganda narratives of an authoritarian government become widely used around the world?

I believe there are a number of different risks associated with it. And what we've seen is that in order to release a model in China, you have to pass a test essentially adhering to those laws that you referenced. I think Scale as a company is now doing more AI evaluations than nearly anyone else, and we evaluate all of the world's leading models to better understand their capabilities. AI evaluation is an amazing tool for this. It's really important, though, that we're putting models out to the world that we believe have democratic values and ideals and uphold American values. And that's again why it's important that we not only match but exceed China's intensity in diffusing our tech globally.

Thank you. Mr. Malchano, unlike traditional Internet of Things devices, robots are not just collecting data. They are physical machines capable of movement and interaction with the environment. If a malicious actor gained control of a connected robot operating in a sensitive environment, such as a research laboratory, industrial facility, or transportation hub, what kinds of physical or operational risk could arise?

Thank you, Mr. Chairman. Those risks are very similar to the kinds of risks that would happen if you allowed someone into your environment that was a hostile person, right? And so that robot could go around, manipulate things, look at things. To make that very realistic, if you had an assembly line, that robot could destroy the assembly line or otherwise take malicious photos of materials going through. So in many ways, a robot represents a kind of risk that is beyond the passive risks of IoT devices if compromised.

Thank you. Mr. Robbins, we've already seen what can happen when Chinese technology rapidly dominates critical markets, like I mentioned with DJI drones. What lessons should policymakers take from that experience when thinking about AI systems, robotics platforms, and autonomous technologies developed by Chinese companies?

Yeah, thank you for that question, Mr. Fong. The number one lesson is don't wait until the problem is upon us and we get behind. Take action now to level the playing field for U.S. robotics and AI companies and autonomy companies so they can compete globally on a level playing field now instead of having to fight from behind like American drone companies are doing today. There are policy actions Congress can take. There's actions the executive branch can take now with existing authorities to level the playing field. We spell those out in our Partnerships for Robotics Competitiveness white paper.

Let me throw this out to everyone, but in terms of procurement, should Congress consider a policy similar to existing restrictions on Huawei or DJI where certain Chinese technologies are prohibited from procurement by federal agencies? I've got 15 seconds, so maybe I'll just open it up to everybody.

Absolutely. Yes.

Yes.

Yes, in line with the things that have been done in the past with DJI.

Yes.

Thank you. With that, I now recognize the ranking member, the gentleman from Virginia, Mr. Walkinshaw, for his five minutes of questions.

I'd just like to add that the supply chain for robotics is very broad and involves many high-tech areas, which is all a reason for us to consider the national strategy as a way to create a central point of contact for both sides of that supply chain to interact with the U.S. government and figure out the best ways to make that move forward on American soil.

Thank you, Mr. Chairman, and thank you to our witnesses. Really powerful opening statements. I want to start just on international collaboration and supply chains. We know, and you all ably outlined, the ways in which the PRC's threat poses a significant challenge not just to the United States but to our allies around the world. While we have significant resources and capabilities that perhaps we need to do a better job of bringing to bear, our allies also have resources and manufacturing capabilities and supply chain capabilities that could help us in this effort. Dr. Doshi, your testimony discusses the importance of working with allies like Japan to develop alternative supply chains to avoid reliance on the PRC. Can you just elaborate on the importance and maybe the opportunity of international collaboration in combating PRC supply chain risk? How can we ensure that our policies are implemented in conjunction with key allies and partners around the globe?

Thank you very much for the question, Ranking Member Walkinshaw. Let me begin with one quick point, which is that China outscales the United States when it comes to manufacturing. They are two times American manufacturing value add. How do you deal with a competitor that outscales you in manufacturing? The only way is to add up American manufacturing with allied manufacturing. Collectively, that block completely outscales China. So what I'd like to see is a preferential system, a trading block if you will, among allies where we lower the barriers to each other and raise them against China so that we have that chance to sell to each other and build scale within our own collective market. Right now, there are allies like Japan which dominate key inputs for the robotics industry. We should help those companies succeed. Maybe they can set up plants in the United States. But either way, if those companies disappear, then the only game left in town for the robotics supply chain is China, which we simply cannot allow. Mr. Ranking Member, we've seen this story in industry after industry. We're seeing it in cars, we're seeing it in pharmaceuticals, we're seeing it in biotech. We don't want to see it in robotics because it is a critical enabling technology. So one way we could do this kind of collaboration, in addition to tariff-based or trade-based policy, is regulatory policy. We have a tool, the ICTS tool, which basically allows us to prohibit or import control certain finished products from foreign countries that are threats when connected to our networks. We can align that regulatory tool with our allies and partners and create a bit of a wall that keeps certain Chinese products out and gives ourselves and our friends time to rebuild. Thank you very much for the question, Ranking Member Walkinshaw. Let me begin with one quick point, which is that China outscales the United States when it comes to manufacturing. They are two times American manufacturing value add. How do you deal with a competitor that outscales you in manufacturing? The only way is to add up American manufacturing with allied manufacturing. Collectively, that block completely outscales China. So what I'd like to see is a preferential system, a trading block if you will, among allies where we lower the barriers to each other and raise them against China so that we have that chance to sell to each other and build scale within our own collective market. Right now, there are allies like Japan which dominate key inputs for the robotics industry. We should help those companies succeed. Maybe they can set up plants in the United States. But either way, if those companies disappear, then the only game left in town for the robotics supply chain is China, which we simply cannot allow. Mr. Ranking Member, we've seen this story in industry after industry. We're seeing it in cars, we're seeing it in pharmaceuticals, we're seeing it in biotech. We don't want to see it in robotics because it is a critical enabling technology. So one way we could do this kind of collaboration, in addition to tariff-based or trade-based policy, is regulatory policy. We have a tool, the ICTS tool, which basically allows us to prohibit or import control certain finished products from foreign countries that are threats when connected to our networks. We can align that regulatory tool with our allies and partners and create a bit of a wall that keeps certain Chinese products out and gives ourselves and our friends time to rebuild.

Anyone else with a thought on that? Mr. Robbins, you looked antsier there. Anyone else with a thought on that? Mr. Robbins, you looked antsier there.

I gave a speech last May where I quoted a very smart individual where I noted that if our authoritarian rivals can work together to undermine democracy, then democratic rivals must work together to protect democracy. That person was Dr. Rush Doshi. So I completely agree with what Rush said. We have an I at the end of our name. We work very aggressively to enhance partnerships with international democratic allies like Taiwan, like Japan, in Europe, and throughout the world. That is really the key to the future of the robotics and autonomy and AI industry is ensuring that America leads and that Western values are driving this technology forward. But as Rush said, we're stronger when we do that together and we work with our Western allies on this front. I gave a speech last May where I quoted a very smart individual where I noted that if our authoritarian rivals can work together to undermine democracy, then democratic rivals must work together to protect democracy. That person was Dr. Rush Doshi. So I completely agree with what Rush said. We have an I at the end of our name. We work very aggressively to enhance partnerships with international democratic allies like Taiwan, like Japan, in Europe, and throughout the world. That is really the key to the future of the robotics and autonomy and AI industry is ensuring that America leads and that Western values are driving this technology forward. But as Rush said, we're stronger when we do that together and we work with our Western allies on this front.

Dr. Doshi, your written statement referenced extending export controls to inference chips. I want to give Mr. Fenkell and Mr. Malchano a quick opportunity to describe for us why are inference chips so critical when it comes to robotics relative to training chips? What is the distinction? Why is it relevant to robotics?

So training chips are relevant to robotics in order to train, but in terms of what is deployed inside the robot is a different kind of chip, an embedded processor designed to be able to run ML applications quickly at a low power state inside of a robot. Those chips are just as important and exist one per each robot that you build. So although America is doing a good job of thinking about the plans with regards to the chips used for training, we would encourage also thinking about inference chips as are distributed in an embedded context.

Mr. Fenkell, anything on that? Training versus inference chips?

Nothing to add specifically. I think he covered it. I do think getting back to the allies point, there's two sides where ally cooperation is very important. One is diffusion of the technology stack, and we've spoken about that already. The second is on standards collaboration. I think we look at the Center for AI Standards and Innovation, the US AISI, which is housed at the Department of Commerce. There are also AI safety institutes and security institutes around the world, and it's critical that we're working together on the measurement science for these technologies because ultimately what we saw in 5G is China began to win that race when they won on standards. On AI, those standards are currently being written, and it's critical that they're being written by America and Western-aligned countries.

Thank you. Thank you, Mr. Chairman.

Thank you very much. Maybe we'll do one more round of questions before we conclude this hearing. I wanted to follow up on the last question I asked in terms of procurement. You certainly all answered in the affirmative that we should have a policy restricting certain technologies from procurement by federal agencies. I wanted to ask about the supply chain then in that aspect. China has spent years positioning itself at key points in global technology supply chains. In a geopolitical crisis, what technology supply chain choke points could China potentially exploit to pressure the United States or our allies? How do we go about this? I'll open that question to all of the panelists and leave that to you as to how you guys want to answer it.

Maybe I'll start with the data portion of it. Obviously, Scale is one of the leading companies working on providing training and evaluation data to nearly every frontier AI lab as well as the robotics industry today. We work with a number of robotics companies and have done over 250,000 hours of robotics data to those companies. What we've seen today, though, is still China owns 90 percent of the commercially available AI data market for robotics, and obviously the cost is much higher at about 60 percent. So that is a massive dependency on the Chinese supply chain, and obviously during a geopolitical context, that can become an issue.

You just made sure you said 90 percent?

That's based off our internal estimates, yes. So what we've seen is that in a geopolitical context, that's a big issue. So in our testimony, we've asked the national security apparatus to begin to look into this dependency and then also not only counter but exceed their intensity on data. They're outspending us tremendously. They value this far greater. So as we look at it, every federal data asset should be made AI-ready. Every asset that can be open-sourced should be open-sourced. Then ultimately, we have to value data as a critical part of the supply chain.

I'll just add, sir, that in my, as I said in my opening statement, it has to be a policy of both carrots and sticks. You asked previously about the restrictions, and that's an absolutely critical piece of getting this right. But there also have to be incentives to scale domestic manufacturing to break some of those critical choke points. One good example where there is some good work being done right now is in the area of rare earth magnets. We experienced this in real time in our industry last year in May when China began throttling export controls on rare earth magnets as part of the broader trade war with China between the US. That was sending shockwaves through our industry because, as I mentioned in my opening statement, without those rare earth magnets, we're not building robots, we're not building drones. We're also not building Columbia-class submarines or F-35s. It is a critical dependency on China. So in the United States, companies like Vulcan Elements have worked with the U.S. government to leverage existing tools like the Office of Strategic Capital and leftover funds from the CHIPS Act to drive scaling manufacturing here in the US. We need to double down on that in a significant way both on that industry as well as other industries as well, including in batteries, in lithium and graphite, some of the critical minerals that make up batteries as well. Thank you, sir.

Mr. Chairman, I'll just add one quick point. I agree with what Mr. Robbins said. Right now, if we were to just do a blanket prohibition on Chinese robots, it would be effective in securing elements of our national security. Of course, those robots are cheaper than the ones that we make here at home. So the question then is what do we do about that price differential? That is where the government has to step in and provide a solid industrial policy framework to help drive this industry. Let's just look at the historical record. Again, 10 years ago, China was not a leader in robotics. They were importing most of their industrial robots, more than 65 percent. Today, they produce those robots at home. The way they did that was a careful combination of market restrictions and market access protection combined with industrial policy. So some combination of those tools will have to be in our toolkit going forward.

Then on that note, is it possible then, I mean the phrase decouple, but is it possible to nearshore or decouple from China in the context of supply chain?

It is possible. It will take time and it requires, as the other witness said, a concerted national strategy. If we go about this in a sort of one-off way here and there, we won't get there and we won't get there in time.

Mr. Chairman, if I could just add one point. China still has its own dependencies in its robotics supply chain, dependencies not just on our chips but on certain products manufactured in Japan. So when we think about the overall supply chain, we should think where do we have advantages and choke points that we can harness, where do the Chinese have choke points, and how do we mitigate those choke points?

Any of the other two, do you have anything to add? You want to jump in?

I think the only thing I'd add to that is as we look at the supply chain, we have to recognize all aspects of it. We typically focus on the platform-based aspects of it. Obviously from the Scale side, we definitely would recommend that we also include the data supply chain in there just given that AI ultimately is only as good as the data that it's trained on.

I'd also just add that in terms of a national strategy, demand-side activities like encouraging the additional use of robots in the federal government as well as incentivizing people to adapt robots and adopt their industries for the coming high-tech future are all key ways for us to push that supply chain to produce on American soil.

Thank you. Well, I want to thank the witnesses for your testimony and for the questions from the ranking member. Critically important short-term and long-term challenges that we need to address and we need to take care of these issues in real time. So thank you for your expertise. Members of the subcommittee may have some additional questions for the witnesses, and we would ask the witnesses to respond to these in writing. Pursuant to committee rule 7E, the hearing record will be held open for 10 days. Without objection, this subcommittee stands adjourned.