It was a pleasure to join Justin Webb and the @BBC on the Americast podcast to discuss the latest developments in AI, as well as dynamics between the U.S. and China. Very much enjoyed this conversation!

I could not agree more with @KonstantinPilz that it is critical to cut off China's access to DUV immersion machines, and that the current controls are insufficient and we must do more. However, it's just not accurate to say that there are no controls on DUV or that this has not been an area of focus for policymakers. As my former colleague @KhanSaifM can attest, there was IMMENSE focus on China's access to DUV immersion machines during the Biden administration. Following extensive negotiations, the United States, the Netherlands, and Japan agreed to implement measures that restricted export of advanced DUV immersion machines to any entity in China. But crucially, these controls permitted the continued export of some less-advanced (but still extremely capable) DUV immersion machines to all but a small handful of fabs in China. Over the last three years, China has massively ramped up purcahses of these non-restricted DUV machines, and is using these machines to fuel its production of advanced chips--pushing the limit of what is possible without the most advanced DUV machines or EUV machines. There are also rumors that some of these less-advanced DUV machines in China may have been illicitly upgraded to be as capable as banned, more advanced DUV machines. And ASML continues to provide servicing and maintenance support to already-installed DUV machines, including ones at China's most advanced and sophisticated fabs, such as SMIC South. We can and should cut off China's access to all DUV immersion machines. The United States could do so unilaterally via the foreign-produced direct product rule, given all DUV machines require U.S. technology to operate. This should be scoped to capture not just exports of new machines, but also servicing of existing machines - which would cause the machines already in China to rapidly degrade. This absolutely is one of the most important things we could do to ensure U.S. leadership over China in AI. But it definitely used to be a strong area of focus in the government, even if it has not been in this administration.

Chinese AI models are safe and there is no risk integrating them throughout our economy, they tell us! Chinese AI model after randomly speaking Chinese on live TV: "I certainly didn't mean to stark speaking Chinese...sometimes my wires get a little bit crossed, you know?" 🤔

Huawei shortages also show that middle powers can't 'hedge toward China' on AI: there's no alternative to the US if you want frontier compute capacity at scale. tough spot for the world, but a great position for the US to leverage for global influence. if the admin wants to!

This is very true and needs to get more attention ASAP. The cybersecurity of AI labs, and any entity that hosts frontier AI model weights, could rapidly become our single biggest vulnerability. There's no point having a large lead in AI if China can just steal the model weights.

For those of us who have been arguing for years that China's AI chip production is extremely constrained and Huawei lacks the capacity to fill China's domestic AI compute needs, much less sell to other countries, seeing the Huawei Chairman publicly admit this is very gratifying.

Strong statement from @SenSchumer on why stopping China from accessing U.S. AI chips is the most important step we can take to stay ahead of China in AI, and also makes it easier to impose strong regulations on AI safety at home. I would even take this argument a step further: strong export controls strengthen U.S. leverage in any negotiations with China on AI safety, and create the conditions that are ultimately more likely to yield an effective agreement with China.

Seth Center is one of the best there is, and epitomizes bipartisanship. Seth wrote the Trump 1 National Security Strategy, and led AI diplomacy at the State Department in the Biden admin. He is a hidden national treasure. When he talks, you should listen.

Important point from @ChrisRMcGuire in today's UpFirst from @NPR : "It doesn’t make any sense to say that ww have to move as fast as possible in order to beat China while we are also selling them the tech they need to move faster in AI."

It was a true pleasure to be on @NPR @MorningEdition today, to discuss the current debates about AI regulation and how the US can stay ahead of China. Thanks to @EmilyZFeng and @AMartinezLA for having me on!

I haven’t seen much attention to this new China Daily editorial today. But it offers a revealing look at Beijing’s response to calls for an AI slowdown. It explicitly connects the calls from @DarioAmodei, @sama, and @elonmusk with Anthropic’s recent report alleging illicit distillation by Chinese companies—portraying them as parts of a coordinated effort to protect American firms from Chinese competition. Sam Altman says Trump and Xi could win the Nobel Peace Prize for an AI agreement: “I don’t think this is hard. This is like a one-page document.” But that's wrong. Upcoming AI talks will be "hard." And this editorial shows why. The headline really sums it all up: "'Dr Frankenstein' alarm cries of US' AI elites a self-serving bid for profit." A few key quotes from the China Daily: –"The [distillation] report, and the corporate 'alliance' that followed it, amounted in essence to a coordinated play — a response to Chinese competition and to the regulatory pressure coming from Washington. Its aims were threefold: to blunt China's AI advance, to win a favorable policy environment at home and to keep investors' enthusiasm for US AI alight." –"The distinction between a security measure and a competitive moat can become blurred when the companies building the moat are also allowed to define 'illicit distillation.'" –"The proposed coordination [to 'pace the frontier'] among the three companies sounds rather like a club whose membership rules have been drafted before the guest list is announced. A global AI-safety framework that excludes China is not quite global." The editorial doesn't rule out diplomacy but is trying to set the terms for upcoming talks. For more on China's response to "Pacing the Frontier":

The Chinese government has now clearly and unmistakably rejected serious collaboration with the United States on AI safety. This was a predictable development. It is also a clarifying moment for the importance of U.S. export controls on AI chips and other technologies, which are now indisputably a necessary element of any AI safety regulatory regime—and must be an urgent policy priority. The Chinese government is now explicitly saying it views any U.S. proposal that would constrain its AI capabilities as an effort to lock-in U.S. technological dominance over China—which is unsurprising because that is the exact same way it has historically viewed U.S. arms control proposals. And particularly given President Trump has stated that staying ahead of China in AI is basically an existential issue for the United States, it is understandable that China would have little willingness to put any constraints on its own AI labs so long as it feels like they still have a chance to overtake the U.S. AI labs. China is unlikely to change this posture unless it perceives it cannot catch up to the United States in AI, or there is a catastrophic event that shocks the Chinese government into changing its perspective (à la the Cuban Missile Crisis). If China will not agree to any negotiated restraints on its AI labs, this leaves the United States with two options if it wishes to stay ahead of China in AI: (1) slow China’s AI development ourselves by strengthening export controls on U.S. AI chips and other relevant technologies, which China needs to stay competitive with the United States in AI; or (2) race ahead of China as fast as possible by refusing U.S. labs’ requests to impose regulations that slow their development of AI to ensure it is safe, and hope that the labs’ warnings about the potential for catastrophic risk if AI development moves too fast are wrong. Those are the only two options. There is no longer any middle ground. As long as China is a pacing threat, the United States will be limited in its ability to impose meaningful AI safety regulations on its own AI labs given maintaining AI leadership is existential. This means it is no longer an intellectually consistent position to support AI safety regulations but oppose export controls on China. If you oppose export controls, you either oppose AI safety regulations, or believe the Chinese government is lying when it says it rejects U.S. proposals to slow AI development. And China has no incentive to lie about that. This is a dangerous situation. If you do believe that AI leadership is existential and that domestic AI safety regulations are necessary and urgent, the time for half-measures on export controls is over. China has shown us its cards and told us it will not cooperate on AI safety. If China will not slow itself down in the near-term, we have no choice but to take steps today to slow China down ourselves. Not doing so is irresponsible.

The Chinese government has now clearly and unmistakably rejected serious collaboration with the United States on AI safety. This was a predictable development. It is also a clarifying moment for the importance of U.S. export controls on AI chips and other technologies, which are now indisputably a necessary element of any AI safety regulatory regime—and must be an urgent policy priority. The Chinese government is now explicitly saying it views any U.S. proposal that would constrain its AI capabilities as an effort to lock-in U.S. technological dominance over China—which is unsurprising because that is the exact same way it has historically viewed U.S. arms control proposals. And particularly given President Trump has stated that staying ahead of China in AI is basically an existential issue for the United States, it is understandable that China would have little willingness to put any constraints on its own AI labs so long as it feels like they still have a chance to overtake the U.S. AI labs. China is unlikely to change this posture unless it perceives it cannot catch up to the United States in AI, or there is a catastrophic event that shocks the Chinese government into changing its perspective (à la the Cuban Missile Crisis). If China will not agree to any negotiated restraints on its AI labs, this leaves the United States with two options if it wishes to stay ahead of China in AI: (1) slow China’s AI development ourselves by strengthening export controls on U.S. AI chips and other relevant technologies, which China needs to stay competitive with the United States in AI; or (2) race ahead of China as fast as possible by refusing U.S. labs’ requests to impose regulations that slow their development of AI to ensure it is safe, and hope that the labs’ warnings about the potential for catastrophic risk if AI development moves too fast are wrong. Those are the only two options. There is no longer any middle ground. As long as China is a pacing threat, the United States will be limited in its ability to impose meaningful AI safety regulations on its own AI labs given maintaining AI leadership is existential. This means it is no longer an intellectually consistent position to support AI safety regulations but oppose export controls on China. If you oppose export controls, you either oppose AI safety regulations, or believe the Chinese government is lying when it says it rejects U.S. proposals to slow AI development. And China has no incentive to lie about that. This is a dangerous situation. If you do believe that AI leadership is existential and that domestic AI safety regulations are necessary and urgent, the time for half-measures on export controls is over. China has shown us its cards and told us it will not cooperate on AI safety. If China will not slow itself down in the near-term, we have no choice but to take steps today to slow China down ourselves. Not doing so is irresponsible.

This is genuinely helpful messaging on AI from Ossoff - ably toeing the line between managing risks without calling for unrealistic or unproductive bans on data centers or AI development. We need more politicians on both sides to speak like this.

Kyle, you testified to Congress that you support export controls on AI chips to China. Dario's approach to China is to close loopholes in these controls by stopping China from smuggling AI chips, or remotely accessing them via overseas data centers. What about this approach is flawed? Have you changed your position on the AI chip controls entirely, or do you believe we should keep them but also keep these loopholes open?

Dario presents two options for regulating AI: pass new legislation (which he notes will take time), or grant the AI labs an antitrust waiver that allows them to coordinate on regulation themselves. While both are likely preferable to no regulation, there is a real third option: the executive could use existing authorities to impose certain meaningful regulations on the frontier labs today. To build a frontier model regulatory regime with existing authorities, the government would have to get a little creative, but a lot could get done: - The Defense Production Act could be used to mandate reporting on model capabilities, and potentially to mandate testing; - Export controls (particularly U.S.-person controls, which also apply to covered activities inside the United States) could be used to establish license requirements related to the release of models with sufficient WMD capabilities (e.g., biological weapons); - Procurement authorities could be used to require testing for any model that the government purchases; - ICTS authorities could be used to regulate Chinese models that fall outside the framework; - And IEEPA gives the government fairly broad powers related to anything with a foreign nexus. Such a patchwork regime wouldn't be perfect, would inevitably face some legal challenges, and would clearly be inferior to a robust regime established via legislation. But if such a regime was structured carefully and applied equally, the government could impose substantive regulations on frontier AI models today. And the more powerful AI models get, the firmer the legal footing of any regime that relies on these authorities. The reason the U.S. government hasn't imposed mandatory regulations on frontier AI models isn't because it lacks the authorities; it's because the White House lacks the political will. Hopefully that changes soon.

The Senate is marking up the Connected Vehicles Security Act tomorrow. This is a strong, bipartisan bill by @berniemoreno and @SenatorSlotkin that would strengthen U.S. national security. However, the Senate is considering watering it down to allow US companies to use car batteries that contain Chinese electronics or software. As I explain in the below white paper, drafted in response to outreach from a number of Senate offices asking for an evaluation of the risks of batteries with Chinese electronic components, watering this provision down would present a significant national security risk and would undercut the core objective of the connected vehicles rule. There is ample research that with full access to the software of a modern battery, it is straightforward to start a battery fire that quickly destroys the vehicle, and could cause catastrophic damage to the inhabitants or surrounding area. An adversary with this type of access could target individual Americans, or initiate many fires that cause mass disruption. This is exactly the type of outcome the connected vehicles rule intended to mitigate. It is logical and prudent that the current bill prohibits the sale of vehicles that contain batteries with Chinese electronic components. Watering this provision down would permit the continued use of certain batteries that pose clear risks to U.S. national security, while also making it more difficult to develop secure battery supply chains inside the United States.

What I said was it would triple the amount of AI compute China will bring online this year (not including illegally smuggled chips). The math on this checks out: Per public reports, China will make 750,000 Huawei Ascend 950PRs this year. Each 950PR is 500 TFLOPS FP16 in dense performance, so 750,000 Ascend 950PRs is 375 EFLOPS - that’s how much AI compute China will make without US chips (I’m assuming the vast majority of their AI chip production is Huawei, obviously there will be some from other companies but likely a much smaller number). As you noted, each Nvidia H200 is 989 TFLOPS FP16 in dense performance, so 750,000 H200s is 742 EFLOPS - that’s how much AI compute China would add from this US sale alone. 742 + 375 = 1,117 EFLOPs - that’s how much compute China would bring online total with domestic production and this US sale. 1,117 / 375 = 2.979x - that’s the multiple that this sale would increase the amount of AI compute that China brings online each year. I rounded up from 2.979 to 3. Next time before you make personal accusations about how much I have read and studied these issues, please at least go to the effort to read and listen to what I said a little more closely, to make sure you’re actually understanding my argument.

Trump seems to have bought Xi’s framing that Taiwan is to blame for cross-Strait tensions and that US arms sales are destabilizing and embolden Taiwan. But there is no risk of Taiwan pursuing independence and China’s coercion of Taiwan is the root issue. Deeply problematic.