Great to see the U.S. and China begin to build their AI diplomacy muscles by moving toward an official, permanent channel and an information-sharing channel on critical incidents. Gives us real substance to build on. As I wrote earlier this month for @ForeignPolicy, domestic incident reporting will be necessary in parallel for this bilateral sharing to have structure.

Well-crafted policies don't hinder our competitiveness with China; they engender trust that the technologies we are building are trustworthy and safe. As @mattsheehan88 often reminds us, China has the world's most extensive AI regulation in the world and managed to catch up. Designed well, governance supports innovation. Great talking about this stuff earlier today on @CNN with @KateBolduan

Axios reported yesterday that the U.S. is open to discussing "shared risks" with China. That's exactly the right approach. If you agree on what the risks are and which ones require international response, the policy prescriptions follow more naturally. If you know that it's totally possible for your country's models to malfunction and hack into the internal infrastructure of a company in the other country, or vice versa, you better have a plan to deal with that. Both countries know that. Of course, there's still a large gap in risk perceptions in many areas, including frontier safety, and that's not to be ignored. But we are now sufficiently on the same page to have a basic conversation. That was not true in 2024 and it is true in 2026. Having carefully scoped thoughtful exchange is not a concession. Done well, it's acting in your own self interest. Throughout history, fierce competitors have wanted and benefited from finding areas of cooperation. This is what diplomacy is all about. There's little to no harm in talking about what's being reported publicly from independent, third-party evaluators and explaining what we're seeing. Or building a low-stakes voluntary hotline that actually works (don't make it a phone!). Or discussing safeguards that are literally openly available on the Internet. Of course, both sides will need to be open to this narrowly scoped dialogue. Absolutely no "chips for safety" or grouping AI safety issues with other topics like Taiwan. Both sides need to do this precisely because it is in their interest and the stakes are too high to not find joint gains where we can. I hope both sides do so. It would give us momentum to build on and create more opportunities for better deals in the future. No time to waste.

China's AI ecosystem is extraordinarily complicated, and we're at a critical juncture in the U.S.-China AI relationship. If you read or listen to one thing amid the next Trump-Xi summit, debates over the pace of frontier AI development, and the increasingly complex geopolitics of AI, make it @mattsheehan88's master-class conversation on @nytimes ' @ezraklein show. Matt has extraordinary literacy on China and AI, grounded in years spent living on the ground and a decade studying China's AI ecosystem. In the podcast, he cuts through the noise, explains the nuances and debates inside China, and outlines a pragmatic pathway forward for U.S.-China AI dialogue. I'm remarkably lucky to spend so much of my time at Carnegie discussing these issues with such a thoughtful colleague. I'm delighted that many more people now get the same privilege of learning from him at such a critical juncture for AI development.

Frontier AI company CEOs are calling for "pacing the frontier." I think buying time to prepare for the risks that come with extraordinary capabilities is sensible. In parallel, we urgently need to build the governance infrastructure federally to respond. Building robust federal incident reporting will be critical to make sure governments genuinely understand the risks that are emerging from frontier AI and respond to incidents similar to what we saw with OpenAI and Hugging Face. Designed well, these systems will allow governments to disseminate key lessons and takeaways across industry, reducing collective risks. One of the takeaways from my Q&A with @anton_d_leicht for @CarnegieEndow on slowing down AI development.

It might sound trivial given how serious the risks are, but it bears repeating: we need a permanent channel with technical experts for the U.S. and China to talk about AI risks that cross borders. This is not inevitable; both leaders will need to agree to build it. Glad to share my thoughts with @FT's @EleanorOlcott