A MESSAGE FROM ANTHROPIC

Claude, the AI without ads

A space to think. Anthropic keeps conversations with Claude ad-free: no sponsored links, no advertisers shaping answers, no paid product placements you didn't ask for. When you bring your hardest problem to an AI, you shouldn't have to wonder who it's working for.

Learn more

And now, the main event…

President Trump’s two-day summit last week with Chinese President Xi Jinping—his first trip to China since 2017—included all sorts of agenda items: the Strait of Hormuz, the future of Taiwan, global trade, etcetera. But like everything else these days, the real focus came down to artificial intelligence, which has served as the central battleground of the increasingly tense cold war between the two countries. From Pennsylvania Avenue to Sand Hill Road, the binary race for A.I. supremacy has become the existential national-security justification for the White House’s broadly anything-goes regulatory posture.

Against this backdrop—and especially considering the Trump administration’s growing anxiety over Claude Mythos—there was modest (perhaps quixotic) hope that the meeting would result in some sort of precedent-defining alignment on cybersecurity and A.I. safety. Such an agreement was at least within the “spectrum of possibilities,” according to Scott Singer, a fellow at the Carnegie Endowment for International Peace. The business of A.I. was also expected to be a major point of discussion, given Trump’s entourage of a dozen executives including Meta’s Dina Powell McCormick, Micron’s Sanjay Mehrotra, Coherent’s Jim Anderson, Nvidia’s Jensen Huang (a last-minute addition who carpooled in from Alaska), and, of course, Elon Musk.

A MESSAGE FROM ANTHROPIC

Claude, the AI without ads

A space to think. Anthropic keeps conversations with Claude ad-free: no sponsored links, no advertisers shaping answers, no paid product placements you didn't ask for. When you bring your hardest problem to an AI, you shouldn't have to wonder who it's working for.

Learn more

But two days and 30-ish hours of flight time later, the idea of a future deal seemed like an afterthought. The post-summit fact sheet published by the White House mentioned U.S. planes, agriculture, beef, and Iran—but not A.I. On Air Force One, Trump told reporters, “We talked about possibly working together for guardrails.” When asked to specify which guardrails, he said, “The standard guardrails that we talk about all the time.” The president then noted with characteristic bravado: “A.I. is fantastic. … We probably will—we’re going to work together.”

Meanwhile, in an appearance on CNBC during the talks, Treasury Secretary Scott Bessent offered a slightly clearer impression of the negotiations. “The two A.I. superpowers are going to start talking,” he said. “We’re going to set up a protocol in terms of, how do we go forward with best practices for A.I. to make sure nonstate actors don’t get a hold of these models?” Singer found the lack of tangible results “relatively unsurprising” given the requisite theater embedded in this type of diplomacy. But, like many people I spoke with, he also believed that much more “could have been done that would have been in the national interests of both countries.”

The status quo outcome may have been inevitable. After all, the American and Chinese governments have been effectively micro-negotiating over A.I. soft power for years in plain sight. In 2022, the Biden administration imposed export limits on semiconductor chips and the equipment required to manufacture them. Those restraints were tightened over the following two years, and Trump added more in 2025—all of which hit China’s semiconductor industry and triggered a countrywide effort to produce chips independently. Trump relaxed those restrictions only last July, following an intense lobbying campaign from Huang and AMD’s Lisa Su. And while the U.S. has since cleared 10 Chinese companies to acquire Nvidia’s more advanced H200 chips, China has yet to approve the potential purchases. Trump’s decision to display the brain power of the American innovation industry served as merely the latest flex in this game of geopolitical tit-for-tat.

There’s also the fact that the White House itself is still refining its position on the matter. Kyle Chan, a Brookings research fellow and China expert, acknowledged the challenges inherent in a negotiation that mixes economics and national security. Ever since industry booster David Sacks left the administration, he said, “there’s no unified voice on the U.S. side.” Chris McGuire, a senior fellow at the Council on Foreign Relations, echoed the sentiment. “It seems like the administration isn’t totally decided on what its own stance or policy is,” he told me. “It’s not even clear who the administration’s lead on A.I. policy is.”

Mythos, Anthropic’s recent model that the company deemed too dangerous for public release, appears to be the source of this newfound equivocation. “I think the Trump administration didn’t quite understand the urgency of it until Mythos,” McGuire said, adding that the unreleased model “changed the whole conversation within the administration.”

During his first year in office, Trump’s approach to A.I. policy was “premised on this assumption that A.I. was going to be important down the road, but wasn’t that important right now,” McGuire continued, calling this “the only reason you would advocate for complete deregulation and sharing the technology with other countries, with China, etcetera.” But after the release of Mythos, he said, “it’s very clear that a lot of senior people in government are very convinced that this is an urgent issue, and we need to have a policy framework structured accordingly. But they’re starting basically from scratch.”

We’ve seen evidence of the administration’s scramble in recent weeks, particularly through its advocacy for pre-deployment model safety testing. Still, not everyone in Trump’s orbit is on the same page, and Trump himself “seems to be unclear where he stands on the issue,” Chan noted. Meanwhile, political factions beyond the White House, including prominent Republican groups and a wide array of lawmakers, are also pointing to Mythos as a justification for federal intervention.

Meaningful conversations with China will have to take place eventually, and when they do, Chan told me, they will be heavily influenced by “the conversation in the U.S. about A.I. safety.” He said he wouldn’t be surprised to see some near-term progress on those talks, mainly because the Chinese are “sort of interested in A.I. as an area of potential cooperation,” and, interestingly, don’t really view it as a “divisive issue.”

But many of the researchers I’ve spoken with are asking what an overarching A.I. deal would look like—and what it means that we seem so far away from achieving one. As McGuire pointed out, if the motivation were as simple as a shared interest in mitigating existential risk, “we would have seen something, so the fact that we haven’t means this is complicated.” He added that China is likely to view any U.S. proposal with suspicion and would be unwilling to make too many concessions, given that its “top priority is to catch up to the United States.” Basically, there are stumbling blocks everywhere. “These are hard issues, and they’re moving in real time. They’re very technically complicated,” McGuire told me. “I think the administration really needs to appoint someone who’s their lead on A.I.”

For Singer, though, the idea of some grand agreement is immaterial. “At the end of the day, the most important work is going to be what the U.S. and China do, themselves, to make the models safer,” he said. “It’s a much harder question to figure out how to share information across borders, but simply having governments that understand what’s happening inside their own countries is going to be really important.”

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St., New York, NY 10006