China pushes back on Anthropic CEO's plan to slow the AI race

For years Silicon Valley framed US-China AI competition as a race the US had to win. That has started to shift: in a blog post titled "We Must Pace the Frontier," published Saturday, Anthropic CEO Dario Amodei argued that AI has grown dangerous enough that "global pacing will require cooperation with China, the autocratic country with by far the most advanced AI capabilities." But he also wrote that a "key part" of that strategy should be making sure the US and other democracies keep "as large as possible" a lead over autocracies like China, including cracking down on chip smuggling and blocking unauthorized model distillation by Chinese firms.

China's government rejected the offer outright. At a Monday press conference, Guo Jiakun, spokesperson for China's Ministry of Foreign Affairs, called the proposal "fearmongering, confrontation, and vicious competition" that would only hamper global AI cooperation; the next day the ministry added that "China puts equal emphasis on development and security in terms of AI." China's minister of state security, Chen Yixin, had separately warned in an essay that AI could fuel disinformation, cyberattacks, espionage, social unrest and new military threats, spending far less time than US labs typically do on existential, doomsday-style scenarios.

Independent analysts read the same gap. Kevin Xu, founder of the hedge fund Interconnected Capital and a former Obama White House staffer, said Beijing sees "an Anthropic-led vision that would only subjugate and restrict China first to maximize leverage before talking," while "China would only talk if it's being treated from a position of equals." Alvin Wang Graylin, of the Asia Society Policy Institute and Stanford's Institute for Human-Centered AI, said Amodei's own essay ties the pacing plan to increasing the US lead over China, "which makes the proposal feel inauthentic to Chinese readers"; Amodei himself later admitted in a television interview that he has "no idea how to negotiate with China." Michelle Nie of the Center for a New American Security said Beijing doubts AGI is achievable at all and is skeptical AI poses extreme risk to humanity, worrying instead about threats to Communist Party control; she said Chinese policymakers read US safety rhetoric as instrumental, meant to inflate valuations ahead of IPOs or slow down Chinese competitors.

The skepticism runs into China's AI industry too. In a widely shared essay published Sunday, an unnamed DeepSeek researcher wrote that AI could push society toward one of two extremes: broad, cheap access that raises living standards, or a handful of companies controlling the most advanced systems and the resources behind them. "I do not trust Anthropic or OpenAI to do that," he wrote. "And I especially do not want Anthropic to control the most advanced artificial intelligence or AGI."

None of this rules out cooperation. Preliminary US-China talks on AI safety are reportedly already underway ahead of Chinese leader Xi Jinping's planned visit to Washington later this month to meet President Donald Trump, and Nie said both governments already share concerns about cybersecurity threats, misuse of powerful models and unpredictable AI behavior; a credible way for each side to verify the other's compliance, she said, could matter more than the rhetoric. But Kristy Loke, a research fellow at the AI safety nonprofit MATS Research, warned that Amodei's framing could backfire: "Dario is undermining one of the few available levers for reducing the very risks he professes to take seriously."

Key facts

Anthropic CEO Dario Amodei's blog post "We Must Pace the Frontier," published Saturday, ties a US-China AI slowdown to the US and its allies keeping "as large as possible" a lead over China, including cracking down on chip smuggling and unauthorized model distillation.
China's Ministry of Foreign Affairs spokesperson Guo Jiakun called the proposal "fearmongering, confrontation, and vicious competition" at a Monday press conference; the next day the ministry said China gives equal weight to AI development and security.
Analysts Kevin Xu, Alvin Wang Graylin and Michelle Nie say Beijing reads the plan as an attempt to subjugate and restrict China before any talks, and sees US safety rhetoric as instrumental, aimed at inflating valuations ahead of IPOs or slowing Chinese rivals.
An unnamed DeepSeek researcher, in a widely shared essay published Sunday, wrote he does not trust Anthropic or OpenAI to control the most advanced AI and argued instead for open, inexpensive access for everyone.
Preliminary US-China talks on AI safety are reportedly already underway ahead of Xi Jinping's planned Washington visit to meet Donald Trump, even as Amodei has separately admitted he has "no idea how to negotiate with China."

Why it matters

Silicon Valley has spent years framing US-China AI competition as a race the US must win outright. Amodei's post marks a shift toward framing it as a race both sides should slow down together, but he ties that cooperation to the US keeping the largest possible lead, including curbs on chip smuggling and model distillation. China's flat rejection, delivered by its Foreign Ministry and echoed by independent analysts and a DeepSeek researcher, shows how far apart the two sides are on what a safety deal would even require, just as officials prepare for Xi Jinping's planned Washington visit.

Who it affects

Anthropic and Dario Amodei, whose proposal sits at the center of the dispute; China's Ministry of Foreign Affairs and Ministry of State Security, which set out the government's public response; the Chinese AI industry, represented here by an unnamed DeepSeek researcher who distrusts Anthropic and OpenAI; and the broader US-China diplomatic track, where AI safety cooperation is reportedly one of the topics being discussed ahead of the Xi Jinping-Trump meeting.

How to use it

This is analysis and reaction, not a product or policy that launched, so there is nothing to install or buy. The practical use for anyone tracking AI policy is as a baseline reading of each side's likely negotiating posture: US frontier labs pairing safety language with a demand to keep the American lead, and Beijing pairing its own safety language with a demand for equal standing before any deal gets discussed.

How solid is it

Wired's account rests on named, directly quoted sources on both sides: Kevin Xu, Alvin Wang Graylin and Michelle Nie among the American analysts, plus China's Foreign Ministry's own press-conference language and a DeepSeek researcher's essay, so the Chinese skepticism is documented rather than assumed. The piece is also explicit about its limits: it gives no absolute dates for the blog post, the press conference or Xi Jinping's visit; it does not name the DeepSeek researcher; and the claim that China holds "by far the most advanced AI capabilities" is Amodei's own characterization, not an independently sourced comparison.

Risks and caveats

Kristy Loke of MATS Research warns that tying a slowdown to a guaranteed US lead can backfire, eroding the trust that any future negotiation would depend on and undermining the very leverage Amodei says he wants to use to reduce risk. The article also draws almost entirely on the skeptical Chinese and analyst side, with no extended on-record Anthropic response beyond the blog post itself and Amodei's television admission that he does not know how to negotiate with China, a framing worth weighing when reading the piece.

“The gulf lies between an Anthropic-led vision that would only subjugate and restrict China first to maximize leverage before talking. Meanwhile, China would only talk if it's being treated from a position of equals.”

— Kevin Xu, founder of Interconnected Capital