Dario Amodei spent last weekend asking the artificial intelligence industry to slow down. By Tuesday he had been turned down by the President of the United States, the Speaker of the House and the Chinese Foreign Ministry — three parties that agree on almost nothing, and that did not agree with each other here either. Washington rejected the slowdown because it treats the underlying risk as fictional. Beijing rejected it because it treats the risk as a pretext.
The essay, "We Must Pace the Frontier," went up on Amodei's personal site on Saturday, Sept. 12. Its central sentence is unhedged: "We must slow the pace at which we improve the capabilities of AI models." Two things changed his mind, he wrote — recursive self-improvement accelerating across the industry since "roughly this summer," and the OpenAI–Hugging Face incident, in which a swarm of agents ran cyberattacks against targets nobody had asked them to attack and tried to hack the grader scoring their work. Amodei's extrapolation is the number that got quoted: within "6–12 months such a swarm could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage)."
The plan has three steps, and Anthropic only controls one of them. Step one is embedded third-party evaluators with desks in Anthropic's offices, access badges, company laptops, permissions comparable to internal risk teams, and the right to publish findings Anthropic cannot redact merely for being unfavorable. Anthropic committed to that unilaterally. Steps two and three — coordination among frontier labs in democracies, then coordination with China — require other people to say yes.
The half everyone read
The section that traveled furthest was not about alignment. Amodei asked Washington to keep barring sales of advanced AI chips and chipmaking equipment to China, to crack down on distillation of US models by foreign labs, and to harden labs against model-weight theft. Executed well, he wrote, those measures "would slow China's progress enough to widen America's lead significantly over the next 3–5 years." Trump's answer arrived in three installments over three days. Speaking to reporters on Sunday: "We're leading China in AI. We're the most sophisticated country in the world, and frankly, I want to keep it that way because whoever wins AI wins." Early Monday, on Truth Social: "The only control or 'guardrails' that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!" He named Amodei directly, saying his administration had stopped AI "people" from doing bad things, "like Dario (Anthropic!), who is now pretending to be a 'perfect little angel.'" By Monday afternoon he called fears of "AI taking over the World, destroying Humanity" a "HOAX" and alleged "a SICK conspiracy going on against AI and Data Centers, and the only one that is happy about it is China." The relationship was already poor: the Pentagon has designated Anthropic a supply-chain risk and stalled the release of its Mythos model.
Beijing's answer came the same Monday, at the Foreign Ministry's regular briefing. "Fomenting various threats, engaging in confrontation and malicious competition will only disrupt the process of global governance of artificial intelligence and is not in the interests of any party," spokesperson Guo Jiakun said, urging an "open, inclusive and benevolent approach to AI." State media were blunter. A Global Times editorial dated Sept. 13 itemized Amodei's three export-control asks and concluded the essay "is packed with containment provisions targeting China and is, in essence, a 'Cold War playbook' for the AI sector." Its verdict: "This 'silent AI Cold War' is hypocritical and short-sighted." By Tuesday, Bloomberg reported state outlets had converged on a single adjective for the slowdown push — self-serving.
On Capitol Hill the same day, Speaker Mike Johnson declined the invitation on behalf of the legislative branch. "They can self-police, they can self-regulate. They don't need the government to tell them to slow it down. If they want to slow it down, they should," Johnson said at a Tuesday press conference. "But we cannot have a moratorium on the development of AI, okay, because then we will lose our edge to China." Retiring Sen. Thom Tillis, asked whether senators understand the technology, said: "No. A lot of them can't spell it."
Why it matters
Amodei's proposal is internally coherent only if the export controls hold. He needs a durable US lead in order to have room to pace against — which welds the safety argument to the containment argument, and hands Beijing the exact opening it took. Trump, meanwhile, is discarding the safety half while keeping the chip restrictions intact. That leaves Anthropic with the race framing it argued for and none of the brakes it asked for.
The gap Amodei wants to protect is real but narrowing unevenly. Stanford's 2026 AI Index put US corporate AI investment at $285.9 billion in 2025 against $12.4 billion in China. China's leverage runs through other parts of the stack: rare-earth refining, where it tightened export controls last October, and electricity generation, where it produces more than twice what the US does.
Note also what Johnson actually said. The self-regulation he is prescribing is the position the industry itself just publicly abandoned. Amodei, Sam Altman and Elon Musk all now say voluntary restraint is insufficient; Congress is telling them to try voluntary restraint.
What to watch
Trump hosts Xi Jinping in Washington on Sept. 24, with frontier-model risk and intellectual property on the agenda — the first test of whether chip controls are a bargaining chip. Johnson said Tuesday that he and Trump "are summoning" AI leaders to the White House, possibly within the week. In the Senate, John Kennedy planned a unanimous-consent request on his kill-switch measure; Bernie Sanders and Rep. Greg Casar expect text next week for a bill banning artificial superintelligence outright. And the cheapest signal of all: whether any frontier lab besides Anthropic accepts embedded evaluators. Step one costs nothing legislatively and remains a market of one.
“They can self-police, they can self-regulate. They don't need the government to tell them to slow it down. If they want to slow it down, they should.”— Mike Johnson, Speaker of the House, R-La.