Beijing and Trump Both Reject the AI Slowdown

The fastest part of Dario Amodei's slowdown proposal was the agreement. Anthropic's chief executive posted "We Must Pace the Frontier" on Saturday; Elon Musk and Sam Altman endorsed it the same day, and Satya Nadella followed on Sunday. The slowest part is arriving now. By Monday, the two governments whose cooperation the plan actually requires had both said no — one in an editorial, the other from a golf course.

China went first. The state-run Global Times published a piece on Sunday reading the essay as containment dressed as safety: a "Cold War playbook" for the AI sector with a "hidden agenda," which "politicizes technical issues" and is designed, in the paper's account, to "exclude China from the global AI governance system." The editorial itemised what it considered the real programme — bans on selling advanced chips and equipment to China, a crackdown on model distillation, and hardened security against weight theft — and called the combination hypocritical and short-sighted. A China Daily editorial on Monday sharpened the commercial charge, arguing the push is self-serving and that with chip and equipment curbs having failed to slow Chinese labs, the restrictions are simply migrating to software.

The complaint is not invented. Amodei's essay does not only ask labs to slow down; it also argues that a Chinese lead in AI would be dangerous for the United States and the world, and calls for tighter limits on selling key technology to China. That is the part Beijing read first. Foreign ministry spokesperson Guo Jiakun gave the official line on Monday: "Fearmongering, confrontation and malicious competition will only disrupt the process of global AI governance and serve no one's interests." State security minister Chen Yixin, who spent the same week describing AI as a new track for strategic rivalry, argued China should tighten its own AI security controls while continuing to push capability — which is, stripped of the framing, close to the opposite of pacing.

Washington's answer was blunter and came from the other direction. Speaking at his golf resort in Ireland, President Trump dismissed the warnings outright — "We're leading China in AI, we're the most sophisticated country in the world" — and complained about negative forces raising alarms without justification. His summary, delivered in the four words that have become the administration's entire AI doctrine, was that whoever wins AI wins. He returned to it on Truth Social the next day: the only guardrail AI needs, he wrote, is a strong and smart president.

The interesting response was David Sacks's. Rather than dispute the premise, the administration's AI policy lead conceded it and removed the government from the equation: if the unreleased models are frightening enough to justify pacing, he wrote, go ahead and pace — but stop pretending anyone else's permission is required, and stop presenting the motive as purely altruistic when product liability is sitting right there. Speaker Mike Johnson closed off the legislative route with the same reflex the labs have heard for three years: an emergency session to regulate AI, he said, would lose the race to China. Between them, the executive and the House have declined both halves of the request — no mandate, and no blessing.

Which leaves the labs holding a commitment they made on their own. Reporting in the Washington Post indicates Anthropic, OpenAI and Google have been meeting on a working-group basis since at least July on a shared industry standards body — the second of Amodei's three steps, the one that lives entirely inside the private sector. The third step, coordination between democratic and authoritarian governments, is the one this week just priced. As BitsMinds noted when the essay landed, the embedded-evaluator pledge is the falsifiable part, and it does not need a statute or a summit to happen.

Read narrowly, nothing has changed: no company has withdrawn its endorsement, and the badge-and-desk commitment Anthropic made is unilateral by design. Read at the scale Amodei was writing at, the week has been clarifying. He argued that pacing only works if defection is visible and eventually governed. Inside 72 hours, one government called the proposal an instrument of containment and the other called it unnecessary, and the labs' answer to both is that they will do it anyway. That is a smaller plan than the essay described, and it is the only version currently on the table.