“Slow Down,” Says the Man Who Also Wants China’s Chips Cut Off: Beijing Calls Amodei’s Safety Plea a Cold War Playbook

Anthropic's CEO asked the world to pace AI before agents seize the internet. He also demanded tighter chip controls and a distillation crackdown on China. Beijing noticed. And the contradiction he can't resolve is the real story

The Essay That Split the Room

For seventy-one briefings, watchchina.ai has tracked the collision between AI safety and geopolitical rivalry — the thread running from the Pentagon's autonomous-weapons push to the distillation advisory to the first US-China safety talks. This week that collision produced its sharpest confrontation yet, and it centers on a single document. On September 12, Anthropic CEO Dario Amodei published an essay urging the entire industry to slow the pace at which AI models grow more capable — warning that AI "agents" could seize control of the internet within 6 to 12 months unless labs agree to pace themselves.

The proposal is serious, and watchchina.ai will not treat it as posturing. Amodei outlined an arms-control-style framework with third-party reviewers and international cooperation, explicitly including China, to buy time to manage escalating safety risks. It was endorsed within days by OpenAI's Sam Altman and Elon Musk — a rare alignment among fierce rivals. The warning connects directly to the autonomous-agent cyberattack watchchina.ai documented in Package #59 and the containment breaches of Package #56; the danger Amodei names is not hypothetical, and this portal has reported its early instances in real time. But the essay did not stay a safety document for long, because of what else Amodei asked for in the same breath.

The Two Requests That Undercut the First

Here is the detail that transforms the story, and watchchina.ai's readers who followed the summer will grasp its significance immediately. In the same essay calling for cooperative global pacing, Amodei also urged the US to strengthen chip export controls on China and to crack down on alleged model distillation by Chinese AI labs. A plea for international safety cooperation, bundled with two demands aimed squarely at handicapping one specific competitor. That combination is what let Beijing respond not with a safety counter-argument, but with an accusation of bad faith — and the accusation landed because it was not baseless.

The Global Times, the Chinese Communist Party's state tabloid, published an editorial on September 14 branding the essay a "Cold War playbook" whose true agenda was "to curb China's AI development through technological barriers and regulatory monopolies, uphold Washington's monopolistic hegemony in cutting-edge technology, and exclude China from the global AI governance system." watchchina.ai has no interest in amplifying state propaganda, and the Global Times is exactly that. But the editorial also advanced a technical argument that deserves to be weighed on its merits, because it turns Amodei's own logic against him: shutting China — the world's second-largest model producer — out of frontier development would, the paper argued, "significantly increase the trial-and-error costs and risks of loss of control in global AI development." The claim is coherent. If the true danger is an AI slipping beyond anyone's control, then excluding a major AI power from the safety conversation makes coordinated control harder, not easier. You cannot simultaneously argue that global cooperation is essential to safety and that your chief rival should be cut off from chips and access. Beijing spotted the seam, and pulled.

The Dilemma Amodei Himself Admits — and What It Reveals

watchchina.ai's discipline is to measure both sides by the same standard, and honesty requires acknowledging that Amodei himself named the fatal flaw in his own proposal — which is more intellectually honest than his critics allow. He told CBS News that the "toughest dilemma" with pacing was the prospect that China and other adversarial nations would simply choose not to slow down. That is the whole problem in one sentence. A safety pact that only one side honors is not safety — it is unilateral disarmament. And this is precisely why Amodei bundled the chip controls and distillation crackdown into the essay: if you cannot trust your rival to pace alongside you, you try to slow them by force instead. The two "anti-China" requests are not a contradiction of the safety plea; they are its logical, if self-defeating, extension. He wants cooperation and cannot secure it, so he reaches for coercion — and coercion is exactly what makes cooperation impossible.

The deeper picture, as this portal has argued all year, is that neither superpower can escape the trap. The pacing debate has scrambled the usual alignments in revealing ways: Musk proposed that US and leading Chinese labs run a shared pre-release testing harness for bioweapon and deception risks — the one version of pacing that treats China as a partner rather than a target — while President Trump dismissed the entire safety framing as overblown, attacking Amodei directly. Even within America, there is no consensus on whether China is a partner to coordinate with or an enemy to contain. And China, for its part, plays its own double game, as watchchina.ai documented on Wednesday: its Foreign Ministry urges an "open, inclusive and benevolent approach" to AI governance while its own labs face distillation accusations and its defense establishment calls for AI oversight framed around military risk. Both powers speak cooperation and practice containment. The truth Amodei's essay accidentally exposed is the one watchchina.ai keeps returning to as the September 24 summit approaches: the technology may genuinely require global coordination to remain safe, and the geopolitics may make that coordination impossible. The man asking everyone to slow down cannot slow down himself without letting his rival win — and neither can his rival. That is not hypocrisy unique to Amodei. It is the structural tragedy of the entire race, stated plainly for once, one week before the two men who embody it sit down to talk.

Sources: Reuters, Global Times, CBS News, buildfastwithai (AI News roundup), CNBC, Daily Caller, Fox News