Chinese officials dismiss Amodei's proposed AI slowdown as fear-mongering

Somebody on your team owns the one line in next year's plan that says how capable the models will be. That line sets hiring, evaluation budgets, and what you are willing to promise a customer about accuracy. Amodei's blog post is an offer to make that line predictable: slow the most powerful models and buy time to make them safer ^[1]. The pause, he wrote, would "require cooperation with China, the autocratic country with by far the most advanced AI capabilities" ^[2].

The pitch is a pause. What Amodei asked for is an end to sales of powerful AI chips and semiconductor manufacturing equipment to China, a crackdown on chip smuggling, and restrictions on remote access to data centres outside China. Those measures, he said, would "give us the breathing room we need in order to pace effectively" ^[3]. Chinese officials, state media and engineers pushed back on the proposal as another move to preserve the U.S. technological lead ^[5]. The day after the post, Global Times ran an editorial titled "Targeting China's AI: U.S. 'tech right' unfolds Cold War playbook" ^[8].

Three days after Guo Jiakun's briefing, China's national cybersecurity standards body published the third edition of its Safety Governance Framework. It warned that AI could autonomously learn, optimize itself and undergo "recursive self-improvement" at a speed or in a direction that exceeds human expectations and control ^[10][17]. "I do not think China and the U.S. see AI risks in fundamentally different ways," Zilan Qian, a London-based research scholar at GovAI, told Rest of World. "But their socioeconomic conditions and political institutions make some risks more visible and some voices more prominent in each country" ^[11]. Qian said Chinese policymakers worry about the information ecosystem, people forming romantic relationships with chatbots, labour effects and cybercrime, and that humans losing control of AI gets less discussion than it does at U.S. frontier labs ^[16][12].

For anyone pricing a pause into a plan, the useful read is Alvin Wang Graylin's. Graylin, a senior fellow at the Asia Society Policy Institute's Center for China Analysis, told Rest of World he saw no reason why China would cooperate with Amodei's proposal and be held down more than it already is ^[6]. He added that this does not mean China would never agree to pace AI development; it means China is not agreeing to abide by Amodei's terms as they stand ^[7].

At home, the proposal split the field too. Sam Altman and Elon Musk backed it, and Mark Zuckerberg and Donald Trump rejected it in favour of rapid, unfettered growth ^[4]. Two for, two against among the four U.S. figures in the account ^[18]. Days before the essay, State Security Minister Chen Yixin warned that Anthropic's Mythos and OpenAI's GPT-5.5-Cyber could pose serious risks to China's critical information infrastructure, and called for tighter government oversight and greater technological self-reliance ^[13]. "Modern warfare has entered an era increasingly shaped by artificial intelligence," Chen wrote in the state-run magazine China Cyberspace on September 13 ^[14].

What this comes down to is ownership. Every capability assumption in a plan needs somebody's consent, and the last thing that party said in public is the best guide to whether the consent is coming. The assumptions a team controls are the ones it can plan on. The ones that need a counterparty whose foreign ministry has called the idea vicious competition are worth nothing, in my view, and Graylin's caveat leaves one narrower opening: a pacing arrangement whose terms Beijing helps write.