China Rejects Anthropic Slowdown Push as AI Safety Fight Turns Geopolitical
Chinese officials and engineers are rejecting Dario Amodei’s AI slowdown proposal, casting it as a U.S. effort to protect its lead while Beijing expands its own AI safety framework.
Chinese resistance to Anthropic CEO Dario Amodei’s proposed pause in frontier AI development is turning a safety debate into a contest over who sets the rules for the next stage of artificial intelligence, Rest of World reported.
The dispute began after Amodei argued that the industry should slow work on its most powerful models to create more time for safety measures.
His proposal would depend on Chinese cooperation, while also pressing for tighter limits on powerful AI chips, semiconductor manufacturing equipment, chip smuggling and remote access to data centres outside China.
That framing has turned a safety argument into a technology-sovereignty fight.
Chinese officials, state media and engineers have treated the proposal less as a neutral risk-control plan than as another attempt to preserve the United States’ lead in artificial intelligence.
Alvin Wang Graylin, a technology fellow at the Asia Society Policy Institute, said China had little reason to accept Amodei’s terms, though he left open the possibility that Beijing could support pacing AI development under a different framework.
The American response is divided as well.
Amodei drew support from OpenAI CEO Sam Altman and xAI founder Elon Musk, while Meta CEO Mark Zuckerberg and President Donald Trump favored faster, less restricted AI growth.
A September 24 Washington meeting between Trump and Chinese President Xi Jinping was expected to include discussion of shared AI risks.
Beijing’s formal response was quick.
Global Times framed Amodei’s approach as a Cold War-style attempt to target China’s AI sector.
Foreign Ministry spokesperson Guo Jiakun called the proposal “fear-mongering, confrontation and vicious competition,” arguing that it would damage global AI governance rather than improve it.
China is not dismissing AI risk altogether.
On September 14, the country’s cybersecurity standards authority issued an updated Safety Governance Framework that warned of systems capable of self-learning, self-optimization and recursive improvement beyond human expectations or control.
Zilan Qian, a London-based GovAI research scholar, told Rest of World that the U.S. and China do not see AI risks in fundamentally different ways, but that each country’s institutions and socioeconomic pressures make different risks more visible.
Those priorities include information disorder, chatbot relationships, labor effects, cybercrime and national-security exposure.
Chinese State Security Minister Chen Yixin warned in China Cyberspace that U.S. systems such as Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber could threaten China’s critical information infrastructure.
He called for stronger oversight, technological self-reliance and defenses against AI-enabled attacks and influence operations.
Engineers at Chinese AI companies have been more direct than most companies themselves.
Tao He, head of R&D at MiniMax Code, questioned why U.S. voices could move from urging a race toward AGI to demanding a slowdown once the terms changed.
DeepSeek engineer Liu Shengyu described a different dystopia: a world in which a handful of companies control frontier intelligence and the economic power attached to it.
His answer was to make powerful AI cheap and broadly available, not to let Anthropic or OpenAI control the frontier.
The divide leaves AI safety entangled with export controls, military risk and industrial catch-up.
Huawei rotating chairman Eric Xu suggested Chinese developers may first need to reach the level at which they can perceive the same frontier risks.
For Chinese engineers already working under chip restrictions, a slowdown can sound less like shared caution than another rule written by the side already ahead.
