Anthropic chief executive Dario Amodei published a blog outlining a series of steps intended to slow the pace of artificial-intelligence development in China. His plan called for halting sales of high-performance AI chips and related manufacturing tools, tightening enforcement against chip smuggling, and limiting remote connections to overseas data centers. Amodei argued that such measures would create breathing room for the global community to manage AI risks more effectively.
The proposal attracted unexpected endorsement from two prominent U.S. AI leaders. OpenAI chief Sam Altman and xAI founder Elon Musk publicly expressed support for the suggested restrictions. By contrast, Meta chief executive Mark Zuckerberg and former President Donald Trump dismissed the idea, urging instead that AI progress continue without impediment. Their divergent positions highlight a split within the American technology sector over how to handle emerging capabilities.
Chinese authorities responded with a unified rebuke. State-run Global Times published an editorial accusing the plan of being a Cold-War style attempt to preserve U.S. technological dominance. Foreign Ministry spokesperson Guo Jiakun labeled the call as fear-mongering and competitive aggression that could undermine worldwide AI governance. The backlash comes ahead of a scheduled meeting between President Trump and President Xi, where Washington is expected to raise shared AI risks with Beijing.
On September 14, China’s national cybersecurity standards organization issued the third edition of its Safety Governance Framework. The document warned that AI systems might achieve recursive self-improvement at speeds beyond human anticipation, echoing concerns raised earlier by a former Anthropic researcher. The framework seeks to codify safeguards against uncontrolled optimization, reflecting Beijing’s growing emphasis on formal risk-management structures for advanced models.
Chinese policymakers have outlined a broad spectrum of AI-related hazards. Concerns range from the technology’s impact on the information environment and the emergence of romantic relationships with chatbots to potential disruptions in the labor market and the use of AI for cyber-crime. State Security Minister Chen Yixin recently warned that cutting-edge U.S. models, including Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber, could threaten China’s critical information infrastructure, prompting calls for tighter oversight and greater self-reliance.
Individual engineers in China have begun voicing personal perspectives. Tao He, head of research at MiniMax Code, responded publicly to the debate, while DeepSeek developer Liu Shengyu posted an essay on WeChat describing a dystopian future where a handful of firms dominate the most powerful AI. Liu wrote, “I don’t trust Anthropic or OpenAI to do that. In particular, I do not want Anthropic to control the world’s most advanced AI or AGI,” reflecting apprehension among practitioners.
Huawei’s rotating chairman Eric Xu offered a contrasting view, suggesting that Chinese AI developers may need to accelerate rather than decelerate. Xu argued that leading U.S. labs are likely encountering risks that domestic teams have not yet perceived, and that Chinese model providers should advance their capabilities to experience comparable challenges. His comments underline a sentiment within parts of China’s tech community that catching up is essential before meaningful risk mitigation can occur.