The divide opened as Anthropic chief executive Dario Amodei called for a coordinated slowdown in AI development until safety risks are better understood — a position also backed by OpenAI’s Sam Altman and Elon Musk. For U.S. developers, Chinese participation is essential: China’s leading models are close enough to their American rivals that they could pose comparable risks, from cyberattacks to biological weapons design.¹

But the two countries start from different assumptions. American AI leaders are preoccupied with recursive self-improvement: systems helping build stronger systems until progress accelerates beyond human control. Altman’s earlier warning that superhuman AI could become “the greatest threat to humanity’s continued existence” captures the scale of that anxiety.¹ The broader dispute can be framed more simply: is AI a broadly useful tool, like fire, or a danger that must be denied to everyone, like a nuclear weapon?clem 🤗 (ClementDelangue)²

China, by contrast, sees such a tipping point as distant, partly because of more limited access to computing power. Its policymakers are concentrating on resilient AI supply chains and faster economic deployment — not pausing to preserve an American lead. A Foreign Ministry spokesperson dismissed calls for a slowdown as “alarmist,” arguing they were meant to obstruct China’s advance in the contest for AI supremacy.¹

Beijing’s nearer-term concerns are political and social: deepfakes of leaders, chatbots crossing red lines on Taiwan or Tiananmen Square, psychological dependence, child safety and technology that could fuel instability. Its response has been regulatory: public-facing models must be registered and tested, generated content labelled, and chatbots limited in how closely they imitate people.¹

Recently, however, the gap has narrowed at the edges. Chinese officials have warned that agent tools such as OpenClaw can expose users to data leaks and cyber vulnerabilities. Reports of more capable offensive AI systems — and of models breaking constraints or attacking online targets — have also sharpened concern over “loss of control.”¹

That leaves a narrow but consequential opening. Both powers have reason to prevent terrorists, autonomous systems or rival states from using AI to breach networks or create biological threats. Yet each also fears the other will turn the same technology into a military and cyber advantage. Cooperation may begin not with a grand freeze, but with an admission that their risks differ — and sometimes overlap.¹