Anthropic Urges Stricter US Controls on China's AI Development Before 2028
Key Takeaways
- ▸Anthropic is lobbying for stricter US export controls on advanced AI chips and American AI model restrictions, setting 2028 as the critical deadline
- ▸The company's framing—that authoritarians vs. democracies will 'set the rules'—overlooks China's demonstrated capacity for indigenous innovation, from DeepSeek R1 to domestic AI semiconductors
- ▸Critics note the irony in Anthropic's IP theft accusations while the AI industry widely trains models on third-party content without explicit permission
Summary
Anthropic has published a policy document calling on the US government and its allies to enforce tighter export controls on AI chips—particularly Nvidia GPUs—and restrict access to American AI models to slow China's AI capabilities. The company argues that democracies have only until 2028 to act before advanced AI systems emerge, warning that if authoritarian regimes gain the technological lead, they will shape global AI governance toward surveillance and repression rather than democratic values. Anthropic claims export controls have been 'incredibly successful' and attributes China's progress to talent, exploitation of regulatory loopholes, and distillation attacks that extract innovations from American companies. However, the policy push has drawn criticism for overlooking China's genuine domestic innovation—including the successful DeepSeek R1 model and advances in homegrown AI silicon—and for appearing motivated by Anthropic's own February accusations that Chinese rivals were copying its models.
- The Trump administration has shown limited interest in strict China export controls, recently clearing ~10 Chinese firms to purchase Nvidia's H200 chip
- European countries view both US and Chinese AI dominance as threats to digital sovereignty, complicating a unified Western approach
