Anthropic CEO Dario Amodei is again warning about risks from powerful open-weight AI models while saying he does not support a ban on them. In a new post, he says models without dangerous capabilities are a public good and that a US ban on Chinese models would not address the core risks.
Amodei’s concerns are focused on two scenarios: authoritarian states using stronger models for military or surveillance advantage, and powerful open models being misused for cyber or biological attacks. His argument is that released weights are difficult to control or recall, making safeguards harder than with closed models.
The position remains controversial because open models also support researchers, startups, and defenders. Amodei is instead calling for stricter chip export controls and action against industrial distillation. Critics argue that such restrictions may protect incumbents or accelerate China’s domestic chip efforts. The debate is becoming central to AI policy because it pits openness, competition, safety, and national security against one another.