Anthropic's Dario Amodei: I'm Not Trying to Ban Open-Weight AI, But China Terrifies Me

Dario Amodei wants to set the record straight. The Anthropic CEO published a blog post Monday pushing back against industry chatter that his company is lobbying for broad bans on open-weight AI models. “Anthropic has never advocated for a ban on open-weights models,” he wrote, emphasis his own. The clarification came days after Nvidia’s Jensen Huang shared a joint industry letter urging policymakers to avoid “premature restrictions” on open-weight systems — a letter Anthropic notably did not sign.

But don’t mistake Amodei’s stance on open-weight for a relaxed view on risk. He draws a sharp distinction between harmless open models — which he calls “a public good” — and the existential threats he believes authoritarian governments pose. His real fear isn’t startups tinkering with Llama. It’s the Chinese Communist Party achieving “permanent military superiority” through AI or using it to repress its own citizens. He’s also deeply worried about biological attacks, an area where he argues open-weight models are uniquely dangerous because guardrails can’t be retroactively applied and releases can’t be unwound.

The solution, in Amodei’s view, isn’t a ban. It’s a layered defense: keep choking off China’s access to advanced chips, formally crack down on IP theft via distillation, and build a global safety testing body. On that last point, he struck a surprisingly diplomatic tone. He says he’s “heartened” by recent Trump administration moves and believes even China might cooperate on preventing AI-enabled bioweapons — because it’s in Beijing’s interest too. Skeptics will note that getting the CCP to submit to international testing is a monumental ask. But Amodei seems to think limited, pragmatic cooperation is less fantasy than it sounds.