Dario Amodei sets out Anthropic's position on open-weight models

Dario Amodei published a statement of Anthropic’s position on open-weight models, after the company’s absence from an industry letter — backed by Nvidia, Microsoft, Meta and OpenAI — opposing restrictions on open-weight releases had been read by some as tacit support for a ban. He rejected that reading directly:

Anthropic has never advocated for a ban on open-weights models as a category. We should instead focus on keeping powerful chips out of authoritarian hands, stopping industrial-scale distillation, and requiring safety testing of all sufficiently capable models, open and closed.

— Dario Amodei, Anthropic company statement

The essay argued that openness itself was not the relevant axis of risk: an authoritarian government’s most dangerous model would likely be trained in secret for military or surveillance use regardless of what the same country’s public open-weight releases looked like. Amodei instead pointed to semiconductor export controls, restricting large-scale distillation of frontier models by competitors, and mandatory pre-release safety testing for capable models regardless of licence.

The statement was a defensive clarification more than a policy announcement — TechCrunch reported it as a response to specific criticism — but it set out, in more detail than before, the distinction between open weights as a distribution format and the underlying capability of a model as Anthropic’s stated basis for policy.