Our Position on Open-Weights Models

Anthropic CEO Dario Amodei clarifies that Anthropic has never advocated for a ban on open-weights AI models, pushing back against accusations tied to US government discussions about restricting Chinese open-weights models. Amodei identifies two core national security threats: authoritarian governments surpassing US AI capabilities, and misuse of powerful models for cyberattacks or bioweapons. He argues blanket bans are ineffective and instead endorses three targeted measures: enforcing chip export controls to China, cracking down on industrial-scale model distillation operations, and requiring mandatory safety testing for all sufficiently capable models regardless of origin or open/closed status. Amodei partially endorses an open-weights industry letter but disputes its assertion that broad capability access inherently advantages defenders, citing bioweapons asymmetry as a key counterexample requiring empirical testing rather than assumption.