SourceOur position on open-weights models \ Anthropic
This signals that a leading AI firm prefers targeted technical restrictions over broad bans, influencing how governments might regulate open-source AI to manage national security risks.
- Dario Amodei stated Anthropic has never advocated for a ban on open-weights models.
- Proposed three key measures: stopping chip smuggling to China, countering industrial distillation, and establishing mandatory safety testing for capable models.
- Highlighted research on modular pre-training to enhance open-weights model safety.