Anthropic backs open weights—and mandatory testing for powerful models

Why it matters

Anthropic’s broader package includes tighter controls on powerful chips reaching authoritarian governments, action against industrial-scale model distillation, and mandatory safety testing for sufficiently capable models regardless of whether their weights are open or closed. These are the company’s policy positions, not evidence that the proposed controls will be easy to implement or globally accepted.

The decisive details are still unresolved: who defines the capability threshold, which tests predict real misuse, who evaluates the models, how results can be challenged, and how requirements apply across borders. Poorly designed rules could entrench incumbents; weak rules could leave irreversible releases unexamined. The policy has to preserve useful openness while treating genuinely dangerous capability as a measurable release condition.