Anthropic has circulated a draft AI safety bill to Congress, outlining scenarios in which an AI model's release could be blocked by a federal judge. The legislation, first reported by Politico, defines "catastrophic risk" as situations where over 50 people face imminent death or more than $1 billion in property damage could result from a biological weapon, a model escaping developer controls, or a cyberattack.
The thresholds mirror California's Senate Bill 53, recently signed into law, which used identical figures to define catastrophic AI risk. Anthropic's proposal goes further by proposing registered "independent evaluators" overseen by the Commerce Department to assess models before release — evaluators who could be compensated by the AI labs themselves.
Notably, the bill would prevent federal agencies from blocking a model's release without a judge's permission, a provision that critics argue could slow regulatory response. The proposal arrives as lawmakers scramble to establish guardrails for frontier models, with multiple competing bills in both chambers.
Anthropic has positioned itself as the safety-first AI lab, and this legislative push extends that brand into the policy arena. Whether Congress adopts the framework — or any of the competing proposals — remains uncertain, but the specific thresholds and evaluator structure give lawmakers a concrete template to work from.
