Anthropic Draft Bill Would Let Judge Halt AI Model Release

The draft would have Commerce ask a federal judge to halt a model's release when specified risks are identified.

The threshold is a foreseeable, imminent risk that could put more than 50 people at risk of death or cause more than $1 billion in property damage. Meeting that threshold would make government intervention possible; it would not automatically halt a model.

Covered risks include biological weapons, a model evading its developer's controls, or a cyberattack.

Independent evaluators registered with the Commerce Department would assess models. AI developers could pay the evaluators, provided payment did not depend on their findings. An evaluator identifying a qualifying risk would notify Commerce within 72 hours.

Commerce would then ask a federal judge in Washington, D.C., to halt the model's release. Federal agencies could not halt a model's release on their own under the proposed act.

California's SB 53 uses the same numerical thresholds in its definition of catastrophic risk. Anthropic has advocated its Advanced AI Framework since June.

Anthropic has separately given METR staff internal access for voluntary safety reviews. That arrangement does not amount to regulatory approval of a model release. The bill remains a proposal.