Anthropic pitches a federal AI safety bill with a 50-death threshold for blocking model release

TechCrunch reported on Oct 20 that Anthropic is introducing a new AI safety bill to members of the US Congress and hopes lawmakers will consider adopting it. Per TechCrunch's write-up of a Politico report, the draft would rely on independent evaluators registered with the US Commerce Department to judge whether a model poses a foreseeable, imminent catastrophic risk. The triggers include more than 50 people at risk of dying or more than $1 billion in property damage, but hitting those thresholds alone would not automatically trigger a block; the risk would also have to stem from creating or releasing a biological weapon, a model evading developer controls, or a cyberattack against the system.

The enforcement chain runs through the evaluator: if one determines an imminent catastrophic risk exists, it must notify Commerce within 72 hours. Commerce would not be able to block a model on its own and would have to ask a judge in Washington DC to halt a release. The draft reportedly bars any federal agency from blocking model distribution without a judge's permission. Labs may pay evaluators, but the draft reportedly requires that payment not be contingent on the evaluation results. TechCrunch notes it remains unclear how the mechanism would prevent payment from influencing evaluations.

Anthropic said the bill reflects proposals from its Advanced AI Framework and that the company has been advocating them with federal and state lawmakers since June. The report also notes the draft uses thresholds similar to California's Senate Bill 53, which became law and likewise treats more than 50 people at risk of dying or more than $1 billion in property damage as catastrophic risk. This remains a draft rather than enacted federal law, and the page does not disclose a formal bill name or number.