Anthropic proposed a draft AI safety bill to Congress. It defines catastrophic risk as 50 or more deaths or over a billion dollars in damage. The risk must come from a biological weapon, a model evading controls, or a cyberattack. Independent evaluators registered with the Commerce Department would assess models. Evaluators would have 72 hours to notify the department of an imminent risk. The department would then ask a D.C. judge to halt the release. The bill forbids any federal agency from blocking a model without judicial approval. Evaluators could be paid by AI labs if payment is not tied to results. The proposal aligns with Anthropic’s Advanced AI Framework. It mirrors thresholds in California’s Senate Bill 53. Anthropic confirmed the draft bill to Politico.
What changed
Federal agencies could block a model release on their own discretion.
What it unlocks
A judge can halt a model release if catastrophic risk is proven.
- 50+ people at risk of death
- $1B+ in property damage
- 72 hours to notify Commerce Department
What you need to act on it
- a foreseeable, imminent risk of 50+ deaths or $1B+ damage
- risk stems from biological weapon, control evasion, or cyberattack
- independent evaluator assessment
- Commerce Department request to a D.C. judge
Sources