Anthropic pitches its own AI safety bill to US lawmakers

Anthropic has pitched a draft bill to Capitol Hill lawmakers to establish federal regulation for frontier artificial intelligence models. Based on proposals from Anthropic's Advanced AI Framework, the draft bill allows government intervention only if a model poses an imminent risk of over $1 billion in damages or more than 50 casualties from cyberattacks, biological weapons, or lost control. Under the draft legislation, independent evaluators would report risks to the Commerce Department, which could petition a D.C. federal judge to halt dangerous deployments.

Anthropic proposes draft AI bill

• ▪Anthropic representatives quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier artificial intelligence safety and security
• ▪An Anthropic spokesperson stated Anthropic's draft AI safety bill represents proposals outlined in Anthropic's Advanced AI Framework released earlier in 2026
• ▪Anthropic's draft AI safety bill requires frontier AI developers to test models, be transparent about findings, and undergo independent evaluations

Thresholds for government intervention

• ▪Under Anthropic's draft bill, the US government could block an AI model's release or use only if it poses a foreseeable, imminent risk of killing or seriously injuring more than 50 people or causing over $1 billion in damages
• ▪The risk threshold under Anthropic's draft AI safety bill must stem from a cyberattack, a model evading developer controls or going beyond human control, or the creation or release of a biological weapon

Role of independent model evaluators

• ▪AI labs could pay independent evaluators under Anthropic's draft AI safety bill, provided the payment is not tied to the outcome of the evaluation
• ▪Under Anthropic's draft AI safety bill, independent evaluators registered with the Department of Commerce would periodically evaluate AI labs and their frontier models
• ▪If independent evaluators under Anthropic's draft AI safety bill identify an imminent enumerated risk in a company's frontier model, they must report it to the Commerce Department within 72 hours of completing their assessment

Enforcement mechanisms and judicial review

• ▪Anthropic's draft AI safety bill language explicitly prohibits any US government department, agency, or officer other than the Commerce Department and federal court from suspending model development, deployment, or internal use
• ▪Anthropic's draft AI safety bill limits legal suspension authority to specific high-threshold risks evaluated by registered third parties and adjudicated by a D.C. federal court judge
• ▪Under Anthropic's draft AI safety bill, the Commerce Department could request a federal judge to halt an AI model after determining it poses a qualified risk