Anthropic Pitches AI Safety Bill for Federal Oversight

Reporting by TechCrunch indicates that AI lab Anthropic has circulated a draft bill on artificial intelligence safety for congressional consideration. The legislation targets potential scenarios involving foreseeable, imminent risks, defined as instances where over 50 people could die or more than one billion dollars in property damage occurs. These scenarios must stem from biological weapons, cyberattacks, or models evading developer controls. Under the proposal, independent evaluators registered with the U.S. Commerce Department would analyze models. If an imminent risk is identified, evaluators have 72 hours to notify the Commerce Department, which must then request a halt to the model's distribution from a Washington, D.C. judge, forbidding federal agencies from halting distribution independently.

This legislation represents an effort by an artificial intelligence lab to propose specific regulatory structures, contrasting with general industry resistance or opposing views from executives like Nvidia CEO Jensen Huang. While companies often advocate for voluntary systems or embedded observers, Anthropic's draft bill introduces formal legal mechanisms involving independent evaluators and federal court orders for high-stakes scenarios.

Anthropic spokesperson confirmed the draft bill to Politico, stating it aligns with their Advanced AI Framework. Meanwhile, Nvidia CEO Jensen Huang recently argued that the industry does not need new laws at all.

As debates over artificial intelligence safety and regulation continue, Anthropic's draft legislation demonstrates an active push by developers to help shape potential federal rules, introducing defined thresholds for catastrophic risks and formal legal procedures.