Anthropic pitches bill to block AI models only via judge

Anthropic's draft arrives after a year of debate over AI safety and follows the company's Advanced AI Framework, which a spokesperson said it has been advocating with federal and state lawmakers since June. The bill's 50-person and $1 billion thresholds echo Senate Bill 53, the California legislation recently passed into law that treats similar scenarios as catastrophic risk, suggesting the draft borrows from state-level work.

The mechanics differ from a simple agency ban: registered independent evaluators would flag imminent risk to the Commerce Department within 72 hours, and the department would then ask a D.C. judge to halt release, with federal agencies expressly barred from acting alone. Evaluators could be paid by the AI labs so long as payment is not contingent on results, though it is not clear how that would assure payments couldn't influence the evaluation; Demis Hassabis of DeepMind has separately floated a self-governing body modeled on FINRA.

The draft sits against a backdrop of industry resistance to mandatory oversight and voluntary embedded-evaluator schemes, and it goes slightly further than proposals such as Nvidia CEO Jensen Huang's recent argument that the AI industry needed no new laws at all.