Anthropic urges US law for tighter checks on AI labs
Technology
Anthropic wants the US government to step in if any frontier AI model becomes a real danger: think risking more than $1 billion in damage or, foreseeably, imminent risk of killing or seriously injuring more than 50 people.
Their draft bill suggests tighter checks on AI labs to stop things like cyberattacks or even the creation of biological weapons before they happen.
Commerce-registered evaluators trigger 72-hour alerts
Independent evaluators, registered with the Department of Commerce, would regularly review AI models.
If they spot a serious threat, they would have 72 hours to alert the Commerce Department.
Only this department could then ask a federal judge in D.C. to pause an AI system (no other agencies allowed).