Anthropic Proposes AI Bill to Block Models That Risk 50+ Deaths — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic has drafted a proposed AI safety bill that would enable a judge to halt the release of an AI model if it poses a foreseeable, imminent risk of killing over 50 people or causing over $1 billion in property damage.
- Anthropic tied the catastrophic risk threshold to specific scenarios: the creation or release of a biological weapon, a model evading developer controls, or a major cyberattack.
- Independent evaluators registered with the U.S. Commerce Department would assess models for catastrophic risk, with labs allowed to pay them so long as compensation is not tied to the outcome of the evaluation.
- The Commerce Department would have authority to request a Washington, D.C., judge block a model’s release within 72 hours of receiving a risk alert, but no federal agency could unilaterally halt distribution.
- Anthropic confirmed the draft bill aligns with its Advanced AI Framework, which it has been advocating to federal and state lawmakers since June.
- Demis Hassabis, chairman of DeepMind, previously suggested a self-governing AI regulator modeled on FINRA, highlighting alternative industry-backed regulatory visions.
- Nvidia CEO Jensen Huang has argued the AI industry does not need new laws, contrasting with Anthropic’s push for formal legislative safeguards.
Why it matters: Anthropic’s proposal shifts the AI regulation debate from voluntary oversight to enforceable legal thresholds, giving courts—not agencies—the final say on blocking dangerous models. This creates a tangible benchmark for catastrophic risk that could shape future policy, even as rivals like Nvidia resist new rules.
Ask SkimNews
