Analysis
Anthropic has been circulating a draft AI safety bill to lawmakers, first reported by Politico and confirmed to TechCrunch by an Anthropic spokesperson. The company hopes Congress will take it up. The bill would allow a model's release to be blocked when there is a "foreseeable, imminent risk" -- defined as a threat of more than 50 deaths or over $1 billion in property damage -- arising from a biological weapon's creation or release, a model escaping its developer's controls, or a cyberattack.
How Would The Bill Actually Block A Model?
Independent evaluators registered with the U.S. Commerce Department would assess models before release. Labs could pay those evaluators directly, as long as compensation doesn't depend on the outcome of the evaluation. If an evaluator finds an imminent catastrophic risk, it would have 72 hours to notify Commerce, which would then ask a federal judge in Washington, D.C. to halt the release. The bill would simultaneously bar federal agencies from unilaterally halting distribution on their own authority, routing that power through the courts instead.
“If an evaluator finds an imminent catastrophic risk, it would have 72 hours to notify Commerce, which would then ask a federal judge in Washington, D.C.”
The proposal reflects ideas Anthropic has promoted to federal and state lawmakers since June through its Advanced AI Framework, and its risk thresholds closely track California's SB 53, a law enacted earlier this year using similar death and damage figures to define catastrophic risk. That overlap suggests Anthropic is angling to turn a state framework into a national standard before more states pass their own, divergent versions -- a patchwork labs have said repeatedly they want to avoid.
Reaction from the rest of the industry has been predictably split. DeepMind chairman Demis Hassabis has floated a different model: a self-governing industry body similar to FINRA, the financial industry's self-regulator, rather than a court-enforced block. Nvidia CEO Jensen Huang has argued the AI industry needs no new laws at all. More broadly, labs have generally resisted mandatory government oversight in favor of voluntary, embedded third-party evaluators -- making Anthropic's willingness to propose court-enforceable blocking authority notably more aggressive than its peers' public positions.
TechCrunch's reporting also raises a structural question the bill doesn't resolve: how evaluator payment arrangements, even outcome-independent ones, avoid influencing results when the lab being assessed is also the client paying the bill. That question will likely shape how Congress -- and Anthropic's rivals -- respond if the bill gains traction. Anthropic did not immediately respond to TechCrunch's request for further comment, and no lawmaker has yet introduced the draft as formal legislation.
