Anthropic shops safety bill to Congress

Politico reports the AI lab has drafted legislation on AI safety it hopes Congress might adopt. The bill would block model releases in certain catastrophic scenarios.

The fifty-person threshold

The bill blocks release if there is foreseeable, imminent risk of over fifty people dying or over a billion dollars in property damage. The risk must stem from biological weapons, model evasion or cyberattack.

California law sets same metrics

The fifty-person and billion-dollar thresholds mirror California's Senate Bill fifty-three, which recently passed into law and similarly defined those scenarios as catastrophic risk.

Independent evaluators would assess risk

Anthropic's bill proposes independent evaluators registered with the U.S. Commerce Department assess model risk. AI labs could pay evaluators so long as payment is not contingent on evaluation results.

Commerce Department gets seventy-two hours

When evaluators find imminent catastrophic risk, they have seventy-two hours to contact the Commerce Department. The department must then request a judge halt the model's release.

Anthropic confirms draft to Politico

Anthropic did not respond to TechCrunch but confirmed the draft to Politico. A spokesperson said the bill represents proposals in Anthropic's Advanced AI Framework advocated since June.

“This bill represents the proposals outlined in Anthropic's Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June.”

Industry has resisted mandatory oversight

Despite warnings that their technology could threaten human civilization, AI labs have resisted mandatory government oversight. They prefer voluntary systems with embedded evaluators inside organizations.

Anthropic goes further than competitors

Anthropic's proposal goes slightly further than other industry regulatory proposals. Nvidia CEO Jensen Huang recently argued the AI industry did not need any new laws at all.