Anthropic pitches AI safety bill that may block a model if 50+ people could be killed
techcrunch ·
Anthropic has proposed a new draft bill to Congress aimed at regulating artificial intelligence safety by blocking the release of models that pose catastrophic risks. The legislation defines such risks as scenarios where over 50 people could die or more than one billion dollars in property damage could occur due to biological weapons, cyberattacks, or model control evasion. To assess these dangers, the bill suggests using independent evaluators registered with the U.S. Commerce Department, who would be paid by AI labs but not based on evaluation results. If an imminent risk is identified, evaluators must notify the Commerce Department within 72 hours, which would then seek a court order to halt distribution. This approach explicitly prevents federal agencies from unilaterally stopping model releases without judicial oversight. The metrics align with California’s recent Senate Bill 53, though critics question how payment structures might influence evaluator neutrality. While industry leaders like Nvidia have resisted new laws, Anthropic’s proposal represents a more structured regulatory framework compared to voluntary self-governance models.
techcrunch