Anthropic circulates draft AI safety bill among US lawmakers

Anthropic is circulating draft legislation that would establish federal safety and security rules for the most powerful artificial intelligence models, the Claude developer confirmed Friday.

The proposal would allow the US Commerce Department to seek a federal court order blocking a model’s release or use when it presents a foreseeable and imminent risk of killing or seriously injuring more than 50 people or causing more than $1 billion in damage. Politico reported that qualifying threats would have to arise from a biological weapon, a cyberattack or a model evading its developer’s controls.

Frontier AI developers would be required to test their models, disclose findings publicly, maintain security programs and submit to independent evaluations. Evaluators registered with the Commerce Department would periodically assess companies and their models, and would have to report an “imminent enumerated risk” within 72 hours of completing an assessment. A judge in the US District Court for the District of Columbia would decide whether to suspend a model.

The draft would also pre-empt some state AI laws. Anthropic previously said states should retain authority to regulate in the absence of strong federal legislation.

The proposal follows Anthropic’s decision to embed a team from nonprofit evaluator METR in its offices with access to hardware and internal systems. METR can publish its findings without Anthropic’s approval, according to Ars Technica.

Anthropic co-founder Jack Clark separately urged international cooperation on AI rules at the IMF and World Bank annual meetings Saturday, warning policymakers that the window to address biological, cyber and other risks was narrowing.