Anthropic's AI Safety Bill Limits Federal Model Bans to Judge Approval
Anthropic proposes a bill requiring judges to approve AI model blocks by federal agencies.
Why it matters: The bill shapes evolving AI governance, balancing innovation with safety by limiting agency authority and imposing a high threshold for intervention.
- Anthropic's bill lets the Commerce Department block AI models only if risks include 50+ deaths or $1 billion in damages.
- Independent evaluators registered with Commerce assess AI safety and report threats.
- Commerce must seek a federal judge's approval within 72 hours to restrict AI model releases.
- Other federal agencies cannot block AI models without judicial approval under this bill.
Anthropic has introduced a legislative proposal that sets a high bar for federal intervention in AI deployment. Under the bill, issued through Anthropic's Advanced AI Framework, the U.S. Department of Commerce would be the sole federal agency authorized to halt the release of AI models.
Key to the proposal is a strict threshold for action: the Commerce Department may only block AI releases if there is a foreseeable, imminent risk involving either over $1 billion in damages or the death or serious injury of more than 50 people. The risks must connect to scenarios such as biological weapons, AI-enabled cyberattacks, or loss of AI model control by labs.
Independent evaluators registered with the Commerce Department — potentially compensated by AI labs — would assess these risks and report imminent threats. Once reported, the Department must present evidence to a federal judge within 72 hours to obtain approval before blocking AI model releases, effectively limiting executive agency power.
The bill explicitly prohibits other federal agencies from intervening independently without judicial approval. This judicial layer intends to balance AI innovation with public safety, but the high fatality and damage thresholds have sparked controversy over whether this sufficiently protects against AI risks.
This legislative effort comes amid rising concern over AI safety. The Federal Trade Commission is investigating AI companies, including Anthropic, following consumer safety incidents. In July 2026, Anthropic also resolved a $1.5 billion settlement for unauthorized use of copyrighted works in training its AI models.
The proposal demonstrates Anthropic's ongoing engagement with lawmakers since June 2026 to create clear AI safety standards focusing on judicial oversight over regulatory agency powers.
By the numbers:
- 50 deaths or serious injuries — minimum threshold to block AI model release
- $1 billion — damages threshold for federal intervention
- 72 hours — timeframe for judge review after threat report
Yes, but: The bill's high damage and fatality thresholds could delay interventions against lower-scale but still significant AI risks, raising debate about sufficiency of protections.
What's next: Anthropic is expected to continue advocacy for the Advanced AI Framework with lawmakers, while the FTC investigation into AI safety incidents progresses.