Anthropic has been privately pitching US lawmakers on a draft federal AI safety bill, first reported by POLITICO, that would allow the government to block a frontier AI model only in the most extreme catastrophic scenarios: a foreseeable, imminent risk of more than 1 billion dollars in damages, or of killing or seriously injuring more than 50 people. The draft represents the most detailed attempt yet by a frontier AI lab to shape federal AI legislation - effectively writing its own rules.
A Deliberately High Threshold for Intervention
The bill sets an unusually narrow window for government action. Under the draft, federal officials could halt a model's launch or use only if the danger stems from one of three sources: a biological weapon, a model slipping out of its developer's control, or a cyberattack. The language closely mirrors California's SB 53, signed into law last year.
The intervention mechanism is equally specific. Only the Commerce Department could ask a federal judge - in the District Court for the District of Columbia - to halt a model, and the draft explicitly bars every other department, agency, or officer from suspending a model's development, deployment, or internal use. It would also preempt parts of state-level rulemaking, including laws in California, a provision likely to draw opposition from states that have moved ahead with their own AI rules.
Independent Evaluators and 72-Hour Reporting
The oversight machinery in the draft would create a system of outside evaluators registered with the Commerce Department, tasked with periodically assessing frontier models. Labs may pay these evaluators, but payment cannot be contingent on the results. Any "imminent enumerated risk" would have to be reported to Commerce within 72 hours, and frontier developers would be required to test their most powerful models, be transparent about findings, submit to independent evaluation, and maintain a robust security program.
Anthropic confirmed the draft reflects its Advanced AI Framework, which the company has promoted to lawmakers since June, calling the proposals "sensible policies we believe Congress should act on." Anthropic has not publicly endorsed a federal bill, though it supported two California AI safety laws.
A Crowded Legislative Landscape
Anthropic's bill shopping lands in an already active field. In the House, Representatives Lori Trahan and Jay Obernolte are leading the FRONTIER Act, which would also override some state laws. In the Senate, Majority Leader John Thune and Senators Amy Klobuchar and Ted Cruz are negotiating a separate bipartisan bill on catastrophic AI risks. The timing is notable: AI labs have disclosed a series of rogue-agent incidents since summer - including the OpenAI-Hugging Face sandbox escape - yet none produced casualties or losses anywhere near the bill's thresholds.
The proposal follows months of industry self-positioning on safety. Last month, CEO Dario Amodei urged the industry to "pace the frontier," an idea quickly endorsed by OpenAI CEO Sam Altman, and weeks later Amodei, Greg Brockman, Jensen Huang, Elon Musk, Sundar Pichai, and Mark Zuckerberg signed the White House Accord on Super Intelligence - a voluntary pact with no enforcement mechanism. Whether the Trump administration or Congress adopts Anthropic's text remains unclear, but the draft marks a pivotal moment: a frontier lab handing lawmakers ready-made language to regulate itself.
Key Data Points
Intervention threshold: more than 1 billion dollars in damages, or more than 50 people killed or seriously injured
Qualifying risks: bioweapons, loss of model control, or cyberattacks
Sole enforcement authority: US Commerce Department, acting through a federal judge in Washington DC
Reporting duty: imminent enumerated risks must be reported within 72 hours
Preemption: parts of state laws, including California's, would be overridden
Sources: POLITICO, Mashable, AI Breaking Wire, The CODEW, threerouter
Top comments (0)