Anthropic has been quietly pitching U.S. lawmakers on draft federal legislation that would let the government block the release of an AI model — but only in the most extreme circumstances, according to a report by Politico.
The draft proposal, which Politico says it has seen, would permit federal intervention only when a model poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously harming upwards of 50 people. The framework reflects Anthropic's attempt to shape the rules governing its own industry before others write them — an effort the company confirms it has been running since June.
What the Draft Bill Would Do
Under the proposal, the government could block an AI model's release or use only under a narrow set of catastrophic scenarios. The specified risks must relate to the creation or release of a biological weapon, an AI model evading its developer's controls, or a cyberattack.
The thresholds are deliberately high. Politico's reporting notes that while AI labs have recorded incidents in which their agents hacked websites or misbehaved in other ways, no AI incident to date has caused widespread deaths or more than $1 billion in damages. In other words, the bill defines the government's red line far above where current failures actually occur.
Critics have already seized on that design. Tom's Hardware characterized the proposal as one that would "effectively make the Commerce Department the only agency capable of stopping an AI model," while Mother Jones described the bill's threshold for government action as "alarming." Supporters counter that a high bar is precisely the point: it targets catastrophic, existential-class risks without subjecting routine model updates to regulatory review.
Independent Evaluations, Registered With Commerce
The bill's enforcement mechanism rests on third-party scrutiny. It would call for independent evaluations — registered with the Commerce Department — to periodically review AI labs and their frontier models.
If an evaluator concludes that a company's frontier models pose an "imminent enumerated risk," the evaluator must report that finding to the Commerce Department within 72 hours of completing the assessment, according to Politico. That reporting clock, and the registration requirement, would make the Commerce Department the federal government's central node for frontier AI oversight.
The approach echoes the direction of other pending legislation. The Artificial Intelligence Risk Management and Security Act of 2026, introduced by a trio of Senate Democrats, similarly contemplates Commerce-directed AI evaluations. A separate bill from Senator Maria Cantwell would task the National Institute of Standards and Technology with developing standards against catastrophic AI harms, while Senator Josh Hawley's AI Risk Evaluation Act would put the Department of Energy in charge of stress-testing the most powerful systems before deployment.
Anthropic's Defense: Sensible Policies, Advocated Since June
An Anthropic spokesperson told Politico that the draft "represents the proposals outlined in Anthropic's Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June," adding: "These are sensible policies we believe Congress should act on."
The company's position is consistent with its public posture. Anthropic has repeatedly argued that frontier AI risks are real but narrow — concentrated in bioweapons assistance, loss of control and offensive cyber operations — and that regulation should focus there rather than on the broad swath of everyday AI applications.
Writing its own bill is the logical next step. Rather than reacting to proposals from others, Anthropic hands lawmakers language calibrated to its own risk framework, with preemption of patchwork state rules an implicit benefit of federal legislation. Reuters and a string of other outlets amplified the Politico report within a day, a sign of how closely Washington and the AI industry are now watching each other's moves.
The Road Ahead in a Divided Washington
The draft faces steep political odds regardless of which party controls Congress. Politico's reporting notes that any AI safety bill is likely to face pushback from the White House, which remains opposed to mandatory AI regulation and favors a voluntary, standards-based approach.
Even so, the proposal matters as a marker. It quantifies — in dollars and casualty counts — where one of the world's leading AI labs believes the line between acceptable risk and government intervention should sit. It also creates a reference point for the competing bills now circulating on Capitol Hill, several of which propose far more aggressive oversight with lower thresholds.
Policy analysts will be watching three things as the debate develops: whether the 50-victim and $1 billion figures survive contact with congressional drafting; whether evaluators registered with Commerce get real authority or merely an advisory role; and whether federal preemption language absorbs or collides with state-level AI laws already on the books. Anthropic's bet is that a narrow, catastrophic-risk-only framework is the version of regulation the industry can actually live with — and that shaping that framework early beats having one imposed later.
Stay Ahead of AI
The AI landscape moves fast — follow AI Buzz Wire for breaking AI news, model releases and industry analysis.
Read more AI news →
