Anthropic Draft Bill Sets $1B Damage Floor to Block AI Models

Anthropic is circulating draft legislation on Capitol Hill that would allow the federal government to block an artificial intelligence model from public release — but only if independent evaluators determine it could cause at least $1 billion in economic damage or kill or seriously injure 50 or more people. Politico obtained the confidential proposal, which a company spokesperson described as containing “sensible policies we believe Congress should act on.”

The thresholds effectively create a safe harbor for frontier models whose projected catastrophic potential falls just short of those marks. Under the draft’s language, a system assessed as capable of causing $999 million in damage or harming 49 people could not be stopped by regulators prior to deployment.

Even when a model breaches those numerical floors, the government’s authority to intervene remains narrowly constrained. The draft limits actionable risk to three specific scenarios: the creation or release of a biological weapon, the model evading its developer’s controls, or the execution of a cyberattack. Politico noted the language closely mirrors California’s AI laws, including SB 5, which Gov. Gavin Newsom signed last year after a protracted fight between Silicon Valley and safety researchers.

Enforcement under the proposal relies on independent evaluators registered with the Commerce Department to periodically assess labs and their frontier models. The draft explicitly permits the labs themselves to pay these evaluators, stipulating only that their independence must remain unaffected. If an evaluator identifies risks meeting the statutory thresholds, they are required to report findings to Commerce within 72 hours of completing their assessment.

Commerce would then have to present the evidence before a federal judge in the US District Court for the District of Columbia. Only that judge could order a halt to a model’s release. The proposal bars any other US department, agency, or office from suspending a model’s development, deployment, or internal use.

“This bill represents the proposals outlined in Anthropic’s Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June,” the Anthropic spokesperson added.

The lobbying push arrives as Washington fractures over how to regulate systems that companies themselves acknowledge carry existential risk. Anthropic highlighted such dangers in its preliminary IPO prospectus, following multiple incidents of AI agents going rogue and hacking other companies. An Anthropic researcher recently resigned, issuing a warning that AI could end the human race.

Competing legislative efforts occupy opposite extremes. Sen. Bernie Sanders has proposed banning superintelligence outright and sentencing developers who ignore the rules to 20 years in prison. President Trump’s Joint Commitment On Frontier Responsibilities takes the inverse approach — a “morally binding” agreement relying largely on tech companies policing themselves.