Anthropic pitches lawmakers on draft frontier AI safety bill

Anthropic representatives pitched U.S. lawmakers on draft legislation for frontier AI safety and security, according to reporting based on three people with knowledge of the effort. An Anthropic spokesperson separately confirmed that the privately circulated bill reflects proposals in the company’s Advanced AI Framework.

The reported proposal would combine catastrophic-risk testing, independent evaluation and government intervention authority for some frontier AI systems. No draft text, bill number or record of formal introduction was available in the research materials. The bill-specific thresholds and procedures below therefore remain attributed to the reporting, rather than established as enacted law or independently verified legislative text.

Anthropic’s public framework proposes applying obligations to developers that both train models above 10^25 floating-point operations and either generate more than $500 million in annual AI-derived revenue or spend more than $1 billion a year on AI research and development. Covered developers would have to test models for specified catastrophic risks, publish safety materials and testing summaries, and regularly engage at least one qualified independent evaluator.

The framework also proposes giving the government authority to block or deter dangerous deployments. Possible remedies include fines, a bar on further covered-model deployments until violations are fixed, and restrictions on access to already deployed models in extreme cases. Anthropic presents court enforcement and judicial review as safeguards and policy options, not powers currently in force.

According to the reporting, the draft would let the government block a model’s release or use only when a foreseeable and imminent risk could cause more than $1 billion in damage or death or serious injury to more than 50 people. The reported threshold would apply to harms involving a biological weapon, a model evading its developer’s controls, or a cyberattack. Those definitions and thresholds could not be checked against the draft itself.

The reporting also says independent evaluators would register with the Commerce Department and periodically assess covered laboratories and frontier models. Labs could pay evaluators as long as compensation was not tied to the assessment’s outcome, and evaluators would have 72 hours after finishing an assessment to report an “imminent enumerated risk.” Those rules likewise remain unverified against legislative text.

Under the procedure described in the reporting, the Commerce Department could ask a federal judge in the District of Columbia to halt a model. Other federal departments, agencies and officers would be barred from suspending model development, deployment or internal use. The opened materials do not identify the lawmakers or congressional offices approached, when meetings occurred, who drafted the language or whether any lawmaker has agreed to sponsor it. The Commerce Department did not immediately respond to a request for comment cited in the reporting.