Anthropic has submitted a draft federal AI safety bill to US lawmakers, dated October 2026. The proposal relies on the company's Advanced AI Framework, introduced in June 2026. The core of the legislation defines a covered developer as an entity training models exceeding 10²⁵ FLOPs that also meets specific financial criteria.
To be classified as a covered developer, a company must generate over $500 million in annual AI-derived revenue or spend more than $1 billion annually on AI research and development. These firms would be subject to new transparency obligations. The bill also limits the Commerce Department's power to restrict model deployment to cases involving foreseeable catastrophic risks.
Catastrophic risks are defined as incidents causing damages exceeding $1 billion or resulting in the loss of more than 50 lives. Critics argue that the 10²⁵ FLOP threshold is outdated, noting that by June 2025, at least 33 models from 12 developers had already surpassed this mark. The figure is significantly lower than the 10²⁶ FLOP threshold set by California’s SB 53.
The proposal faces criticism for potentially favoring large incumbents who can afford the costs of compliance, such as safety frameworks and independent evaluations. Some observers also suggest the draft is narrower than the FRONTIER Act (H.R. 9925). Unlike the EU AI Act, which uses 10²⁵ FLOPs as a rebuttable presumption, the Anthropic bill functions as a mandatory compliance trigger.
