On October 16, 2026, POLITICO reported that representatives for Anthropic "have quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI safety and security," citing "three people with knowledge of the efforts." An Anthropic spokesperson confirmed the intent: "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."
What the draft would require
Per POLITICO's reading of the confidential text, the bill would require frontier labs to "create safety frameworks, publish transparency reports for new frontier models, implement protections for whistleblowers and report critical safety incidents to Commerce." Independent evaluators would register with the Commerce Department and could be "paid for their work by the lab itself so long as the payment is not contingent on the evaluations' results." If they find an "imminent enumerated risk," they must report it to Commerce "within 72 hours of completing their assessment."
A high, narrow trigger -- and a court in the loop
The bill sets a deliberately high bar for intervention. The government could block a model's release or use "only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people," and only from three sources: a biological weapon, the model evading its developer's controls, or a cyberattack. POLITICO notes the language "mirrors that found in California's AI laws including SB 53."
Two structural choices matter as much as the threshold. First, a court, not the agency, pulls the trigger: Commerce could "bring a request to the judge to halt the release of a model," but "the language explicitly prohibits any other U.S. department, agency, or officer from suspending the model's development, deployment or internal use." Only a federal judge in the District of Columbia could order a suspension. Second, the bill "would also preempt some state AI laws related to public disclosure of risk, third-party evaluation and reporting of critical incidents" -- a flashpoint, POLITICO notes, with Democrats wary of proposals that "neuter regulations passed by the likes of California, Illinois and New York."
Anthropic's own June framework is consistent on the preemption point: it says Congress "should not preempt state law unless it enacts a rigorous federal regime that meets or exceeds the strongest measures" in the framework, and even then only for specific functions such as "catastrophic-risk testing of covered models, evaluator licensing, or closely related reporting obligations." The framework scopes obligations to developers above 10^25 training FLOP who also earn more than $500 million in annual AI revenue or spend more than $1 billion a year on R&D.
What remains uncertain. The draft is confidential and not introduced; POLITICO says Anthropic "has yet to publicly back particular legislative efforts in Congress." Competing vehicles are already moving -- the House FRONTIER Act from Reps. Trahan and Obernolte, and a Senate effort from Thune, Klobuchar and Cruz that faces resistance from Sen. Cantwell -- so whether any of Anthropic's language survives is open.
Analysis: The court-only trigger is the tell. After the summer, when Anthropic's own models were cut off by export controls in what POLITICO earlier described as "a whirlwind series of phone calls," the company is proposing rules that would route any future shutdown through a judge and a damages threshold rather than an official's phone. That is a reasonable due-process instinct and a self-interested one at once. The harder question for lawmakers is the $1 billion / 50-death floor: this summer's disclosed incidents, by the labs' own accounts, caused neither, so a regime built only for catastrophes would not have touched any of them. A company writing the rules for its own technology is worth reading closely on exactly what those rules leave out.
POLITICO reported October 16 that Anthropic has quietly circulated a confidential draft frontier-AI bill based on its June Advanced AI Framework; it would mandate safety frameworks, registered evaluators and 72-hour incident reports to Commerce, but let the government block a model only via a federal judge and only above a $1 billion or 50-death risk from bioweapons, loss of control or cyberattack, while preempting some state disclosure laws.