TL;DR
Anthropic representatives have been privately briefing members of Congress on draft federal legislation covering the safety and security of frontier models, POLITICO reported on 16 October, citing three people familiar with the talks and a copy of the confidential text. It sets a high bar before Washington could stop a model reaching the public, and it would override parts of state AI regulation.
What the draft says
The threshold is steep. Federal action to halt a model would be allowed only where it presented a foreseeable, imminent danger of over $1bn in damage, or of death or serious injury to over 50 people. That risk would have to arise from a biological weapon, from the model escaping its developer’s control, or from a cyberattack. POLITICO notes the wording mirrors California’s SB 53.
Independent evaluators, registered with Commerce, would periodically assess labs and their frontier models. Labs could pay them, provided payment did not depend on the results. Evaluators finding an “imminent enumerated risk” would have 72 hours to report it to Commerce.
Commerce could ask a federal judge in the District of Columbia to order a suspension. The draft explicitly bars any other US “department, agency, or officer” from pausing a model, whether in development, deployment or internal use. Labs would also need safety frameworks, whistleblower protections, a published transparency report for each new frontier model, and incident reporting to Commerce.
The bill would override some state laws on public risk disclosure, third-party evaluation and critical incident reporting, a contentious point, since Democrats have rejected proposals they feared would weaken laws passed in New York, Illinois and California.
Anthropic’s position
An Anthropic spokesperson said the bill reflects its Advanced AI Framework, which it says it has been “advocating for with federal and state lawmakers since June”, and called the measures “sensible policies”. POLITICO also recalls that export controls this summer shut down two of Anthropic’s most advanced models, a decision taken in a rapid series of calls that worried parts of the industry.
Looking forward
The draft enters a crowded field that includes the House FRONTIER Act and a Senate bill being negotiated by John Thune, Amy Klobuchar and Ted Cruz. In our view, a developer drafting the conditions under which its own models could be stopped deserves close scrutiny wherever it happens. For UK readers, the draft is a concrete example of the thresholds and reporting rules a frontier AI law could contain, as MPs on the BIST Committee pursue their inquiry into long-term AI safety.