While most AI giants spend their lobbying budget asking Washington to move faster, Anthropic has quietly been doing the opposite since June 2026 — pushing its own draft federal AI safety bill on Capitol Hill. The details only became public on October 16, when Politico broke the story.
On paper the pitch sounds narrow. The government could only stop a model's release if the risk is "foreseeable and imminent": more than 50 deaths or serious injuries, or over $1 billion in damage, and only if it stems from bioweapons, a model slipping out of its operators' control, or a cyberattack. Independent evaluators registered with the Commerce Department would do the assessing, with 72 hours to flag a threat before the agency has to go to a federal judge. Agencies themselves get no power to block a release on their own.
The backlash came from both directions at once. California state senator Scott Wiener, the author of SB 53, said the draft "falls far short" of protecting public safety, pointing out it sets no floor for safety plans or auditor standards and effectively strips state courts of emergency shutdown powers. From the opposite camp, former White House AI czar David Sacks warned the damage threshold was so low it would sweep in most of Silicon Valley, startups included.
On October 18, Anthony Cimino, who runs Anthropic's federal affairs team, admitted the preemption language was "not precise enough" and promised a rewrite. That did little to calm things down — Encode AI's Nathan Calvin said the draft made him "viscerally angry" for seemingly blocking state courts from ordering an emergency pause.
By October 21, CEO Dario Amodei himself weighed in with a lengthy post on X, conceding the company had made mistakes while insisting that racing for frontier capability doesn't contradict its safety mission. The awkward part: a bipartisan bill, the FRONTIER Act, is already sitting in Congress — and by most accounts, Anthropic's own version looks noticeably weaker.
