Anthropic’s draft federal AI safety bill, reported by Politico and confirmed in Reuters coverage on October 16, 2026, would let the Commerce Department block a frontier model only on a showing of “foreseeable, imminent” harm reaching $1 billion in damages or 50 deaths or serious injuries, with evaluators required to report imminent risk within 72 hours and challenges heard in federal court in Washington, D.C.

The headline threshold is the distraction. The quieter provision is the preemption language, which would displace state AI rules of the sort small businesses already comply with.

That’s the part that matters to any owner of a 5-to-30-person business running AI-assisted outreach. Connecticut’s CART Act, New Jersey’s disclosure bills with $5,000-a-day fines, and Colorado’s ADMT rulemaking all regulate deployers, not just frontier labs. Preempt them with a catastrophic-risk floor and the transparency obligations shaping what AI marketing tools can legally do disappear by implication.

The backlash came fast. California State Senator Scott Wiener, author of SB 53, said the draft “falls far short of protecting public health and safety” and accused it of borrowing a definition from his transparency law to “gut” its own shutdown powers. Nathan Calvin of Encode called himself “viscerally angry.” LawAI’s Charlie Bullock warned the evaluator market would “offer the least demanding checks at the lowest price.” David Sacks, the former White House AI czar, said Anthropic’s compute threshold would “sweep in most of Silicon Valley.”

On October 18, Anthropic’s federal affairs head Anthony Cimino conceded on X that the preemption wording was “not precise enough” and said any preemption should be “narrow, even scalpel-like.” A spokesperson said the bill reflects Anthropic’s June 2026 “Advanced AI Framework.” On October 20, OpenAI’s Chris Lehane added that “no one company (including OpenAI) should write the rules alone,” a line that sits awkwardly beside the bipartisan FRONTIER Act that Representatives Jay Obernolte and Lori Trahan introduced in July.

Services like LemonLime, which prepare AI-driven outreach and content for small businesses, sit squarely inside the deployer zone those state laws govern. Whether AEGIS passes, fails, or gets rewritten decides which rulebook applies.

Sources

Sources