Senate negotiators are drafting bipartisan legislation that would impose a federal duty of care on the developers of the most advanced A.I. models and give Washington the authority to block their release in court, according to two Senate aides and a lobbyist involved in the talks cited by Reuters. Majority Leader John Thune, Commerce Chairman Ted Cruz, and Senator Amy Klobuchar are the principal architects. Semafor reports bill text could drop as early as next week.

The scope is narrow by design. It targets frontier systems, with Google, Anthropic, and OpenAI named as the American developers currently in that tier. Cruz, posting on X, said the bill would “address catastrophic risks involving biological or nuclear threats.” Speaking on ABC’s “The View” on Wednesday, he characterized covered models as ones that “would require government supervision and approval before they can be implemented,” while conceding of passage: “It is possible, but it ain’t easy.”

The mechanics are already contested. A Democratic committee aide told Nextgov the current draft leans on companies running their own safety tests and submitting the results to the Commerce secretary for deployment sign-off, a structure closer to industry self-attestation than independent review. Senator Maria Cantwell, the top Democrat on Commerce, wants mandatory pre-deployment testing by the national laboratories and national security agencies, and is resisting language that would preempt tougher state A.I. statutes. A person familiar with the matter told Semafor that Cantwell, Anthropic, and outside safety groups are “refusing to play ball,” while Klobuchar, in a statement to Reuters, framed the goal as getting companies “to work with government experts to verify and test models to make sure A.I. is safe.”

The calendar is the real antagonist. The House has one week in session before the Nov. 3 midterms; the Senate, three. Semafor calls this the only vehicle with a plausible shot before 2027.

Read structurally, the fight rhymes with the 1996 Telecommunications Act debates: a preemption clause that quietly reshapes the regulatory map matters as much as the headline safety regime. It also arrives into a policy environment already crowded by the Pentagon’s contested Anthropic contract, the upcoming U.S.–China safety talks, and a market where Anthropic’s Fable 5.1 release cut the cost of running agentic workloads. Every one of those threads runs through the same handful of models the Senate now proposes to gate.

Cruz’s “it ain’t easy” is the tell. The negotiators know the votes aren’t there yet. They’re drafting anyway because the alternative is arriving at 2027 with nothing.

Sources