Senate Majority Leader John Thune (R-SD), Sen. Amy Klobuchar (D-MN), and Commerce Chair Ted Cruz (R-TX) · 2026
Thune–Klobuchar–Cruz Frontier AI Safety Bill (Senate draft)
Thune–Klobuchar
A bipartisan Senate frontier-safety bill negotiated through the summer of 2026 by the Majority Leader, the Commerce Committee chair, and Sen. Klobuchar, building on the Thune–Klobuchar AI Research, Innovation, and Accountability Act that cleared committee in the previous Congress but organized around catastrophic risk rather than that bill's risk-tiered transparency. The text has not been published or introduced; draft language circulated to stakeholders in mid-September, and its provisions are known from reporting. As reported, it would impose a legal duty of care on frontier developers to manage catastrophic risks — cyberattacks, biological and nuclear weapons, and loss of control — require them to promptly notify the Commerce Department of significant capability advances, and have developers test their own models and present the results to the Commerce Secretary. Commerce could go to federal court to block the release of a model whose developer failed its risk-management obligations, with companies able to challenge the decision. It would preempt state laws on the same catastrophic risks. Talks stalled in late September: Cruz said any markup would have to wait for the lame duck, Klobuchar blamed the President's 'opposition to common-sense, bipartisan guardrails,' Ranking Member Cantwell opposed both the self-testing model and the preemption language, and a coalition of researchers and 19 advocacy groups wrote to Senate leaders objecting to the draft.
Key Provisions
Legal duty of care on frontier developers to manage catastrophic risks, including cyberattacks, biological and nuclear weapons, and loss of control
Prompt notification to the Commerce Department of significant capability advances in frontier models
Developers test their own models and present the results to the Commerce Secretary; Klobuchar describes developers working 'with government experts to verify and test models'
Commerce may seek a federal court injunction blocking the release of a model whose developer has failed its risk-management obligations, and companies may challenge blocking decisions in court
Preemption of state laws on catastrophic AI risk, establishing a single national standard — scope contested in negotiations
Text unpublished and not introduced as of October 2026; any committee markup deferred to the post-election lame duck
Regulatory Philosophy
Liability-backed prevention through the courts. Rather than build a licensing agency or a pre-release approval process, the draft converts catastrophic-risk management into a legal duty and gives the Commerce Department a judicial route to stop a release when that duty is breached. The design keeps the test in the developer's hands and puts the brake in a judge's — a structure Cruz has defended against proposals that would concentrate authority in Commerce itself. Preemption is the price of the federal duty, following the trade most federal frontier proposals now make: a real national obligation in exchange for displacing state law on the same risks.
Where the burden falls
FrontierPrimaryAppsNoneHyperscalersNoneChipsNone
Base assessed
Frontier developers (threshold not yet public)
Why it lands there
Every reported obligation — the duty of care, capability notification, self-testing, and exposure to an injunction — attaches to whoever trains the frontier model, so the incidence is pure frontier layer, including the labs inside hyperscaler parents. What the reporting does not reveal is the coverage threshold, which determines whether the duty reaches three companies or thirty. Preemption is the transfer in the other direction: removing state frontier laws relieves the same layer of obligations it already carries.
Thune–Klobuchar vs. Cantwell: the Commerce Committee's two answers
Both come out of the Senate Commerce Committee and both target the same catastrophic risks, and Cantwell's framework was released while the draft sat stalled. They split on who tests and what it costs the states. The draft has developers test their own models and report results to Commerce, with a court injunction as the brake; Cantwell holds that covered models should not be released until an independent audit confirms they meet NIST standards, and has pressed for testing by national-lab scientists. The draft preempts state catastrophic-risk laws; Cantwell's framework says nothing about preemption, and she has publicly opposed the draft's version of it.
+Led by the Majority Leader and the Commerce chair, giving it the clearest path to the floor of any frontier-safety bill in the Senate
+A duty of care creates liability at the design stage rather than only after harm, reaching decisions made before a model ships
+Gives the government a real, if judicial, mechanism to stop an unsafe release — something the June executive order expressly forecloses and the White House accord never contemplates
+Capability-advance notification gives Commerce early warning without requiring it to run its own evaluations of every model
+Judicial review of blocking decisions answers the due-process objection that sank broader licensing proposals
Weaknesses
From the perspective of political opposition
−The text is unpublished — the public, and most of Congress, are evaluating a bill they cannot read, and every provision here is reported rather than verified
−Self-testing keeps the evaluation in the hands of the developer whose release is at stake; Cantwell's alternative, testing by national-lab scientists before deployment, was rejected in the draft
−A court injunction is a slow brake for a fast risk: Commerce must first establish that a developer failed its obligations, and litigation runs on a timeline frontier releases do not
−Preempting state catastrophic-risk laws would displace California and New York's frontier statutes in exchange for a federal duty whose content is still being negotiated — the 'weak federal standard that becomes a backdoor for wiping out stronger state protections' Cantwell warned against
−Stalled before the election and dependent on a White House that prefers voluntary commitments, it may not survive into the next Congress
−Reported scope is catastrophic risk only — no workers, children, consumers, or copyright