Senate Negotiators Are Drafting a Bill That Could Block a Model From Shipping

Senate Negotiators Are Drafting a Bill That Could Block a Model From Shipping

For three years the American debate over artificial intelligence has been a debate about disclosure — what labs must publish, report, or promise. According to reporting published September 11, a small group of senators is now negotiating something categorically different: a federal veto over whether a model gets released at all.

Reuters correspondent Courtney Rozen reported that Senate negotiators are debating legislation imposing a legal duty of care on AI developers, requiring them to design products with the goal of preventing catastrophic risks. Negotiators aim to give the federal government power to block the release of certain models deemed unsafe, with companies able to challenge that decision in federal court. The proposal is being worked by Senate Majority Leader John Thune, Senate Commerce Chairman Ted Cruz and Senator Amy Klobuchar, the lead Democrat in the talks.

Almost none of it is confirmed on the record. Reuters attributed the duty-of-care provision to two Senate aides, the blocking authority to a source familiar with deliberations and one aide, and the court-challenge mechanism to an aide and a lobbyist in the negotiations. There is no bill number and no public text. Thune's spokesperson declined to comment.

What is reportedly in the draft

The measure would apply only to models with the most advanced capabilities — a tier that in practice covers Alphabet, Anthropic and OpenAI. Reuters reported that part of the bill would also block states from enforcing their own laws governing certain AI risks, with one aide citing a bad actor using AI to design nuclear or biological weapons as an example.

The most contested question is who does the testing. Nextgov/FCW reported the same day, citing multiple people familiar with negotiations, that the draft would have companies test their own models and present results to the Commerce secretary for deployment approval. Senator Maria Cantwell, the Commerce Committee ranking member, wants that replaced with mandatory vetting by federal agencies, including the national laboratories.

Cruz confirmed the effort on X, saying he is working with Klobuchar and Thune on legislation to address catastrophic risks involving biological or nuclear threats. Speaking Wednesday, he framed the balance: This is scary stuff, but we're also not going to be able to stick our head in the sand and pretend technology isn't happening. So we've got to put some guardrails on it.

Klobuchar told Reuters she is working toward bipartisan agreement on government oversight of the greatest risks posed by AI models, including requiring developers to work with government experts to verify and test models to make sure AI is safe. Cantwell, posting Thursday, welcomed the urgency while rejecting what she called a weak federal standard, saying meaningful legislation would require the most powerful models to undergo testing by scientists and experts at our national laboratories. A person familiar with negotiations told Nextgov/FCW that Republicans have created legal duties for companies to manage catastrophic risks, and that insisting on mandatory pre-deployment fixes would sink the bipartisan product.

Why this matters

If enacted as described, the blocking authority would be the first pre-market approval regime for software in American law. Drugs, aircraft and medical devices require federal clearance before sale. Code never has. That is the novelty, and it is why constitutional objections arrive immediately.

The sharpest is prior restraint. Commentators including the Foundation for Individual Rights and Expression argue that conditioning release of a system that generates text and code on government clearance implicates a doctrine carrying a heavy presumption of invalidity. The counterpoint is that publishing model weights may not be protected expression at all, in which case the doctrine never attaches — a question no court has squarely answered. A separate problem lurks in the delegation: instructing the Commerce secretary to block unsafe models without a defined threshold invites challenge as standardless. The reported federal-court appeal route reads as an attempt to pre-empt that objection.

The usual comparison is the EU AI Act, whose systemic-risk tier imposes evaluation, incident reporting and mitigation duties on the largest general-purpose models. But Brussels built no licensing gate. Obligations attach; a regulator does not hold release. As described, the Senate draft would go further than the EU on authority while covering fewer models.

Then there is federalism, and the timing is pointed. On September 9 — two days before the Reuters story — Governor Gavin Newsom signed California SB 813 and AB 1405, creating certified independent verification organizations and a state AI auditor registry, with first certifications due by January 1, 2028. Newsom paired the signings with a call for Washington to act. The Justice Department has been moving the other way: Attorney General Pam Bondi announced an AI Litigation Task Force on January 9, 2026, and on April 24 DOJ intervened in xAI's challenge to Colorado's AI Act, with enforcement suspended three days later. A preemption clause would land in the middle of an active fight, not a vacuum.

What to watch

Congress has failed here before. Thune and Klobuchar introduced the AI Research, Innovation and Accountability Act in 2023; it never passed. A proposed ten-year moratorium on state AI regulation collapsed in 2025. A Commerce markup set for before the August recess was cancelled, Cruz saying he wanted to focus on child safety.

The calendar is brutal. Reuters notes the House is scheduled for one week in session before the November 3 midterms; the Senate for three. Semafor reported September 10 that the bill could be introduced as early as next week, and that frontier labs and advocacy groups are already commenting on unreleased text. OpenAI global affairs chief Chris Lehane wrote this week that the AI policy window is open, for now. To Klobuchar's left, Senator Bernie Sanders is preparing a bill to pause development outright.

Three markers will tell you whether this is real: whether text is introduced before the recess; whether the testing regime lands on self-certification or national-lab vetting, the difference between a reporting statute and a licensing one; and how broadly preemption is drawn. Until a bill number exists, all of it remains sourced to people still arguing about it.