Summary: On June 4, 2026, Representatives Lori Trahan and Jay Obernolte released a 269-page Great American Artificial Intelligence Act discussion draft that would require major developers to run safety plans and undergo six-month audits while preempting state model-development rules for three years. It would also fund NIST’s CAISI with $300 million for frontier-model evaluations. Opposition from labor unions, consumer advocates, and a House Democratic commission appeared within hours, while Politico separately reported objections from state lawmakers; Axios reported on June 17 that progressive pressure continued.
Axios reported on June 17 that Representative Lori Trahan’s push for federal AI regulation was still drawing fire from progressives. On June 4, she and Republican Jay Obernolte had put forward a 269-page discussion draft — formally the Great American Artificial Intelligence Act of 2026 — and it stands as the weightiest bipartisan attempt at AI legislation this Congress has seen, as well as the last realistic opening for federal AI rules ahead of the August recess and the midterm elections.
At bottom, the draft proposes a trade. The biggest AI developers would take on binding safety plans and recurring third-party audits; what states give up is, for three years, the authority to enact fresh legislation on how models are built. Opposition surfaced within hours of the draft’s release — and it came, strikingly, from inside Trahan’s own party too.
What the Draft’s 269 Pages Actually Contain
Under the bill’s terms, the most powerful AI developers would have to draw up — and actually operate — safety plans targeting the potentially catastrophic risks their advanced models carry, first among them the prospect of fresh systems amplifying cybersecurity threats. Independent auditors would check compliance on a six-month cycle, and developers owe the government a running account of frontier model development as it unfolds.
Another provision converts an executive-branch initiative into law: the Center for AI Standards and Innovation (CAISI), housed inside the Commerce Department’s NIST, would receive $300 million to evaluate frontier models over the coming three years. A week earlier, Trump had tried to create the same office by executive order — but without money from Congress it was an empty shell, and this draft is what would actually fund it. Joining as co-sponsors at release were four more cross-party lawmakers: Suhas Subramanyam, Scott Franklin, Scott Peters, and Erin Houchin.
How the Preemption Clause Is Built: Development In, Deployment Out
The clause drawing the most fire is the preemption: for three years, states would lose the power to legislate specifically on AI model development — how systems are trained, built, and weighted — and the freeze would lapse automatically once the period ran out. Laws addressing how AI systems are deployed or used sit explicitly outside it: civil rights enforcement, labor protections, copyright, child sexual abuse material prohibitions, and consumer privacy all stay in state hands.
Sponsors present that line-drawing as a considered compromise. The FAQ published by Trahan’s own office, though, points to who gets caught in the crossfire, California’s AB 2013 first of all — the measure that requires model developers to post public summaries of their training data. That is precisely the brand of development-side transparency law the draft would shut out.
The Pushback, Inside and Outside the Party: “Turning the Floor Into a Ceiling”
Per TechTimes, rejection from nearly every direction landed within hours of publication — labor unions, consumer advocates, and even a formal House Democratic commission lined up against it. Politico’s reporting added that state lawmakers from Massachusetts and New York took their warning straight to Trahan: hands off their power to regulate AI developers.
The most withering critique came from Brad Carson, president of Americans for Responsible Innovation: as written, the bill would hoist “the current floor on state AI legislation” into a ceiling, and stripping states of their authority would amount to “a generational mistake.” The sweep of the initial rejection stood out for a discussion draft — a document whose whole purpose is to absorb criticism ahead of formal introduction. The coalition that usually lines up behind AI safety legislation fractured over whether any federal floor justifies surrendering state authority for three years. And as of June 17, Axios made clear, that progressive pressure had lost none of its force.
The Read for Developers and Product Teams
Three readings matter in practice. Timing: stakeholders can weigh in before formal introduction, with comments going to GAAIA@mail.house.gov; should nothing cohere before the August recess, the Congress that emerges after the midterms will be tougher terrain for legislation, not easier. Structure: passed as drafted, the bill would place model development under a single federal standard in place of fifty state regimes that keep accreting — exactly the attraction for the largest labs. Status quo: until the bill passes, nothing moves — state AI laws keep coming into force and keep multiplying, and compliance teams keep charting their obligations state by state. The calendar adds a wrinkle of its own: a bill that slips past the August recess does not simply carry over — a post-midterm Congress could restart from a different baseline, or never restart at all. This standoff between federal and state authority likewise carries forward the policy-tightening tone we noted in our opening outlook for 2026.