The House went home on September 17 for a recess that runs until after the November midterms, and it left behind the most contested question in American AI policy: whether Congress will strip the states of their authority to regulate how frontier AI models are built. The Great American AI Act, the 269-page bipartisan discussion draft that reopened the preemption fight in June, never moved as a single bill. Its safety and preemption provisions now sit in a separate measure that has had no hearing, no markup, and — by its own lead sponsor's account — no path to the House floor before December.
Rep. Jay Obernolte (R-Calif.) said on September 17 that eleven bills carrying most of the package cleared the House Science Committee in July, that the frontier-safety bill could get a hearing in November, and that the pieces might then be combined into a single measure eligible for a December floor vote. None of the eleven has reached the floor.
What the bill actually does — and what it would preempt
Obernolte and Rep. Lori Trahan (D-Mass.) released the Great American AI Act as a discussion draft on June 4, 2026. Filed as one bill it would have gone to as many as eight committees, because it reaches federal AI research funding, cybersecurity, classroom curricula, teacher training grants, workforce programs and safety rules for the largest models. So it was broken apart.
The frontier-safety piece was introduced July 23 as H.R. 9925, the FRONTIER Act, with Reps. Scott Franklin (R-Fla.), Scott Peters (D-Calif.), Erin Houchin (R-Ind.) and Suhas Subramanyam (D-Va.) as original cosponsors. It went to Energy and Commerce and to Science, Space, and Technology — not, contrary to a persistent Washington shorthand, to Judiciary, whose Courts, Intellectual Property, AI and the Internet subcommittee held the memorable "AI at a Crossroads: A Nationwide Strategy or Californication?" preemption hearing in September 2025 but has not touched the bill since.
The preemption language shrank sharply between draft and bill. The June draft, as the Future of Privacy Forum documented, would have barred states for three years from enforcing any law that "specifically regulates the development" of an AI model, while preserving laws of general applicability and anything governing post-deployment use. The introduced version dropped the three-year window and narrowed the scope to frontier-safety requirements — transparency, incident reporting, and third-party auditing and verification — leaving state authority over deployment and use largely intact.
The substantive obligations track the state laws they would displace, mirroring California's SB 53, New York's RAISE Act and Illinois' SB 315: published frontier AI frameworks, pre-deployment transparency reports, confidential incident-reporting channels, whistleblower protections, and critical safety incident disclosure to the Center for AI Standards and Innovation within 15 days, or 24 hours where there is imminent risk of death or serious injury. It adds a $50 million revenue floor for "frontier developer" status and caps penalties at $1 million per violation, each day counting separately.
Both sides of the argument, at their strongest
The case for preemption is a coordination argument, and its sponsors make it plainly. "The FRONTIER Act is a targeted bill with clear, uniform transparency, and safety standards, instead of a fifty-state patchwork," Peters said on introduction. Obernolte's framing is that oversight should fall on "the largest developers and most advanced models" under one set of rules, so a company shipping a model does not face conflicting audit regimes in Sacramento, Albany and Springfield. Speaker Mike Johnson (R-La.) makes a harder version of the same point: "we cannot have a moratorium on the development of AI, okay, because then we will lose our edge to China."
The case against is a federalism-and-floor argument, and it is bipartisan. A coalition of 36 state and territorial attorneys general told congressional leaders in a November 25, 2025 letter that broad preemption would undermine states' ability to respond quickly to emerging AI risks, citing laws already on the books covering deepfakes, algorithmic rent-setting, AI disclosure and voter deception. State lawmakers sent their own letter opposing the Act's preemption clause on June 16, 2026. Rep. Sara Jacobs (D-Calif.) put the floor test in one sentence on September 16: "As a representative of California, I can't support anything that will mean that my constituents will have less protection than they do today."
Why this matters
Even the narrowed language leaves interpretive risk. As FPF noted, state AI laws rarely separate development from deployment cleanly; a statute requiring pre-deployment risk assessment for a hiring tool sits on both sides of the line, and courts would draw it.
The politics are moving faster than the text. The AI super PAC network Leading the Future has raised more than $125 million and spent over $24 million in federal primaries through June, with a single-issue objective: a national framework that preempts state law. It has also endorsed 18 Republican state legislative candidates across eight states. That money makes preemption a campaign issue rather than a drafting question, and it hardens Democratic resistance — the Senate voted 99-1 to strip a 10-year state AI moratorium from the 2025 reconciliation package, and nobody has forgotten the number.
Meanwhile the states keep legislating. Gov. Gavin Newsom signed California's SB 813 and AB 1405 on September 9, creating the first state registry for independent AI auditors, with covered audits restricted to registered auditors from January 1, 2029. Every month of federal delay adds another law a future preemption clause would have to reach backward to cover.
What to watch
Three things. Whether the frontier-safety bill actually gets the November hearing Obernolte described, and in which committee. Whether Trahan's lame-duck scenario materializes — she said on September 16 that a deal "will likely need it to be bipartisan, even if we win back the majority in November." And whether Senate Commerce produces the Thune-Klobuchar text that has been promised but never released; Majority Whip John Barrasso (R-Wyo.) said flatly that he does not see a solution "to the point that it's going to have a vote."
“As a representative of California, I can't support anything that will mean that my constituents will have less protection than they do today.”— Sara Jacobs, U.S. Representative (D-Calif.), House AI Caucus