Issue · June 14, 2026

The preemption argument

A bipartisan draft would set one federal rulebook for building AI and bar states from writing their own for three years. The states already doing the regulating call it a giveaway.

What the bill would do

On June 4, Representatives Jay Obernolte and Lori Trahan released a 269-page draft they call the Great American Artificial Intelligence Act. Obernolte is a Republican from California. Trahan is a Democrat from Massachusetts. The draft does two things at once. It would require the largest AI developers to publish a plan for managing catastrophic risk and submit to independent third-party audits twice a year. It would also bar states for three years from passing new laws that “specifically regulate the development” of AI models. The freeze is narrower than the headlines suggest. It covers rules about how models are built. It leaves states free to regulate how AI gets used in hiring, housing, lending, and health care. Obernolte and Trahan put the text out as a discussion draft, which means it is meant to draw fire before anyone tries to pass it. Roll Call has the provisions. It drew plenty within the day.

The case for one rulebook

The argument for preemption is not new, and it is not only an industry talking point. Fifty states writing fifty AI laws creates real compliance cost, and the burden falls hardest on smaller companies that cannot staff a policy team in every capital. Backers also frame it as national security, arguing that one clear standard helps the United States move faster than China. The libertarian Reason Foundation makes the cleanest version of the case, treating model development as interstate commerce that no single state should govern for the whole country. Their claim is that a moratorium protects both innovation and the national market. Industry has wanted a federal floor for over a year. Sam Altman used a policy blueprint this spring to call for one durable federal framework and a set of capability tests that powerful models would have to pass. Congress already tried a blunter version last year. A ten-year ban on state AI laws rode along inside the big budget package before the Senate pulled it back out. The audit requirement in the new draft is the sweetener. It lets supporters say preemption arrives with real federal oversight attached, not a simple green light.

The case against the freeze

The pushback was loud and bipartisan in its own way. Labor unions, consumer groups, and a House Democratic commission rejected the draft within hours. The advocacy group Public Knowledge opposed it outright, arguing that a freeze on state law removes protections that exist now in exchange for federal rules that might never arrive. Their objection is that preemption trades a real safeguard for a promise. The states have a point about who showed up first. Colorado, California, and New York wrote enforceable AI rules while Congress argued. Jina John of the ACLU told the Christian Science Monitor that on AI, “the federal government really has been playing catch-up”. The awkward part for the state camp is what Colorado is doing right now. Its landmark law, SB 205, finally takes effect on June 30. Lawmakers there spent the spring trying to soften it, with a rewrite that would drop the duty to guard against algorithmic discrimination and push the whole thing into 2027. Preemption backers read that as proof the states cannot hold a line. The states read it as a law learning to walk. Both are describing the same bill.