House weighs blocking state AI laws

Rep. Jay Obernolte and Rep. Lori Trahan are discussing a House artificial intelligence bill that could block a narrow set of state AI laws, according to a May 15 Politico report citing lobbyists and advocates familiar with the talks. The proposal under discussion would preempt laws in states including California and New York that require major AI developers to disclose information about advanced models and related safety risks, Politico reported. The talks are unfolding as the Trump White House weighs its own approach to reviewing powerful models before release. That overlap has turned a technical question about model evaluations into a federal-state fight over who gets to set the rules. ### Which state laws are House negotiators said to be targeting? California enacted Senate Bill 53 on Sept. 29, 2025, creating the Transparency in Frontier Artificial Intelligence Act, according to Governor Gavin Newsom’s office. The law established disclosure and transparency requirements for frontier-model developers and was described by the state as first-in-the-nation frontier AI safety legislation. (politico.com) New York signed the RAISE Act on Dec. 19, 2025, and later finalized amendments on March 27, 2026, according to Governor Kathy Hochul’s office and legal analyses of the statute. The law requires large frontier-model developers to publish information about safety protocols, report certain incidents to the state within 72 hours, and submit to oversight by a new office in the Department of Financial Services. The law takes effect Jan. 1, 2027, according to Wiley’s summary of the final version. (gov.ca.gov) Politico reported that the House discussions are focused on state laws that regulate frontier-model developers directly, not the broader universe of state AI rules. That distinction matters because many state proposals address uses of AI in employment, housing, insurance or government services rather than the development of base models themselves. (governor.ny.gov) ### What is the fight between Obernolte and Trahan? Politico reported on May 15 that Obernolte and Trahan were considering a sunset that would let states regulate frontier AI development again after two years. The same report said the two lawmakers were struggling to agree on whether any federal vetting regime should be mandatory. (politico.com) Politico said Obernolte favored a lighter-touch or voluntary approach under which companies could decide whether to disclose some information to the government. Trahan, according to the report, wanted stronger accountability, including mandatory data-sharing requirements. Spokespeople for both lawmakers declined to comment to Politico. (politico.com) Punchbowl reported on May 13 that Obernolte said the talks were “going really well” and that he was eyeing the following week for possible bill introduction. Earlier reporting from Punchbowl, summarized in search results, said Trahan had acknowledged differences over how far to go in preempting state rules. ### Why is model vetting suddenly at the center of this? (politico.com) The White House has been discussing executive actions on frontier AI models since at least early May, according to Politico. Those discussions included a possible vetting regime to review national security risks from advanced models and, in some versions of the talks, a requirement that companies get a government green light before release. (punchbowl.news) Kevin Hassett, director of the National Economic Council, said in a Fox Business interview cited by Politico that the administration was studying an executive order so future AIs that create vulnerabilities would go through a process before release. A senior White House official later told Politico the administration was seeking “partnership” with companies rather than “government regulation,” underscoring that the internal debate was still unresolved as of May 7. (politico.com) Politico reported that the House negotiations have mirrored that White House split. In both places, the question is not whether powerful models should be evaluated, but whether disclosure and review would be voluntary, mandatory, or tied to some form of pre-clearance. ### What does “blocking state AI laws” mean in practice? H.R. 5388, introduced in the House on Sept. 16, 2025, would establish a temporary moratorium preempting certain state laws that restrict AI models and systems engaged in interstate commerce, according to Congress.gov. (politico.com) The bill says Congress found that a “patchwork of divergent State AI rules” creates conflicting requirements for firms. (politico.com) A narrower version of preemption aimed only at frontier-model developer rules would not erase every state AI law. Based on Politico’s description, it would instead carve out one category of state regulation — laws that require top model developers to make disclosures or follow state-set safety procedures for advanced systems. That is an inference from the reported scope of the talks and the text of existing state laws. (congress.gov) The political stakes are already visible. Politico reported on May 1 that Massachusetts state lawmakers had warned Trahan against helping override state AI rules, and advocacy groups sent letters opposing federal preemption of state AI laws. ### Where does this fit in the broader House AI push? Ted Lieu and Obernolte introduced the American Leadership in AI Act on April 27, 2026, as a package built from recommendations of the Bipartisan AI Task Force. (politico.com) Their release said the package covered standards, evaluation, research infrastructure, risk management, workforce issues and AI-enabled crime. (politico.com) The task force’s work already pointed toward AI evaluations as a federal policy subject. Obernolte said in a Jan. 14 House hearing statement that the 2024 task force report recommended strengthening the science of AI evaluation and standards. That does not settle the preemption fight, but it shows why model testing and disclosure have become central to current House talks. (lieu.house.gov) May 2026 is the next concrete marker. Obernolte said on May 13 he was looking at the following week for a possible bill introduction, while the White House was still weighing whether to issue an executive order on frontier-model vetting. (punchbowl.news) (congress.gov)