Bipartisan AI Bill Would Override State Laws, Mandate Safety Plans

House unveils sweeping AI framework amid preemption controversy

Two House lawmakers released a comprehensive artificial intelligence bill Thursday that would establish the first federal safety requirements for advanced AI systems while blocking states from enacting their own developer regulations—a provision drawing sharp criticism from both parties.

Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) introduced the 269-page discussion draft, which represents Congress's most significant bipartisan AI legislation effort before the August recess and likely the final opportunity to pass federal AI rules ahead of midterm elections, according to POLITICO, which first reported the details.

The framework would require leading AI developers to create and implement plans addressing catastrophic risks from their advanced models, including potential cybersecurity threats. Third-party auditors would verify compliance with these safety plans.

Federal oversight through Commerce Department

The legislation designates the Center for AI Standards and Innovation, an office within the Commerce Department's National Institute of Standards and Technology, as the primary enforcement body. The bill would formally establish CAISI—currently created only through executive order—and authorize $300 million over three years for its operations.

Major AI companies including OpenAI, Anthropic, and Google DeepMind already work with CAISI on voluntary model evaluations. The new framework would make these safety assessments mandatory for top developers.

The bill also addresses open-source security by requiring CAISI and the Cybersecurity and Infrastructure Security Agency to support digital security of open-source code. Eligible U.S.-based software maintainers would gain access to frontier AI models capable of identifying and fixing security vulnerabilities.

State preemption draws bipartisan opposition

The proposal's most contentious element is a three-year sunset provision that would override state AI safety laws, including recent regulations passed in California, New York, and Illinois. State lawmakers in Massachusetts and New York have warned Trahan against undermining their regulatory authority.

Opposition spans the political spectrum. Florida Governor Ron DeSantis criticized similar preemption proposals in December as "AI amnesty," while GOP gubernatorial frontrunner Rep. Byron Donalds publicly disagreed with President Trump's support for federal preemption.

Even moderate Democrats who previously supported bipartisan AI efforts, including Rep. Ted Lieu (D-Calif.), initially resisted the preemption requirements.

Why it matters

This legislation represents a critical test of whether Congress can establish baseline AI safety standards before the technology advances further—or whether the preemption fight will derail federal action entirely. The three-year sunset provision attempts to balance federal consistency with state innovation, but may satisfy neither camp. If the bill stalls, states will continue developing their own patchwork of AI regulations, potentially creating compliance challenges for developers while leaving gaps in safety oversight. The framework also signals how the Trump administration's light-touch regulatory philosophy is being translated into concrete legislative proposals that must navigate bipartisan concerns.

Additional cosponsors and next steps

Reps. Suhas Subramanyam (D-Va.), Scott Peters (D-Calif.), Scott Franklin (R-Fla.), and Erin Houchin (R-Ind.) are expected to cosponsor the framework.

The legislation aligns with President Trump's executive order signed Tuesday requesting voluntary government review of powerful AI models 30 days before public release. House Speaker Mike Johnson has expressed support for federal preemption of state AI laws.

POLITICO reported these details based on interviews with over a dozen people familiar with the draft legislation.