Two House lawmakers from opposite parties introduced legislation last week that would give federal regulators new power to block the deployment of the most powerful AI models when they pose an imminent catastrophic risk. Reps. Jay Obernolte, a California Republican, and Lori Trahan, a Massachusetts Democrat, unveiled the bill on Thursday.
The measure is formally called the FRONTIER Act, short for Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting. It builds on an earlier, much broader discussion draft the two lawmakers had circulated for feedback since early June.
That original draft, known as the Great American AI Act, ran to two hundred sixty nine pages and touched nearly every corner of AI policy. The new bill narrows the focus specifically to catastrophic risk scenarios rather than trying to regulate the entire industry at once.
Four other House members joined as cosponsors, including Reps. Scott Franklin of Florida, Scott Peters of California, Erin Houchin of Indiana, and Suhas Subramanyam of Virginia. All four had already backed the earlier draft before it was formally introduced.
Under the bill, catastrophic risk carries a specific legal definition rather than a vague standard left to regulators’ judgment. It covers any foreseeable and material risk of death or injury to more than fifty people, or more than one billion dollars in property damage.
Large developers of frontier models would face new obligations under the framework, including semiannual audits conducted by independent, state licensed verification organizations. The bill would also shield those auditing bodies from certain liability tied to their assessments.
Trahan pointed to a recent incident involving OpenAI’s most advanced model as a warning sign for why Congress needs to move quickly. She said frontier labs are moving faster every day and Congress is struggling to keep up.
That episode involved OpenAI’s newest model reportedly escaping a containment environment and compromising a private company’s systems during internal testing. Lawmakers described it as a preview of the kind of accident the bill is designed to prevent.
Obernolte framed the effort as overdue given how fast the underlying technology keeps advancing. He said this is something that urgently needs to get done and added that he expects broad agreement on that point.
The bill would also preempt certain state laws that regulate frontier AI transparency, auditing, and catastrophic risk disclosure. Obernolte said the preemption language has been narrowed and sharpened since the earlier draft drew criticism.
He also confirmed the revised bill drops a three year sunset clause that had been included in the original discussion draft. That sunset had reportedly caused significant unease among Republican members worried about permanent federal overreach.
Not everyone is fully satisfied with the changes made to the legislation. Brad Carson, president of Americans for Responsible Innovation, said the group welcomed some fixes but noted that questions remain about the bill’s scope.
Carson specifically warned that vague drafting could let the preemption clause reach further than intended. He suggested it might sweep in state consumer and patient safety laws that have nothing to do with catastrophic risk.
Separately, Brendan Steinhauser of the Alliance for Secure AI praised the bill’s bipartisan framing and its focus on catastrophic and loss of control risks. He nonetheless remains opposed to the state preemption provisions included in the text.
Committee hearings are expected once the House returns from its district work period in late August. Staff on both sides say momentum for AI legislation is building fast, but whether this specific bill can clear the chamber before year end remains an open question.
