Public Knowledge, one of the loudest critics of Congress's first attempt at a federal AI safety framework, has endorsed the revised FRONTIER Act, telling lawmakers that the rewritten bill "addresses this feedback head-on" on the preemption question that sank support for its predecessor. The turnaround from a group that spent the spring and early summer opposing a broad federal override of state AI law marks one of the clearest signals yet that a bipartisan compromise on frontier-model regulation has a real shot in this Congress.
The bill, formally titled the Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting Act and known as the FRONTIER Act, is H.R. 9925 in the 119th Congress. It was introduced July 23, 2026, by Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA), alongside Representatives Scott Franklin (R-FL), Scott Peters (D-CA), Erin Houchin (R-IN) and Suhas Subramanyan (D-VA). The legislation grew out of an earlier discussion draft known as the Great American AI Act, which drew sharp criticism — including from Public Knowledge — for the breadth of the state-law preemption it proposed. After roughly seven weeks of stakeholder feedback, sponsors released the revised version now drawing endorsements.
"Artificial intelligence is advancing rapidly, and Congress must ensure our regulatory framework keeps pace without undermining American innovation," Obernolte said when the bill was introduced. "The FRONTIER Act focuses oversight on the largest developers and most advanced models, requiring transparency, independent evaluation, and timely reporting of serious safety incidents." Trahan struck a similar note, saying the bill "delivers commonsense transparency and independent oversight for the largest AI developers while giving them a single, clear national standard to build on."
What the Bill Actually Requires
The FRONTIER Act defines a "frontier model" using a compute threshold: a foundation model trained using more than 10^26 floating-point operations, counting both the original training run and any subsequent substantial modification. That makes it the first piece of federal legislation to peg regulatory coverage to a specific compute figure, mirroring an approach state laws — most notably California's SB 53 — pioneered first. Coverage is also tied to the size of the developer, with tiered obligations scaled to a company's AI research and development investment, a design Franklin said was meant to spare "the startups and innovators driving America's AI ecosystem" from one-size-fits-all mandates.
Covered developers must create, implement, and publish a written "frontier AI framework" addressing catastrophic-risk thresholds, risk assessment methodology, model-weight cybersecurity, incident response, and deployment decision-making. On top of the framework itself, the bill requires model cards describing capabilities and limitations, independent third-party audits of compliance with the framework, reporting of serious safety incidents, and ongoing reassessment rather than a one-time filing. Enforcement runs through a newly created Under Secretary of Commerce for AI Security, who would write implementing rules on minimum framework requirements, license and oversee the independent verification organizations that conduct audits, and evaluate incident reports as they come in.
The revision also added a mechanism absent from both the original discussion draft and any state law so far: a rules-based system for emergency suspension of a model's availability on safety or security grounds, paired with transparency requirements, defined limits on the executive branch's authority, and judicial review. Nicholas Garcia, senior policy counsel at Public Knowledge, wrote in an August 25 analysis that this "seemingly workable, if somewhat byzantine, system" was added partly in response to recent instances of AI model releases being delayed or restricted under what he described as pressure from the executive branch — though he cautioned the emergency-order authority "grants significant new powers to an administration that has been flagrant in its abuse of power and its disdain for legal limits."
The Preemption Fight
Preemption remains the bill's central fault line. The original discussion draft would have swept broadly across state AI law; the revised FRONTIER Act narrows that to what Garcia called "one-to-one preemption," under which the bill displaces state law only on the same catastrophic-risk transparency, evaluation, and reporting questions the federal framework itself addresses. "That one-to-one preemption, where any law preempted is replaced with an adequate federal framework, is exactly what Public Knowledge has called for in the past," Garcia wrote, calling it "the appropriate trade-off for a national legal regime." He added that "any sound proposal to establish a federal system for transparency, evaluation, and reporting on catastrophic risks in AI will — or should — look a lot like the FRONTIER Act."
Not everyone agrees the narrowing goes far enough. Public Citizen, the AFL-CIO, and the co-chairs of the House's bipartisan AI Commission have continued to argue the preemption clause could be read more broadly than its catastrophic-risk focus suggests, sweeping in state laws addressing algorithmic discrimination, consumer protection, or AI deployment in ways unrelated to frontier-model training risk. Garcia himself flagged this as unresolved, noting "the exact scope of its preemption will be fiercely litigated, likely to the detriment of fast-acting state enforcement," and that the bill "lacks the whistleblower protections present in the state laws." On the other side, some Republicans and industry voices say the bill doesn't preempt enough, since it leaves the bulk of state AI deployment and application law — as opposed to frontier-model development law — untouched. Subramanyan, one of the bill's Democratic co-sponsors, called the underlying threat serious enough to justify moving forward regardless: "This is a four-alarm fire. We need to take action now," he said, adding that the bill, "while not perfect, is an important, bipartisan step in the right direction."
Why It Matters
The FRONTIER Act lands in the middle of an increasingly crowded state regulatory landscape that it is explicitly designed to narrow. California's AI Transparency Act, which requires large AI platforms to offer free content-detection tools and embed watermarks in AI-generated content, became operative August 2, 2026. Colorado is preparing to replace its original AI Act with an amended framework taking effect January 1, 2027. Those are just two entries in a much larger list: state legislatures passed roughly 85 AI-related laws in 2026 alone, according to tracking cited across the policy bar, creating exactly the compliance patchwork that Obernolte, Trahan, and their co-sponsors say a uniform federal standard is meant to resolve. Whether Public Knowledge's endorsement translates into broader civil-society buy-in, or whether critics' preemption concerns keep the bill bottled up, will help determine whether Congress can actually deliver the "single, clear national standard" its sponsors are promising — or whether frontier AI regulation remains a state-by-state contest for the foreseeable future.
What to Watch
Watch for whether Public Knowledge's endorsement pulls other preemption-skeptical groups — Public Citizen and the AFL-CIO chief among them — into the FRONTIER Act's camp, or whether they hold out for further narrowing of the preemption language. Also watch how the bill moves through committee markup, whether the emergency-suspension authority draws sustained scrutiny given ongoing concerns about executive-branch influence over model releases, and how California and Colorado regulators respond as their own frameworks take effect this fall and winter, potentially setting up direct conflicts with whatever federal standard Congress ultimately adopts.
“That one-to-one preemption, where any law preempted is replaced with an adequate federal framework, is exactly what Public Knowledge has called for in the past.”— Nicholas Garcia, Senior Policy Counsel, Public Knowledge