US Bipartisan FRONTIER Act Proposes Federal Audits and Pause Power for Frontier AI

US Representatives Lori Trahan and Jay Obernolte introduced the bipartisan FRONTIER Act to address the potential risks of increasingly powerful AI systems. The legislation integrates previous regulatory approaches, requiring the most powerful models to adhere to responsible development principles. Hailed by multiple AI safety advocates as the strongest federal AI oversight proposal to date, it marks a substantial step forward in US frontier AI regulation.

Confirmed

The bill targets developers of frontier AI models trained with over 10^26 FLOPs, requiring them to publish transparency reports. Core regulatory mechanisms include establishing an Undersecretary for AI Security within the Department of Commerce; integrating transparency rules from California's SB 53 and the RAISE Act; allowing the government to set minimum requirements for frontier AI safety frameworks; and introducing licensed Independent Verification Organizations (IVOs) to assess developers' safety measures. For sufficiently large AI developers, the bill mandates systematic third-party catastrophic risk audits, evaluating not just compliance but whether the company meets an "acceptable catastrophic risk" standard. Additionally, the bill grants the federal government the power to pause the deployment of high-risk frontier AI models. The bill's text has already been published.

Industry Feedback

The Alliance for Secure AI is generally positive about the bill but has not yet formally endorsed it. AI safety advocate Nat Purser stated that, after extensive feedback on earlier federal framework discussion drafts, the FRONTIER Act has become the strongest AI safety proposal he has seen, though he hopes the text will be further refined. Supporters (as relayed by @dhadfieldmenell) argue that the third-party assessment mechanism will allow regulators to truly see inside AI labs' internal deployment situations, making it more efficient than ad-hoc executive branch responses. In communities like Reddit (posted by @ActualWizard), netizens are discussing the bill's chances of passage and the specific operational model of the IVOs.

Unconfirmed

According to views relayed by @MilesBrundage and @StephenLCasper, although the bill incorporates various feedback, it still has areas needing reinforcement. For instance, the bill's preemption clause lacks an expiration date and does not provide an immediate replacement for superseded state laws, suggesting that state-level AI laws should not be preempted before a federal alternative is in place. Furthermore, the bill still leaves the power to release high-risk models to developers. The details of these provisions remain under discussion, and whether it will ultimately pass in Congress is still unknown.

Why it matters

The bill signifies a substantial step in federal frontier AI regulation, with regulatory scope and stringency exceeding previous proposals. If passed, it will establish a systematic federal AI regulatory framework, profoundly impacting the development and deployment of frontier AI models.

2026-07-24 ~ 2026-07-25 · 12 related posts

Full story(2 episodes)→

Primary sources