FRONTIER Act
Sponsored By: Representative Obernolte, Jay [R-CA-23]
Introduced
Summary
Federal oversight of the most powerful AI models sets rules to reduce catastrophic risks by forcing transparency, independent checks, and a fast emergency stop for models that could cause mass harm or major damage. It centers on required disclosures, licensed independent verifiers, audits, and strong enforcement for high-risk "frontier" models.
Show full summary
- Communities and public safety: The bill defines a "catastrophic risk" as harms that could kill or seriously injure more than 50 people or cause over $1.0 billion in property damage. It gives the Commerce Secretary emergency authority to suspend or limit models that pose imminent catastrophic risk.
- Developers and operators: Firms that meet the Act's "frontier" thresholds must register, publish a public Frontier AI Framework, submit to annual third-party audits, and follow strict incident-reporting windows for safety events. The law targets developers above $50.0 million and very large firms above $5.0 billion in revenue.
- Independent verification and enforcement: The Under Secretary will license independent verification organizations with conflict-of-interest rules and GAO will monitor market capacity. Violations of emergency orders carry civil penalties up to $10.0 million and criminal exposure including imprisonment up to 10 years.
Personalized for You
How does this bill affect your finances?
Personalize government policy and PRIA will tell you what this bill means for your household, plus every other piece of legislation we track. PRIA reads each provision against your financial profile to show you exactly what matters to your wallet.
Bill Overview
Analyzed Economic Effects
7 provisions identified: 2 benefits, 1 costs, 4 mixed.
Federal preemption and state powers
The bill would stop States and localities from creating new substantive obligations on AI developers in three covered areas: risk transparency, third-party auditing/verification, and incident reporting for catastrophic risks. At the same time, States would keep power to enforce general laws that do not target developers, to regulate AI use by deployers or users (consumer, privacy, or criminal laws), to protect minors, and to set procurement rules for state use. The law would create a national floor for developer-facing rules while preserving several state authorities.
Emergency orders, fines, and review
This bill would let the Secretary of Commerce quickly suspend or limit development, deployment, or internal use of a frontier model when there is an imminent catastrophic risk. Provisional orders could lapse after about 45 days and final orders generally last up to 90 days and can be renewed. The Attorney General could sue in the D.C. district court to enforce orders, and courts must move cases fast. Willful violations of emergency orders could bring criminal fines up to $1,000,000 and up to 10 years in prison, and civil penalties for emergency-order violations could be up to $10,000,000 per violation per day; other covered violations can carry civil penalties up to $1,000,000 per violation per day.
What counts as catastrophic or critical
The bill would define "catastrophic risk" as a foreseeable risk that a frontier model could cause over 50 deaths or over $1 billion in property damage in a single incident. It would also define "critical safety incident," including stolen model weights, loss of control, or harms from a catastrophic risk. These definitions would guide when developers must report, when the Secretary can act, and when special mitigation and audits apply.
New transparency and audit rules
The bill would require large frontier developers to publish a public "frontier AI framework" within one year of enactment or 90 days after qualifying. Large developers would need annual independent audits (starting within a year or 90 days after qualifying) and must publish a high-level summary and a redacted report within 30 days. Before or when deploying a new or substantially changed frontier model, developers must post a machine-readable report listing intended uses, restrictions, risk assessments, and third-party involvement; some sensitive details may be redacted but unredacted copies must be kept and given to the Under Secretary.
Who counts as a regulated developer
This bill would define which AI models and companies the rules cover. A "frontier model" would be one trained with more than 10^26 operations, counting original training and major updates. A "large frontier developer" would be a company (with affiliates) that in the prior 36 months had more than $50 million in revenue and at least $1 billion in AI development spending; a "very large" developer uses thresholds of $5 billion revenue and $10 billion in AI spending. The Under Secretary could raise these thresholds by rule, with reviews at least every 2 years and any change taking effect not earlier than 180 days after publication.
Rules for independent verifiers (IVOs)
The bill would require any company that acts as a licensed independent verification organization (IVO) to get a license under rules issued within 180 days. Licenses must show independence from the AI industry and technical qualifications, and licenses can be revoked for material misrepresentation or loss of independence. Very large frontier developers would must use licensed IVOs at least every 6 months, give timely access to unredacted materials, and follow strict post-assessment timelines (developers respond in 14 days, IVOs opine in 7 days, post-assessment reports to the Under Secretary in 30 days). Licensed IVOs would get broad immunity from suit for harms they assess, but a narrow federal cause of action would allow claims for death or serious injury caused by willful misconduct.
Confidential safety reports and FOIA limits
The bill would require the Under Secretary to set up a confidential way to report critical safety incidents within 180 days of enactment. It would also exempt assessment reports and supporting materials given to the Under Secretary from disclosure under the Freedom of Information Act. Reporters could amend submissions and developers could keep sensitive material confidential while still giving unredacted copies to the regulator.
Sponsors & CoSponsors
Sponsor
Obernolte, Jay [R-CA-23]
CA • R
Cosponsors
Rep. Trahan, Lori [D-MA-3]
MA • D
Sponsored 7/23/2026
Rep. Houchin, Erin [R-IN-9]
IN • R
Sponsored 7/23/2026
Rep. Peters, Scott H. [D-CA-50]
CA • D
Sponsored 7/23/2026
Rep. Franklin, Scott [R-FL-18]
FL • R
Sponsored 7/23/2026
Rep. Subramanyam, Suhas [D-VA-10]
VA • D
Sponsored 7/23/2026
Roll Call Votes
No roll call votes available for this bill.
View on Congress.gov