FRONTIER Act
- Sponsor
- Rep. Obernolte, Jay [R-CA-23]
- Committees
- Science, Space, and Technology Committee (primary)
- Last action
- Jul 23, 2026
Bottom line
The FRONTIER Act creates a comprehensive federal regulatory structure for high-risk AI models, aiming to balance innovation with public safety through independent verification, transparency, and federal enforcement powers.
What it actually does
This bill establishes a federal oversight framework for the development and deployment of 'frontier artificial intelligence' (AI) models, defined by their computational power. It mandates transparency and risk mitigation frameworks for large frontier developers, requires independent audits for very large frontier developers, and grants the Secretary of Commerce emergency authority to suspend or restrict models posing imminent catastrophic risks. The bill also preempts state laws in specific areas of AI oversight.
Proponents argue
Supporters argue the bill is crucial for proactively addressing the potentially catastrophic risks posed by advanced AI models, ensuring public safety and national security while fostering responsible innovation. They emphasize the need for transparency, independent oversight, and clear accountability mechanisms to prevent misuse or uncontrolled behavior of powerful AI systems.
Opponents contend
Critics might argue that the bill's definitions of 'frontier model' and 'catastrophic risk' are too broad or too narrow, potentially stifling innovation by imposing burdensome regulations on emerging technologies. Concerns could also be raised about the potential for regulatory capture by large developers, the feasibility of establishing a robust independent verification market, or the scope of federal emergency powers.
The bill is moderately complex due to its technical definitions and regulatory structure, requiring careful reading to grasp its full implications.
Sec. 1
Short Title
This section establishes the official short title for the Act as the 'Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting Act' or the 'FRONTIER Act'. This provides a concise and recognizable name for the legislation, clearly indicating its focus on advanced AI risk management.
Supporters argue
A clear and descriptive short title helps communicate the bill's purpose and scope to the public and facilitates legislative discourse.
Critics contend
No specific opposition is typically raised against the short title of a bill.
Tradeoffs
None.
Sec. 2
Key Definitions for AI Oversight
This section defines critical terms that determine the bill's applicability, including 'frontier model' (a foundation model trained with over 10^26 operations), 'catastrophic risk' (foreseeable risk of >50 deaths/serious injuries or >$1 billion property damage from specific AI actions), and thresholds for 'large' and 'very large' frontier developers (based on gross revenues and AI-related development expenditures). These definitions are central to identifying which AI systems and entities fall under the regulatory scope.
Sec. 4(a)
Mandated Frontier AI Risk Framework
This provision requires large frontier developers to write, implement, comply with, and publicly publish a comprehensive frontier AI framework. This framework must detail how the developer identifies risk thresholds, assesses capabilities that could pose catastrophic risks, decides on model deployment, utilizes third parties for assessment, updates the framework, ensures cybersecurity of nonpublic model weights, and responds to critical safety incidents.
Sec. 3(c), Sec. 5(a)-(d)
Independent Verification Organization (IVO) System
The bill directs the Under Secretary of Commerce for AI Security to establish regulations for licensing and overseeing Independent Verification Organizations (IVOs). These IVOs will be responsible for assessing the adequacy of very large frontier developers' AI frameworks and risk mitigation efforts. The regulations will cover conflict-of-interest rules, application procedures, technical qualifications, and grounds for license revocation, ensuring IVOs are independent and competent.
Sec. 5(b), (f)-(i)
Mandatory Independent Assessments for Very Large Developers
Very large frontier developers are required to retain a licensed IVO to conduct ongoing assessments of their AI frameworks, governance, risk-monitoring, and mitigation strategies to ensure acceptable levels of catastrophic risk mitigation. These assessments must occur at least every six months, with reports submitted to the Under Secretary and the developer. The developer must respond with corrective actions, and the IVO provides an opinion on their adequacy. Ad hoc and out-of-cycle assessments are also mandated under certain conditions.
Sec. 8
Secretary's Emergency Order Authority
This section grants the Secretary of Commerce the authority to issue emergency orders to suspend or restrict the development, deployment, or internal use of a frontier model if it presents an 'imminent catastrophic risk.' These orders can be provisional (up to 45 days) or final (up to 90 days, renewable), and are subject to administrative and judicial review. The Secretary must consult with the Under Secretary for AI Security and provide a written basis for the finding.
Sec. 9
Federal Preemption of State AI Oversight
This section establishes federal preemption over state or local laws that impose new substantive obligations on artificial intelligence developers in 'Covered Subject Areas.' These areas include frontier AI risk transparency, third-party auditing/independent verification, and incident reporting. States retain authority over generally applicable laws, consumer protection, civil rights, minor protection, and state government procurement of AI, but cannot create conflicting or additional regulations in the preempted areas.
Defines a 'frontier model' as a foundation model trained using a quantity of computing power greater than 10^26 integer or floating-point operations, including original training and subsequent fine-tuning or substantial modification.
Sec. 2(12)
Why it matters:While technical definitions are standard legislative practice, this specific threshold is critical as it determines the entire scope of the bill's regulatory application.
Case for: Proponents argue that providing a clear, quantifiable metric ensures that oversight is precisely targeted at the most powerful and potentially risky AI systems, allowing for focused regulatory efforts.
Case against: Critics might contend that such a specific numerical threshold could quickly become outdated due to rapid advancements in AI hardware and algorithmic efficiency, potentially requiring frequent updates or becoming irrelevant, leading to regulatory lag.
Estimated impact: This definition determines which AI developers and models are subject to the bill's mandatory assessments, reporting, and emergency order provisions, thereby defining the entire scope of the regulatory framework.
Defines 'catastrophic risk' as a foreseeable and material risk that a frontier model will materially contribute to the death or serious injury of more than 50 people, or more than $1,000,000,000 in property damage, arising from a single incident involving specific AI actions (e.g., providing non-public assistance for CBRN weapons, autonomous cyberattacks, or evading control).
Sec. 2(6)
Why it matters:This is standard for defining legal thresholds, but the precise details are highly consequential as they set the bar for regulatory intervention and enforcement actions.
Case for: Proponents argue that these clear, high-bar criteria focus regulatory efforts on the most severe potential harms, ensuring that the bill's stringent measures are reserved for truly catastrophic scenarios and avoiding over-regulation of less critical issues.
Case against: Opponents might contend that the thresholds are too high, meaning significant harms below these levels would not trigger the bill's most stringent protections. The specific examples of AI actions might also be seen as too narrow or difficult to prove causation for.
Estimated impact: This definition directly shapes the scope of 'catastrophic risk mitigation' and the conditions under which emergency orders can be issued, thus defining the core problem the bill addresses and the severity of incidents that trigger federal intervention.
Grants licensed Independent Verification Organizations (IVOs) immunity from suit and liability under Federal and State law for claims of loss caused by, arising out of, relating to, or resulting from the materialization of a catastrophic risk of a frontier model assessed by the IVO. The sole exception is an exclusive Federal cause of action for death or serious physical injury proximately caused by willful misconduct.
Sec. 5(q)
Why it matters:This is a significant protection for IVOs, potentially controversial, and could be seen as a way to ensure IVOs are willing to operate without excessive liability risk in a nascent and high-stakes field.
Case for: Proponents argue that this immunity is essential to incentivize organizations to become IVOs and perform critical safety assessments without fear of crippling liability, thereby ensuring the availability of independent oversight for complex AI systems.
Case against: Opponents might contend that such broad immunity reduces accountability for IVOs, potentially leading to less rigorous assessments if the consequences of their failures are minimal for the auditor, thereby undermining public trust in the oversight system.
Estimated impact: Significantly reduces the legal risk for IVOs, potentially increasing their participation in the market but also shifting some risk from auditors to affected parties, and potentially impacting the ability of victims to seek redress.