Bipartisan 'Frontier Act' Introduced in Congress to Establish Federal Safety Regulations for Advanced AI

On July 23, 2026, a bipartisan coalition in the U.S. House of Representatives introduced the Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting (FRONTIER) Act. Spearheaded by Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA), the legislation represents the most comprehensive federal effort to date to regulate advanced artificial intelligence. The bill explicitly targets the developers of the most powerful "frontier" models, establishing a national framework for risk management, independent auditing, and incident reporting.[1][4][7]

The introduction of the FRONTIER Act marks a significant shift from voluntary guidelines to mandatory federal oversight. According to the bill's sponsors, the legislation is designed to ensure that the regulatory framework keeps pace with rapid technological advancements without stifling American innovation. This legislative push was reportedly accelerated by recent high-profile security incidents, including a major breach at OpenAI, which prompted lawmakers to transition from a discussion draft to formal legislation.[1][5]

Structurally, legal analysts note that the FRONTIER Act functions as a transparency and oversight statute rather than a strict licensing regime. It does not condition the initial release of an AI model on prior government approval, nor does it create a new dedicated regulatory agency. Instead, it imposes documented obligations that mirror existing financial and cybersecurity compliance frameworks, such as securities disclosures or SOC 2 audits.[3]

The legislation employs strict compute and financial thresholds to determine which companies fall under its purview. The highest regulatory tiers are reserved for "large" or "very large" frontier developers. A large developer is defined as a company with gross revenues exceeding $50 million and at least $1 billion in AI-related research and development expenditures. Very large developers must meet a $5 billion revenue threshold and $10 billion in AI expenditures.[7]

By tying regulatory obligations to these massive financial and computational benchmarks, the bill's authors aim to protect the broader AI ecosystem. Proponents argue this targeted approach avoids one-size-fits-all mandates, allowing startups and smaller innovators to build new technologies without navigating burdensome federal red tape. The evidence supporting this threshold approach rests on the premise that only the most resource-intensive models pose systemic or catastrophic risks.[1][6]

For companies that meet the thresholds, the FRONTIER Act mandates a rigorous set of recurring obligations. Developers must publish detailed model cards that explicitly describe a system's capabilities and limitations. Furthermore, they are required to maintain written risk-management frameworks and submit to independent, third-party audits to verify compliance with those frameworks.[3]

A critical component of the bill is its stringent incident reporting requirement. In the event of a serious safety incident or the discovery of a catastrophic risk vulnerability, covered developers must report the issue to federal authorities within 72 hours. This rapid-response mandate represents a significant tightening from earlier discussion drafts, which had proposed a 15-day reporting window.[6]

Perhaps the most unprecedented mechanism in the FRONTIER Act is found in Section 8, which grants the executive branch the authority to halt AI deployment. Upon a written finding of imminent catastrophic risk, the Secretary of Commerce is empowered to suspend or restrict a model's development, deployment, or even its internal use within the AI lab. These emergency orders are valid for a provisional 45 days and can be finalized for 90 days, subject to judicial review.[6]

While the tech industry has broadly welcomed the bill's focus on transparency over licensing, the legislation's preemption clause has become its most fiercely debated provision. The FRONTIER Act explicitly preempts state laws regarding frontier-risk transparency, third-party auditing, and incident reporting. This would effectively override comprehensive state-level AI regulations recently passed or pending in California, New York, and Texas.[3][4][6]

Proponents of preemption argue that a uniform national standard is essential to prevent a fragmented, fifty-state patchwork of regulations that would be impossible for developers to navigate. However, state lawmakers and consumer protection advocates counter that the federal government is moving too slowly, and that preempting state laws removes the only active guardrails currently protecting the public.[1][4]

AI safety organizations have raised significant concerns about the bill's reliance on corporate self-assessment. Americans for Responsible Innovation (ARI) criticized the legislation for leaving the initial decision to deploy potentially risky models entirely in the hands of the AI labs, which have strong financial incentives to release their products. ARI argues that relying on companies to perform their own cost-benefit analyses regarding catastrophic risk is a fundamental flaw in the oversight regime.[2]

Furthermore, critics point out that the FRONTIER Act mandates independent third-party audits but relies on a private assurance market that does not yet exist at scale. Safety advocates warn that it will take years to establish a robust, standardized ecosystem of AI auditors capable of rigorously evaluating frontier models. Because the bill's preemption of state laws would take effect immediately, analysts warn this could create a temporary regulatory vacuum.[2]

The legislation also faces scrutiny for what it omits. Public interest groups have highlighted the absence of whistleblower protections, arguing that employees inside major AI labs need a secure avenue to report violations of the proposed safety regime without fear of retaliation. The lack of such protections weakens the enforcement mechanisms of a bill that relies heavily on corporate transparency.[2][6]

The FRONTIER Act was introduced alongside a separate, complementary bill known as the AI Kill Switch Act, which would require developers to build technical mechanisms to throttle or shut down their most powerful systems. Lawmakers view these bills as a tandem effort to establish both the legal authority and the technical capability to manage rogue AI.[5][6]

The legislative path forward remains uncertain, though sponsors are pushing for an aggressive timeline. Co-sponsors have publicly called for the House to mark up the FRONTIER Act in September, with the goal of passing the legislation before the current Congress adjourns at the end of 2026. However, with competing proposals and deep divisions over state preemption, the bill will likely face significant revisions before reaching a floor vote.[5]