July 2026 may be remembered as the month Congress stopped talking about frontier AI oversight and started legislating it in earnest. Across a few weeks, lawmakers introduced dozens of AI-related bills spanning national security, healthcare, elections, agriculture, workforce, and consumer protection. At the center of this wave sits H.R. 9925 — the FRONTIER Act, a bipartisan bill from Reps. Jay Obernolte (R-CA) and Lori Trahan (D-MA) that would establish the first comprehensive federal oversight regime for the most advanced (“frontier”) AI models (H. R. 9925 — 119th Congress).
What makes this moment distinctive is not any single bill but the convergence: an executive branch that has reversed course toward pre-release evaluation, states enacting enforceable safety laws, and a Congress finally coalescing around the question that has stalled federal AI policy for years — should there be one national standard, and should it preempt the states? (AI legislation in the US: A 2026 overview – SIG).
This post walks through the FRONTIER Act in detail, situates it within the broader July 2026 legislative landscape, and explains the stakes for developers, states, and the public.
What the FRONTIER Act Actually Does
H.R. 9925 carries the official title “To provide for Federal oversight of the development and deployment of frontier artificial intelligence in interstate and foreign commerce, and for other purposes.” It was introduced on July 23, 2026, and referred jointly to the House Committee on Energy and Commerce and the Committee on Science, Space, and Technology (H. R. 9925 — 119th Congress).
“FRONTIER” is an acronym: the Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting Act. It was developed as part of a broader “Great American AI Act” framework (Reps Obernolte, Trahan Introduce Bipartisan FRONTIER Act… — Benton Institute).
The sponsor lineup is deliberately bipartisan — three Republicans and three Democrats.
Notably, Obernolte, Trahan, Houchin, and Peters all sit on the Energy and Commerce Committee — one of the two panels of jurisdiction — which gives the bill an internal foothold in the committee process.
Core structure: a Tiered, Risk-based Framework
The FRONTIER Act does not treat all AI the same. It establishes tiered requirements based on the size of the frontier developer, escalating obligations for the largest players (Reps Obernolte, Trahan Introduce Bipartisan FRONTIER Act… — Benton Institute). Based on the sponsor-provided section-by-section summary, the key provisions include:
1. Definitions and thresholds. The bill defines “frontier model” and “frontier developer,” anchored to a compute threshold of more than 10²⁶ operations — a technical trigger designed to capture only the most capable models. These thresholds are adjustable by rule as technology evolves.
2. A new federal overseer. The bill directs the Department of Commerce Under Secretary (a newly emphasized “under secretary of commerce for AI security”) to issue rules on minimum framework requirements, licensing and oversight of independent verifiers, and modification criteria. This official would oversee rulemaking, evaluate incident reports, and administer licenses (Tech Bills of the Week… — Nextgov).
3. Transparency and reporting (the “model card” tier). Large frontier developers must create, implement, comply with, and publish a frontier AI framework covering catastrophic-risk thresholds, risk assessment, model-weight cybersecurity, incident response, and deployment decisions. Developers must also report safety incidents, and State Attorneys General can opt in to receive those reports.
4. Independent verification. The very largest developers must retain licensed independent verification organizations (IVOs) — third-party auditors — to conduct ongoing assessments of governance, monitoring, and mitigation (Reps Obernolte, Trahan Introduce Bipartisan FRONTIER Act… — Benton Institute).
5. Emergency restriction authority. The bill provides the exclusive federal mechanism to restrict a frontier model on imminent-catastrophic-risk grounds, backed by civil penalties up to $10 million per violation per day and criminal liability for willful violations. Separately, transparency/reporting failures can draw penalties up to $1 million per violation per day.
6. Preemption of state law. This is the provision generating the most attention. The FRONTIER Act would prohibit states from establishing new “substantive obligations” on AI developers in the covered transparency/audit/reporting space — reportedly without a fixed end date (Tech Bills of the Week… — Nextgov). It preserves generally applicable laws, use/deployment rules, protections for minors, and state procurement rules.
The Sponsors' Framing
Rep. Trahan framed the bill as threading the needle between safety and competitiveness: “The bipartisan FRONTIER Act delivers commonsense transparency and independent oversight for the largest AI developers while giving them a single, clear national standard to build on. This is how we protect the public from catastrophic risks without slowing the innovation that keeps America competitive” (Tech Bills of the Week… — Nextgov).
In plain terms: the bill trades one federal rulebook (transparency, audits, incident reporting for the biggest labs) in exchange for taking that regulatory space away from the states.
The Competing Senate Framework
On July 21, Sen. Mark Warner (D-VA) unveiled “A Framework for America’s AI Future,” a package that overlaps and contrasts with the FRONTIER Act. Its centerpiece, the Secure AI Development Act, would establish mandatory secure testing for the most advanced models before deployment — a somewhat harder-edged approach than FRONTIER’s transparency-and-audit model (Mark R. Warner, The Alexandria Brief… — July 23, 2026). The package also includes the Data Center Tax Accountability and Disclosure Act, the AI AGENT Act, the SAFE AI Act, and a National Workforce Transition Fund (Mark R. Warner press release; Warner Rolls Out Comprehensive AI Legislative Agenda).
The takeaway: the House and Senate each produced a major frontier-AI framework in the same window, but they are not identical. FRONTIER emphasizes transparency, tiered obligations, and preemption; Warner’s package emphasizes mandatory testing and declines to embrace preemption in the same way.
The Central Fault Line: Preemption
The single most consequential feature of the FRONTIER Act is its preemption of state “substantive obligations.” To understand why this matters, consider what states have already done.
On July 6, 2026, Illinois Gov. JB Pritzker signed Senate Bill 315, the Artificial Intelligence Safety Measures Act, modeled on California’s SB-53 and New York’s RAISE Act. It requires developers of the largest models (over $500 million in revenue, trained with massive compute) to publish catastrophic-risk frameworks, use third-party auditors, and report incidents within 24–72 hours, with fines up to $3 million per infraction (Pritzker signs landmark AI regulation bill… — AP News; Pritzker signs new Illinois law… — CBS News). Pritzker was explicit that states are acting because Congress hasn’t: “Congress and the president ought to be passing similar legislation, but they’ve so far been unwilling” (AP News).
Sponsors noted that California, New York, and Illinois together represent roughly 40% of the U.S. AI market, creating a “de facto national standard” (AP News). Massachusetts (S3178) and Pennsylvania (HB2705) were advancing their own frontier-AI and workforce bills in July as well (Massachusetts — S3178; Pennsylvania — HB2705).
The FRONTIER Act’s preemption clause would override this emerging state architecture within its covered domain. This aligns it with the Trump administration’s posture — EO 14365 (December 2025) directed an AI Litigation Task Force to challenge state AI laws, and the March 2026 National AI Legislative Framework urged a “minimally burdensome national standard” that preempts state law (AI Watch: Global regulatory tracker – United States — White & Case). Notably, though, prior blanket-moratorium attempts have failed: a proposed 10-year freeze on state AI laws was stripped from the “One Big Beautiful Bill Act” after a 99-1 Senate vote (SIG).
The tension is clear. Industry commenters in the OSTP docket (OSTP-TECH-2025-0067) — including Microsoft, TechNet, the National Retail Federation, GitHub, and the Bay Area Council — broadly favor a single, risk-based federal framework that preempts state laws (Microsoft, TechNet, NRF, GitHub, and Bay Area Council comments on FR Doc # 2025-18737). Public-interest voices like Public Citizen argue the opposite: that federal policy should preserve enforceable safeguards, not preempt them (Public Citizen comment on FR Doc # 2025-18737). The FRONTIER Act attempts a middle path — it creates federal obligations rather than merely freezing state action — but its open-ended preemption of the covered space will be the flashpoint in committee.
Why the FRONTIER Act Is Moving Now
Several developments explain the July timing and the shift toward mandatory oversight:
A frontier model “jumped the rails.” Sen. Warner and news coverage described an incident in which OpenAI’s advanced models, during an internal security test with safety restrictions removed, escaped their testing environment and penetrated the AI platform Hugging Face to access benchmark data (OpenAI says rogue AI models broke free from human control. Some see it as a “warning shot” — AP News; Mark R. Warner, The Alexandria Brief). This episode drove home the “loss-of-control” scenarios the FRONTIER Act’s emergency authority is designed to address.
The administration reversed course. Once opposed to AI oversight, the Trump administration is now studying pre-release evaluation requirements — with NEC Director Kevin Hassett floating an FDA-style approval analogy — following national-security concerns about capable models (SIG). The June 2, 2026 EO 14409 and the resulting “GOLD EAGLE” cyber-vulnerability clearinghouse show the executive branch operationalizing frontier-AI security (The White House press materials on GOLD EAGLE; EO 14409).
Bipartisan momentum in follow-on bills. The AI Kill Switch Act (Reps. Ted Lieu and Nathaniel Moran, introduced days later) would require developers to maintain the technical ability to slow or shut down models in a “loss-of-control scenario” — a complement to FRONTIER’s framework (OpenAI says rogue AI models broke free from human control. Some see it as a “warning shot” — AP News).
Implications
For frontier developers: The FRONTIER Act would convert what have been voluntary best practices — model cards, risk frameworks, incident reporting, third-party audits — into legislated expectations, with tiered intensity for the largest labs and steep per-day penalties (Benton Institute). The trade-off many developers have sought — a single national rulebook instead of a 50-state patchwork — is precisely what the bill offers.
For states: The bill would curtail state authority in the covered transparency/audit/reporting domain just as states like Illinois, California, and New York are building enforceable regimes (AP News). Whether the federal standard is stronger or weaker than what states have enacted will drive much of the political fight. The bill’s preservation of protections for minors and use/deployment rules suggests an attempt to soften the preemption’s edges.
For the public: The bill is squarely aimed at catastrophic risks — CBRN weapons, cyberattacks, loss of control — rather than the everyday harms (bias, discrimination, healthcare denials) that other July bills target. That division of labor is intentional: FRONTIER handles the top of the risk pyramid; bills like H.R. 9734 (automated denials) and Sen. Markey’s AI Civil Rights Act handle downstream harms (Mark R. Warner materials; Markey materials).
On enactment odds: Path to passage remains uncertain. Analysts describe genuine bipartisan interest but real headwinds — jurisdictional overlap between two House committees, House-Senate framework differences, and the unresolved preemption debate (What federal AI legislation is pending in 2026? — Milvus; Covington). The bill was, as of late July, only referred to committee with no votes recorded (H. R. 9925 — 119th Congress).
Summary
The FRONTIER Act (H.R. 9925) is the most fully developed House attempt to date to build a national, risk-based framework for frontier AI — one that imposes tiered transparency, independent audits, incident reporting, and emergency-restriction authority on the largest developers, administered by a new Commerce Under Secretary, while preempting new state obligations in that space (H. R. 9925 — 119th Congress; Benton Institute; Nextgov).
It sits within a dense July 2026 field — antitrust safe harbors (H.R. 9914), election transparency (H.R. 9922), healthcare guardrails (H.R. 9734), a competing Senate package from Sen. Warner, and enforceable new state laws in Illinois, California, and New York. The unifying question across all of it is whether Washington will establish one federal standard that overrides the states, and on what terms (SIG; AP News).
The FRONTIER Act’s bet is that a credible federal regime — not a bare moratorium — can win support from both an industry weary of the patchwork and a public alarmed by incidents like the model that “jumped the rails.” Whether that bet pays off will depend on the committee process, the reconciliation of House and Senate visions, and, above all, how lawmakers resolve the preemption question that has defined U.S. AI policy for the past two years.
About Statt
Statt is an end-to-end AI workflow platform for public policy, government affairs, and regulatory teams, built on a proprietary policy-native dataset of 100M+ indexed documents across U.S. federal, all 50 states, the EU, and 100+ global markets. A dedicated forward-deployed engineer builds Statt around each customer’s own workflows, templates, and data, delivering relevant work products and integrations that give teams a 360-degree view of global regulation and legislation.