🛡 FRONTIER Act Introduced in U.S. Congress — Federal Oversight Bill for Frontier AI
On July 23, 2026, lawmakers led by Republican Jay Obernolte (R-CA) introduced bipartisan bill H.R. 9925 — the FRONTIER Act (Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting Act), a 74-page document. The bill defines a "frontier model" as a system trained with costs exceeding 10^26 FLOPs, including fine-tuning and RL stages, and divides developers into "large" — revenue over $50 million and AI development spending of at least $1 billion over 36 months — and "very large" — at least $5 billion and $10 billion respectively. The former must publish a "frontier AI framework" with risk thresholds, model weight cybersecurity, and incident response plans, as well as undergo annual independent audits. The latter must hire a Department of Commerce-licensed independent verification organization (IVO) for continuous assessment, including internal model use, with reports on "critical security incidents": weight leakage or theft, loss of control, or model deception of its own developer. Catastrophic risk is defined quantitatively — death of more than 50 people or damage exceeding $1 billion in a single incident, including assistance in creating chemical, biological, or nuclear weapons and cyberattacks without human involvement. In the event of "imminent catastrophic risk," the Secretary of Commerce receives the authority to emergency-suspend model development or deployment by order.
🌍 This is the first attempt in the 119th Congress to build comprehensive federal oversight specifically for frontier AI: a regime with a 10^26 FLOPs compute threshold and monetary thresholds would automatically cover the largest players — OpenAI, Anthropic, Google DeepMind, Meta, and xAI — on their next generation of models if passed. For the industry, this means mandatory safety frameworks within one year of signing, annual conflict-of-interest-free audits, and external verification for the largest developers; however, the methodology for calculating and validating compute claims is not described in the text.
👤 Key points for readers: the bill is currently in committees, and GovTrack gives it only a 4% chance of passage, so it is too early to speak of "new requirements." However, this is the most specific of the active federal plans to regulate frontier AI — with precise thresholds and deadlines: Commerce Department regulations within 180 days of signing, safety frameworks within one year, and a report to Congress by the Deputy Secretary for AI Safety starting January 1, 2028. The full 74-page text is available on GovInfo, and bill status can be tracked on GovTrack and congress.gov.
Source 1: https://www.govinfo.gov/app/details/BILLS-119hr9925ih Source 2: https://www.govinfo.gov/content/pkg/BILLS-119hr9925ih/xml/BILLS-119hr9925ih.xml