AI Safety

Congress Introduces FRONTIER Act for AI Oversight in Record 7 Weeks

Bipartisan bill mandates 72-hour incident reporting and closes the 'Ilya loophole' on AI spending.

Deep Dive

On July 23, Representatives Obernolte and Trahan, joined by four bipartisan cosponsors, introduced the FRONTIER Act — the oversight core of the GAAIA discussion draft. It creates three coverage categories: any developer training past 10^26 ops (frontier developer), those with $50M revenue plus $1B in AI development spend over 36 months (large), and a top tier of $5B revenue / $10B spend (very large). Incident reporting is now 72 hours, down from 15 days. Large and very large developers owe a public safety framework, annual independent audit, and registration in a new federal registry that includes beneficial-ownership disclosure. The bill also preempts some state law, scoped to three specific functions. It’s Congress’s fastest move on AI regulation yet.

Key Points
  • Three-tier coverage: compute threshold of 10^26 ops for frontier developers; $50M revenue/$1B AI spend for large tier; $5B/$10B for very large tier.
  • Incident reporting required within 72 hours (down from 15 days in the GAAIA draft).
  • Closes the 'Ilya loophole' by counting AI development expenditures regardless of accounting treatment, catching well-funded labs with little revenue.

Why It Matters

The FRONTIER Act marks Congress's fastest AI regulation push, with real enforcement and transparency requirements for frontier model developers.

📬 Get the top 10 AI stories daily