Bipartisan AI Safety Bill Would Create Federal Auditing Regime

A bipartisan discussion draft in Congress would establish the first comprehensive federal framework for overseeing frontier AI systems, introducing mandatory transparency requirements and independent third-party audits for companies developing the most advanced models.

The Great American AI Act of 2026 (GAAIA), sponsored by Rep. Jay Obernolte (R-Calif.) and Rep. Lori Trahan (D-Mass.), emerged following Anthropic's April announcement of its Mythos model and subsequent government action to restrict access to the company's Fable 5 model over cybersecurity concerns. The Commerce Department's use of export controls to effectively mandate Anthropic revoke public access highlighted the absence of a proper regulatory framework.

Why it matters

The federal government currently lacks institutional capacity to respond to national security risks from advanced AI systems. When serious concerns arise, agencies must improvise using authorities never designed for frontier AI, creating legal uncertainty and denying companies due process. Building regulatory expertise takes years—waiting until a crisis forces action means responding without the necessary institutional foundation.

Key provisions for frontier AI oversight

GAAIA would relocate and expand the Center for AI Standards and Innovation outside NIST, granting it quasi-regulatory authority with $100 million in annual funding. The center would administer two core programs:

Transparency and incident reporting: Frontier AI companies would publish safety frameworks describing how they evaluate catastrophic risks, plus model cards detailing specific model capabilities and safety evaluation results. Companies must report "critical safety incidents" involving stolen model weights or concerning model behavior to government authorities.

Independent verification organizations: Third-party auditors would regularly evaluate whether companies adequately mitigate risks and comply with their own safety frameworks. CAISI would establish licensing requirements for auditors, conflict-of-interest rules, and minimum audit standards.

Federal versus state approaches

While California, New York, and Illinois have enacted similar transparency laws, GAAIA would leverage federal rulemaking authority unavailable to states. The Department of Commerce could prescribe detailed regulations governing the "form, manner, and minimum quality" of required disclosures.

State auditing regimes face structural limitations. Under Illinois's law, companies pay their own auditors, creating market pressure toward rubber-stamp approvals. GAAIA would grant CAISI authority to license auditors, revoke licenses, and establish rules preventing conflicts of interest—regulatory capacity state governments cannot match for interstate tech companies.

The bill also updates the expiring Cybersecurity Information Sharing Act and directs NIST and the Department of Energy to form international alliances on AI research and technical standards.

The preemption problem

GAAIA includes sweeping preemption of state AI laws, which the author argues makes the bill "net-negative as written" despite its strong federal framework. The discussion draft status indicates sponsors expect substantial revision before any floor vote.

This analysis was first reported by Charlie Bullock writing for Lawfare.