When will independent AI safety audits become U.S. law? — Mandatory independent AI safety evaluations or audits

A qualifying federal law must require a safety evaluation or audit of AI models, systems, or safety practices by an evaluator independent of the entity being reviewed. Government evaluators and independent outside organizations qualify. The review must substantively assess safety risks, testing, safeguards, or risk management; disclosure paperwork or self-attestation alone does not qualify. The law may cover a defined class of models or entities and may delegate technical standards to an agency. Self-assessments, voluntary programs, studies, and reviews limited to privacy, fairness, or transparency without a safety component do not qualify. A later effective date does not delay qualification. An example of this is the FRONTIER Act (H.R. 9925).