California: SB 53 frontier AI law

2025-09-29 · Publisher : État de Californie · Domain : Artificial Intelligence

Gavin Newsom signs SB 53, the Transparency in Frontier Artificial Intelligence Act: large frontier model developers must publish their risk management framework and report critical safety incidents.

On September 29, 2025, California Governor Gavin Newsom signs SB 53, authored by Senator Scott Wiener. A year after vetoing SB 1047, deemed too restrictive, he adopts a text focused on transparency, inspired by the recommendations of an expert group convened at his request. California is home to 32 of the world's top 50 AI companies, according to the governor. ### What happened - **Scope**: "frontier models" trained with more than 10^26 operations. The heaviest obligations target "large developers" with annual revenue above $500 million. - **Transparency**: publication of a framework describing how the developer incorporates safety standards and best practices, and of transparency reports when new models are made available. - **Incidents**: critical safety incidents must be reported to California's Office of Emergency Services within 15 days, or within 24 hours in case of imminent danger of death or serious injury. - **Accountability**: whistleblower protections and enforcement by the Attorney General, with civil penalties of up to $1 million per violation. - **Innovation**: creation of CalCompute, a consortium tasked with designing a public computing infrastructure. ### Why it matters It is the first US state law specifically addressing frontier models. Without a federal law, California, home to most of the major labs, de facto sets a national transparency standard. Its approach partly aligns with the European Code of Practice, which eases joint compliance. ### Caveats The law mainly requires publishing and reporting; it does not subject models to prior authorization. How it fits with the federal administration's push to limit state regulations remains uncertain. ### Key takeaways Since January 1, 2026, when it took effect, major US labs must publish their catastrophic risk management policy and report serious incidents.

Tags : IA, Régulation, États-Unis, Sûreté, Transparence