- jurisdiction
- US, California
- level
- state
- domain
- safety
- status
- in-effect
- scope
- Imposes transparency, safety reporting, and whistleblower-protection obligations on developers of frontier AI models trained using more than 10^26 FLOPs. Additional requirements for large frontier developers with annual revenue over $500 million.
- obligations
- Frontier developers must: publish a transparency report before or at launch of each new or substantially modified frontier model; report any critical safety incident to the California Office of Emergency Services within 15 days (24 hours for imminent risk of death or injury); comply with whistleblower protections for employees disclosing safety risks. Large frontier developers must additionally publish a frontier AI framework describing technical and organizational measures to assess and mitigate catastrophic risks, reviewed at least annually.
- effectiveDate
- 2026-01-01
- enactedDate
- 2025-09-29
- penalties
- Civil penalty not to exceed $1,000,000 per violation for failure to publish required documents, false statements, failure to report incidents, or failure to comply with own framework.
- appliesTo
- frontier developers (training compute greater than 10^26 FLOPs); heightened obligations for large frontier developers (annual revenue greater than $500 million)
- notes
- Signed September 29, 2025. The first US frontier AI safety law. Successor to the vetoed SB 1047; takes a lighter-touch approach focused on transparency and reporting rather than pre-deployment safety tests. CalCompute consortium provision operative only upon budget appropriation.
