Overview of California SB 53 — Transparency in Frontier Artificial Intelligence Act
California SB 53, also known as the Transparency in Frontier Artificial Intelligence Act, is the first law in the United States specifically targeting the regulation of frontier artificial intelligence models. Enacted on January 1, 2026, the law establishes mandatory transparency and safety requirements for developers of advanced AI systems. It is designed to mitigate catastrophic risks associated with AI, including those that could result in significant loss of life, severe injuries, or substantial economic damage.
The law applies to AI models trained using computational resources exceeding 10^26 floating-point operations (FLOPs), including fine-tuned models, and to companies generating annual gross revenue of at least $500 million. These thresholds define what constitutes a 'frontier model' under SB 53, though the California Department of Technology may adjust these criteria annually starting in 2027.
SB 53 was introduced by Senator Scott Wiener on January 7, 2025, following the veto of a prior bill, SB-1047, in 2024. The law was passed by the California Senate on September 13, 2025, with a vote of 29-8, and signed into law by Governor Gavin Newsom on September 25, 2025. Its development was informed by recommendations from the Joint California AI Policy Working Group, which was established after the veto of SB-1047.
Key Provisions and Requirements
The law mandates that developers of frontier AI models must create and maintain publicly accessible safety frameworks. These frameworks must detail how companies assess and mitigate catastrophic risks, including scenarios where AI models may bypass internal safeguards or oversight mechanisms. The frameworks must also outline how the models incorporate national and international safety standards.
Companies subject to SB 53 are required to report critical safety incidents to the California Office of Emergency Services (Cal OES). An incident is classified as critical if it results in more than 50 deaths or serious injuries, or causes over $1 billion in economic damage due to AI-assisted acts such as cyberattacks or the creation of biological weapons. Reports must be submitted within 15 days of discovery, unless the incident poses an imminent physical risk, in which case immediate reporting is required.
In addition to safety frameworks, companies must publish transparency reports for newly released frontier models. These reports must include basic information such as terms of service, supported languages, and intended use. For companies exceeding the $500 million annual revenue threshold, full safety frameworks must also be made public.
Whistleblower Protections and Reporting Mechanisms
SB 53 establishes protections for whistleblowers employed by companies developing frontier AI models. These protections include the requirement for anonymous reporting channels, ensuring employees can report concerns about public health and safety risks without fear of retaliation. Whistleblowers must have reasonable cause to believe their employer is substantially risking public safety before utilizing these channels.
The law emphasizes the importance of accountability and transparency in AI development. By providing structured pathways for reporting safety incidents and protecting whistleblowers, SB 53 aims to foster an environment where potential risks are identified and addressed promptly.
Historical Context and Legislative Development
The passage of SB 53 follows the unsuccessful attempt to enact the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act (SB-1047) in 2024, which was vetoed by Governor Gavin Newsom. In response to this veto, Newsom created the Joint California AI Policy Working Group to provide guidance on AI regulation. The recommendations from this group played a pivotal role in shaping SB 53.
The law represents a significant step in the regulation of AI in the United States, setting a precedent for how states may approach the oversight of advanced AI technologies. It reflects growing concerns about the potential risks posed by frontier AI models and underscores the need for proactive measures to ensure their safe and responsible development.
Required actions
- Large frontier developers publish a frontier AI framework describing how catastrophic risks are assessed and mitigated, and update it yearly - Publish a transparency report before deploying a new or substantially modified frontier model - Report critical safety incidents to the Office of Emergency Services within 15 days (24 hours where there is imminent risk) - Do not retaliate against employees who report catastrophic-risk concerns; provide an anonymous internal reporting channel