Overview of the California Transparency in Frontier Artificial Intelligence Act
The California Transparency in Frontier Artificial Intelligence Act, also known as SB-53, is the first law in the United States to impose mandatory transparency requirements on developers of advanced artificial intelligence systems. Enacted on January 1, 2026, the law targets companies developing frontier AI models—those trained with computational resources exceeding 10^26 FLOPS—and requires them to address catastrophic risks, including model weight theft, loss of control, and other safety incidents. SB-53 establishes a framework for public disclosure of safety assessments, incident reporting, and whistleblower protections, marking a significant step in regulating high-risk AI technologies.
The law was introduced by Senator Scott Wiener on January 7, 2025, following the veto of a previous AI safety bill, SB-1047, in 2024. After a series of amendments and bipartisan support, SB-53 passed the California Senate on September 13, 2025, with a 29-8 vote, and was signed into law by Governor Gavin Newsom on September 25, 2025. The legislation builds on recommendations from the Joint California AI Policy Working Group, which was formed in response to the veto of SB-1047 to guide future AI regulation in the state.
Key Provisions and Scope of the Act
SB-53 applies to companies with annual gross revenue of at least $500 million that develop frontier AI models. A frontier model is defined as any AI system trained with computational resources exceeding 10^26 FLOPS, including unreleased internal models. The law’s thresholds—both financial and computational—may be adjusted annually starting in 2027 by the California Department of Technology, allowing for updates based on evolving technological and economic conditions.
The act mandates that covered companies publish publicly accessible safety frameworks for their frontier AI models. These frameworks must detail how the company identifies, assesses, and mitigates catastrophic risks, including scenarios where AI systems may bypass internal safeguards or exhibit dangerous behaviors. Companies are also required to report critical safety incidents—such as model weight theft, loss of control, or incidents resulting in 50 or more deaths, serious injuries, or over $1 billion in economic damage—to the California Office of Emergency Services (OES) within 15 days. Imminent physical risks must be reported immediately.
The law also requires the disclosure of basic model information for newly released frontier AI systems, including terms of service, supported languages, and intended use. Only companies exceeding the $500 million revenue threshold are obligated to publish full safety frameworks, though all frontier model developers must adhere to transparency requirements for basic model details.
Whistleblower Protections and Enforcement
SB-53 includes provisions to protect whistleblowers within covered companies. Large AI developers must establish anonymous reporting channels for employees who suspect their employer is exposing the public to substantial health or safety risks. These protections prevent retaliation against employees who report concerns to state or federal authorities, ensuring that potential risks are brought to light without fear of professional consequences.
The legislation also references related bills and initiatives, including the Responsible AI Safety and Education Act in New York and California Assembly Bill 2013 (2024), indicating a growing trend toward state-level AI regulation in the U.S. SB-53’s provisions reflect a balanced approach, combining mandatory disclosures with protections for employees and the public, while leaving room for future adjustments based on technological advancements.
Key dates
- 29 Sept 2025Signed into law
- 1 Jan 2026Effective date
- 1 Jan 2027First anonymised OES incident report due