Status at a glance
Governor Gavin Newsom signed SB 53, authored by Senator Scott Wiener, on September 29, 2025, according to the governor's announcement, which calls it the nation's first frontier AI safety law. Law firm analyses, including Jones Day, place its operative date at January 1, 2026. As of November 9, 2026, SB 53 is in force and has not been amended.
California has kept building around it. In September 2026 the governor signed SB 813, creating a framework for certifying independent verification organizations, and AB 1405, creating a registry for AI auditors, and on September 18, 2026 issued Executive Order N-9-26 directing agencies to recommend stronger frontier safeguards, as described in the governor's September 30, 2026 release and a Latham & Watkins alert. None of those measures changes SB 53's text.
What is California SB 53?
SB 53 is a disclosure and reporting statute for the developers of the largest AI models. The Future of Privacy Forum's analysis summarizes four main obligations: large developers publish a frontier AI framework, all covered developers publish transparency reports before deployment, developers report critical safety incidents to the state, and employees who raise catastrophic risk concerns are protected from retaliation. The law also creates CalCompute, a consortium within the Government Operations Agency to develop a framework for a public computing cluster, and directs the Department of Technology to recommend updates to the law each year (governor's release).
The law is organized around catastrophic risk, which FPF describes as a foreseeable and material risk that a frontier model contributes to the death or serious injury of 50 or more people or at least $1 billion in damage through expert-level help with chemical, biological, radiological or nuclear weapons, through autonomous cyberattacks or crimes, or by evading its developer's control. SB 53 does not license models, require pre-training safety measures or mandate a shutdown capability.
Who SB 53 applies to
SB 53 uses two tiers, as set out by FPF. A frontier model is a foundation model trained with more than 10^26 integer or floating-point operations, counting compute from the original training run and later fine-tuning or modification. A frontier developer is a person that trained or initiated the training of a frontier model. A large frontier developer is a frontier developer that, together with affiliates, had annual gross revenue above $500 million.
All frontier developers must publish transparency reports, report critical safety incidents and comply with the whistleblower rules. Only large frontier developers must write and publish a frontier AI framework. SB 53 does not expressly limit itself to developers based in California; a Morrison Foerster comparison with New York's law observes that SB 53 may therefore have extraterritorial reach.
What frontier AI developers must do
Publication requirements
Frontier AI framework. A large frontier developer must publish on its website a framework describing how it identifies, assesses, mitigates and governs catastrophic risk, including how it incorporates national and international standards and industry best practices, its cybersecurity practices, its governance structures, and how it assesses catastrophic risk from internal use of its own models. FPF notes the framework is reviewed annually and that redactions are allowed for trade secrets, cybersecurity and national security.
Transparency report. Before or when deploying a new frontier model or a substantially modified one, every frontier developer must publish a report covering the model's intended uses, modalities and restrictions. Large frontier developers must add summaries of their catastrophic risk assessments, the results, and the role of any third-party evaluators (FPF).
Reporting requirements
Critical safety incidents. Frontier developers must report critical safety incidents to the Office of Emergency Services within 15 days of discovery, or within 24 hours if the incident poses imminent danger of death or serious injury. Covered incidents include unauthorized tampering with a model that causes serious harm, the materialization of a catastrophic risk, loss of control of a frontier model that results in injury or major property damage, and a model deliberately evading its developer's safeguards. OES must also open a channel for the public to report incidents (FPF; governor's release).
Whistleblower provisions
SB 53 prohibits retaliation against employees and contractors who report activity creating catastrophic risk. Developers must give employees notice of their rights and maintain anonymous reporting channels, according to FPF. The governor's office lists whistleblower protection as one of the law's four accountability pillars (release).
Effective date and first reporting deadlines
SB 53 became operative January 1, 2026 (Jones Day). It does not set a single calendar filing date for developers. Instead, its duties run from triggering events: a large frontier developer must have its frontier AI framework published and keep it under at least annual review; any frontier developer must publish a transparency report before or at the deployment of a new or substantially modified frontier model; and the 15-day or 24-hour incident clock starts on discovery of a critical safety incident (FPF).
The practical first deadline for most large developers was therefore the first model release after January 1, 2026. We have not found a public tally of incident reports filed with OES as of November 9, 2026.
Penalties and enforcement
The California Attorney General may bring a civil action for violations, with civil penalties of up to $1 million per violation scaled to the severity of the violation (FPF; governor's release).
The law gives no rulemaking authority to the attorney general over its key definitions. FPF notes that an earlier draft did, and that the final text instead has the Department of Technology recommend definitional updates, which only the legislature can adopt.
SB 53 vs SB 1047
SB 53 is Senator Wiener's second attempt. His 2024 bill, SB 1047, was vetoed on September 29, 2024; in his veto message Governor Newsom asked for an approach informed by empirical analysis of AI capabilities, and a working group he convened published a report on frontier AI policy in June 2025. Our SB 1047 explainer covers that bill in detail.
Comparing the two, FPF lists what SB 53 dropped: the full shutdown (kill switch) requirement, pre-training safety obligations, mandatory annual third-party audits, a 72-hour incident reporting window and penalties tied to a share of compute cost. SB 53 focuses on deployment-stage disclosure, lengthens incident reporting to 15 days and caps penalties at $1 million per violation.
How SB 53 compares with New York and Illinois
New York's RAISE Act, as amended in March 2026, adopts the same compute and revenue thresholds, requires critical safety incident reports within 72 hours, caps penalties at $1 million for a first violation and $3 million for later ones, and takes effect January 1, 2027 (Morrison Foerster). Illinois' Artificial Intelligence Safety Measures Act (SB 315), signed July 6, 2026, goes further by requiring annual independent audits of large frontier developers starting in 2028 (Hunton).
SB 53 has applied since January 1, 2026 to developers that trained a foundation model with more than 10^26 operations. They must publish transparency reports and report critical safety incidents to the state, and those with over $500 million in annual revenue must also publish a frontier AI framework. The California Attorney General can seek civil penalties of up to $1 million per violation.
What changed
- 2026-09-30: Governor signed further AI bills and referenced SB 53 as the base of California frontier AI safety policy; SB 53 text unchanged.
- 2026-09-18: Executive Order N-9-26 directed agencies to recommend stronger frontier AI safeguards.
- 2026-01-01: SB 53 took effect.
- 2025-09-29: Gov. Newsom signed SB 53.
- 2024-09-29: Gov. Newsom vetoed SB 1047, the predecessor bill.
Frequently asked questions
What does SB 53 require?
Frontier developers must publish transparency reports before deploying new frontier models, report critical safety incidents to the Office of Emergency Services, and protect whistleblowers. Large frontier developers must also publish and follow a frontier AI framework for managing catastrophic risk.
When does California SB 53 take effect?
SB 53 was signed September 29, 2025 and took effect January 1, 2026. Its reporting duties are triggered by events such as a model deployment or the discovery of a critical safety incident rather than a fixed annual filing date.
Who does SB 53 apply to?
Developers that trained or began training a foundation model with more than 10^26 operations. The heavier framework duties apply only to those with over $500 million in annual gross revenue, together with affiliates.
What are the penalties under SB 53?
The California Attorney General may seek civil penalties of up to $1 million per violation, scaled to the severity of the violation. There is no private right of action for the disclosure duties.
How is SB 53 different from SB 1047?
SB 1047, vetoed in 2024, would have required pre-training safety measures, a shutdown capability, annual third-party audits and 72-hour incident reports. SB 53 drops those and relies on published frameworks, transparency reports, 15-day incident reporting and penalties capped at $1 million.
Sources
- California Legislative Information: SB 53
- Office of the Governor: Newsom signs SB 53 (Sept. 29, 2025)
- Office of the Governor: SB 53 signing message
- Office of the Governor: SB 1047 veto message (2024)
- Future of Privacy Forum: California SB 53 explained
- Jones Day: California enacts SB 53
- Office of the Governor: AI bills signed Sept. 30, 2026
- Morrison Foerster: New York amends the RAISE Act
- Hunton: Illinois Governor signs frontier AI model law
This page is journalism and commentary, not legal advice. AI law changes quickly; check the primary sources above and consult qualified counsel for your situation.