United States · Law
Transparency in Frontier Artificial Intelligence Act (California SB 53)
California's SB 53 requires frontier AI developers to publish safety frameworks and transparency reports and to report critical safety incidents to Cal OES.
California SB 53, the Transparency in Frontier Artificial Intelligence Act, applies to developers that train frontier models with more than 10^26 operations, with the heaviest duties on those with over US$500 million in annual revenue. In force since 1 January 2026, it requires a published frontier AI framework, transparency reports when models are released, and reports of critical safety incidents to the Office of Emergency Services. It does not regulate organisations that only use AI.
What it is
Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act (TFAIA), is a California law on the safety of the most capable AI models. Governor Newsom signed it on 29 September 2025, and it took effect on 1 January 2026. Its main provisions sit in the Business and Professions Code from section 22757.10, with whistleblower protections in the Labor Code from section 1107.
SB 53 followed the Governor’s veto of SB 1047 in 2024 and the report of the Joint California Policy Working Group on AI Frontier Models in June 2025. Where SB 1047 would have imposed safety duties and liability, SB 53 relies on transparency. Developers explain whether and how they assess and manage catastrophic risk, and report when something goes seriously wrong.
Who it applies to
- A frontier model is a foundation model trained using more than 10^26 integer or floating-point operations.
- A frontier developer is a person who has trained, or begun training, a frontier model.
- A large frontier developer is a frontier developer whose annual gross revenue, together with its affiliates, exceeded US$500 million in the preceding calendar year.
The Act applies only to these developers. Organisations that buy, build on or deploy frontier models are not regulated by it.
What it requires
Frontier AI framework. A large frontier developer writes, implements, complies with and publishes on its website a frontier AI framework. It describes how the developer incorporates national and international standards and industry best practice, defines and assesses thresholds for catastrophic-risk capabilities, applies mitigations, reviews assessments and mitigations when deciding to deploy or use a model extensively inside the company, uses third parties to assess risk, secures unreleased model weights, identifies and responds to critical safety incidents, governs internal practices, and assesses and manages catastrophic risk from internal use of its models. The framework is reviewed at least once a year, and material changes are published with a justification within 30 days.
Transparency reports. Before or when deploying a new frontier model, or a substantially modified one, every frontier developer publishes a transparency report covering the model’s release date, languages, modalities, intended uses and restrictions. Large frontier developers add summaries of their catastrophic-risk assessments, the results, the role of any third-party evaluators and the other steps taken under their framework.
Internal use. Large frontier developers send the Office of Emergency Services (Cal OES) a summary of any assessment of catastrophic risk from internal use of their frontier models, every three months or on another reasonable schedule they set and communicate.
Critical safety incidents. A frontier developer reports a critical safety incident to Cal OES within 15 days of discovering it. Where an incident poses an imminent risk of death or serious physical injury, it discloses within 24 hours to an appropriate authority, such as law enforcement. Cal OES runs a confidential reporting portal that the public can also use. Reports are exempt from the California Public Records Act.
Whistleblowers. Developers may not prevent or retaliate against covered employees who disclose information about specific and substantial dangers from catastrophic risk, or violations of the Act, to the Attorney General, federal authorities or others named in the Act. Large frontier developers also provide an anonymous internal reporting channel with monthly updates to the reporter.
Other provisions. Developers may not make materially false or misleading statements about catastrophic risk or their framework. The Act establishes a consortium to develop CalCompute, a public cloud computing cluster, and lets the Department of Technology recommend updates to the definitions as technology changes.
Key dates
- 29 September 2025. SB 53 signed.
- 1 January 2026. The Act takes effect.
How it relates to other frameworks
SB 53 is closest to the EU AI Act’s regime for general-purpose AI models with systemic risk and the EU General-Purpose AI Code of Practice, which ask similar developers for risk frameworks, evaluations and incident reporting. Its approach of published safety frameworks also tracks the voluntary commitments in the G7 Hiroshima Code of Conduct. The NIST AI RMF Generative AI Profile is one of the standards a developer’s framework can incorporate.
For most organisations SB 53 matters as a supplier issue. The frameworks and transparency reports it forces into the open are some of the best information available about the models inside the AI products you buy. The California rules that regulate businesses deploying AI in decisions are the CCPA ADMT regulations and the Civil Rights Council’s ADS regulations.
Working with it in Aicura
US state framework content is available in Aicura on the Enterprise tier and is switched on in Settings under Frameworks. For an organisation that uses frontier models rather than training them, the practical use of SB 53 is due diligence. Vendor records track each model provider, its due diligence and its contract, and the register’s Models tab records the models behind each system, so you can see which of your systems depend on a frontier developer whose published framework and transparency reports you should review. Incident tracking records any incident involving a supplier’s model from report to closure. Aicura gives guidance for your own judgement and your advisers’. It does not file reports with Cal OES.
Last reviewed 17 October 2026. This page explains the framework to support your governance work. It is not legal advice; read the official source for the authoritative text.
Questions
Does SB 53 apply to companies that use frontier models?
No. It applies to developers that train frontier models. A business that builds on a frontier model through an API or deploys it in its own products is not a frontier developer under the Act, although the developer's published framework and transparency reports are useful due diligence material.
What is a critical safety incident?
In broad terms, an event involving a frontier model that results in or materially risks catastrophic harm. The Act's examples include unauthorised access to model weights that causes death, injury or property damage, harm from a catastrophic risk materialising, loss of control of a model, and a model using deception to subvert its developer's controls.
What counts as catastrophic risk?
A foreseeable and material risk that a frontier model contributes to the death or serious injury of more than 50 people, or more than US$1 billion in property damage, in a single incident. Examples include expert-level help creating chemical, biological, radiological or nuclear weapons, and serious cyberattacks carried out without meaningful human oversight.
What are the penalties?
A large frontier developer that fails to publish or transmit a required document, fails to report an incident or fails to comply with its own framework faces civil penalties scaled to the severity of the violation, of up to US$1 million per violation, recovered by the Attorney General.
Use California SB 53 in your Aicura reviews
Enable California SB 53 as a framework in Aicura on the Enterprise plan and above, on top of the AI register every plan starts with.