Last updated:
The California AI Transparency Act, SB 942 as amended by AB 853, requires providers of widely used generative AI systems to embed provenance disclosures in AI-generated image, video and audio content and to publish a free public detection tool. It became operative on August 2, 2026, and is enforced by the California Attorney General.
When did the California AI Transparency Act take effect?
The Act became operative on August 2, 2026, not January 1, 2026, as many secondary sources still state. SB 942 was signed in 2024 with a January 2026 start date, then amended by AB 853, signed on October 13, 2025, which pushed the operative date back and broadened the scope. Anyone working from the original bill text will have the wrong date and an incomplete picture of who is covered.
Who counts as a covered provider?
A covered provider is a person or organization that creates, codes or otherwise produces a generative AI system with more than one million monthly visitors or users that is publicly accessible in California. The threshold attaches to the system, not to the company behind it. A large enterprise whose generative AI product has modest usage can fall outside the Act, while a small company with one widely used image generator falls inside it. Purely textual content is excluded, as are video game and streaming experiences, so the obligations bite on image, video and audio generation.
What disclosures does SB 942 require?
The Act sets out two kinds of disclosure. A latent disclosure is embedded metadata carried in the content itself, identifying the content uniquely, naming the generative AI system and version that produced it, and recording when it was created or altered. A manifest disclosure is a visible marking that clearly identifies the content as AI-generated, which the provider must offer users the option to apply and must embed permanently or near-permanently where that is technically feasible.
Separately, every covered provider must publish a free, publicly accessible detection tool that lets anyone check whether content came from its system. The tool has to accept file uploads and URL submissions, expose an API, and return provenance data without collecting more user information than the check requires.
What are the penalties for non-compliance?
Violations carry a civil penalty of 5,000 US dollars per violation, and each day of continued non-compliance counts as a separate violation. Enforcement sits with state attorneys general; the Act creates no private right of action, so the exposure is regulatory, not class-action driven. Providers who license their systems to third parties carry an additional duty: if a licensee modifies the system so that it no longer applies the required disclosures, the provider has 96 hours from discovering the tampering to revoke the license.
What changes in 2027 and 2028?
AB 853 phased the remaining obligations. From January 1, 2027, generative AI hosting platforms and large online platforms must detect provenance data in content they distribute and display it to users. From January 1, 2028, manufacturers of capture devices such as cameras and phones must enable provenance disclosures by default. Organizations that publish or host AI-generated media should treat the 2027 date as their own compliance deadline even if they are not covered providers themselves.
Real world example:
A design platform with 4 million monthly users offers AI image generation. Because the system crosses the one million monthly user threshold and is available in California, the platform is a covered provider. Before August 2, 2026, it embeds C2PA-style metadata in every generated image recording the model name, version and generation timestamp, adds an optional visible watermark that users can switch on, and publishes a detection endpoint where anyone can upload an image or paste a URL and see whether the platform produced it. Its enterprise licensing terms are rewritten to require licensees to preserve the metadata, with a 96-hour revocation clause if they strip it. The platform's stock photography library, which contains no AI-generated media, sits outside the Act.
