The California AI Transparency Act took effect on August 2
The California AI Transparency Act requires large providers to make generated content carry information showing it was made by AI. It was enacted in September 2024. It only became operative on August 2, 2026. The legislative premise it rests on is the impact of synthetic content depicting people saying or doing things they did not, without their consent, on state government, California-based businesses, and residents.
The Act's application timeline
| Date | Who | Status |
|---|---|---|
| September 2024 | SB 942 enacted | Done |
| October 13, 2025 | Governor approves the amending bill AB 853 | Done |
| August 2, 2026 | GenAI providers with over 1M monthly users | Operative |
| January 1, 2027 | Large online platforms / GenAI hosting platforms | Upcoming |
| January 1, 2028 | Capture device manufacturers (devices first produced for sale on or after that date) | Upcoming |
[ Approved by Governor September 19, 2024. Filed with Secretary of State September 19, 2024. ] / deepfakes, defined to mean audio or visual content that has been generated or manipulated by artificial intelligence that would falsely appear to be authentic or truthful and that features depictions of people appearing to say or do things they did not say or do without their consent, on state government, California-based businesses, and residents of the state — From the date of the Governor's approval and the legislative premise regarding the impact of deepfakes on state government, businesses, and residents
The date was pushed back from January 1
The original operative date was January 1, 2026. AB 853, approved by the Governor on October 13, 2025, rewrote it to August 2, 2026. That puts California's provenance obligations on the same day the EU's labelling obligations for AI-generated content started running. The statute says nothing about the overlap, so whether it was deliberate alignment cannot be determined from published documents. For what began in the EU that day, see the EU's labelling obligations for AI-generated content.
AB 853, Wicks. California AI Transparency Act. / [ Approved by Governor October 13, 2025. Filed with Secretary of State October 13, 2025. ] / Existing law makes the California AI Transparency Act operative on January 1, 2026. This bill would delay the operation of the California AI Transparency Act until August 2, 2026. — From the bill title, the Governor's approval date, and the change to the operative date
These transparency obligations, applicable from 2 August 2026, complement other rules like those for high-risk AI systems or general-purpose AI models — From the statement that the EU transparency obligations apply from 2 August 2026
Covered providers are those with over a million monthly users
The obligations fall on the "covered provider." That means a person who creates, codes, or otherwise produces a generative AI system with over 1,000,000 monthly visitors or users that is publicly accessible within the geographic boundaries of the state. The line is not where the company sits but whether the service is publicly accessible in California. A business based in Japan running a generative AI service reachable from the state falls inside it once it meets the size threshold.
The statute defines a generative AI system as an AI that can generate derived synthetic content — text, images, video, audio — emulating the structure and characteristics of its training data.
The California AI Transparency Act requires a person that creates, codes, or otherwise produces a generative artificial intelligence system that has over 1,000,000 monthly visitors or users and is publicly accessible within the geographic boundaries of the state to make available an AI detection tool at no cost to the user / “Generative artificial intelligence system” or “GenAI system” means an artificial intelligence that can generate derived synthetic content, including text, images, video, and audio, that emulates the structure and characteristics of the system’s training data. — From the covered provider threshold and the definition of a generative AI system
The three obligations placed on providers
What the statute asks for is a visible disclosure, an invisible embedded one, and a detection tool open to everyone. The design lets someone looking at a piece of content trace how it came to be.
The three obligations on covered providers
Manifest and latent do different jobs
The statute separates two kinds of disclosure. One is manifest, defined as easily perceived, understood, or recognized by a natural person. The provider has to offer this as an option; whether it actually goes in is the user's call. It has to identify content as AI-generated and be clear, conspicuous, appropriate for the medium, and understandable to a reasonable person.
The other is latent, defined as present but not manifest. This one is not optional — the covered provider has to include it. To the extent technically feasible and reasonable, it conveys information about the content's provenance, either directly or through a link to a permanent internet website.
The bill would require a covered provider to offer the user an option to include a manifest disclosure in image, video, or audio content, or content that is any combination thereof, created or altered by the covered provider’s generative artificial intelligence (GenAI) system that, among other things, identifies content as AI-generated content and is clear, conspicuous, appropriate for the medium of the content, and understandable to a reasonable person. The bill would require a covered provider to include a latent disclosure in AI-generated image, video, audio content, or content that is any combination thereof, created by the covered provider’s GenAI system that, among other things, to the extent that it is technically feasible and reasonable conveys certain information, either directly or through a link to a permanent internet website, regarding the provenance of the content. — From the respective obligations for manifest (optional) and latent (embedded provenance) disclosures
(i) “Latent” means present but not manifest. (j) “Manifest” means easily perceived, understood, or recognized by a natural person. — From the definitions of latent and manifest
Publishing a free AI detection tool
The third obligation is the detection tool. Covered providers have to make one available at no cost to the user. It has to let a user assess whether a given image, video, or audio was created or altered by that provider's generative AI system, and output any system provenance data detected in the content. Being publicly accessible is a requirement too.
The structure hands verification of a service's output back to the service itself. Rather than routing it through a third-party verifier, the statute asks the provider to publish the means of checking.
to make available an AI detection tool at no cost to the user that, among other things, allows a user to assess whether image, video, or audio content, or content that is a combination thereof, was created or altered by that person’s generative artificial intelligence system and outputs any system provenance data that is detected in the content. / require a covered provider, as defined, to make available an artificial intelligence (AI) detection tool at no cost to the user that meets certain criteria, including that the AI detection tool is publicly accessible. — The first passage is the detection tool requirement from the AB-853 Digest; the second is the public accessibility requirement from the SB-942 Digest
Break the disclosure, lose the license within 96 hours
There are provisions for licensing a generative AI system to others. A provider who knows a third-party licensee modified a licensed system so that it can no longer include the disclosures has to revoke the license within 96 hours of discovering it. The licensee then has to stop using the system.
A licensee who does not stop can be sued by the Attorney General, a county counsel, or a city attorney for injunctive relief and attorney's fees.
The bill would require a covered provider that knows a third-party licensee modified a licensed GenAI system such that it is no longer capable of including the disclosures described above in content the system creates or alters to revoke the license within 96 hours of discovering the licensee’s action and would require a third-party licensee to cease using a licensed GenAI system after the license for the system has been revoked by the covered provider. / The bill would, for a violation by a third-party licensee of the requirement to cease using a licensed GenAI system after the license of the system has been revoked, authorize the Attorney General, a county counsel, or a city attorney to bring a civil action for injunctive relief and reasonable attorney’s fees and costs. — From the 96-hour revocation requirement and the enforcement route against licensee violations
What widens from 2027, and what it means outside the US
The law does not switch on all at once. AB 853 stages further expansions in 2027 and 2028.
Who gets pulled in, and when
| Who | Start | Main obligation |
|---|---|---|
| Large online platforms | January 1, 2027 | Detect provenance data; do not strip provenance information |
| GenAI hosting platforms | January 1, 2027 | Do not make available GenAI systems that omit disclosures |
| Capture device manufacturers | January 1, 2028 | Offer an option for embedded latent provenance disclosure |
Platforms may not strip provenance
From January 1, 2027, large online platforms come in. The definition covers public-facing social media platforms, file-sharing platforms, mass messaging platforms, and stand-alone search engines that exceeded 2,000,000 unique monthly users over the preceding 12 months. They have to detect provenance data conforming to widely adopted standards, and, to the extent technically feasible, must not knowingly strip that provenance data or digital signature. Broadband internet access services and telecommunications services are excluded from the definition.
GenAI hosting platforms start the same day. Those are websites and applications that make source code or model weights available for download, and they must not knowingly make available a GenAI system that omits the required disclosures.
(h) (1) “Large online platform” means a public-facing social media platform, file-sharing platform, mass messaging platform, or stand-alone search engine that distributes content to users who did not create or collaborate in creating the content that exceeded 2,000,000 unique monthly users during the preceding 12 months. / (b) A large online platform shall not, to the extent technically feasible, knowingly strip any system provenance data or digital signature that is compliant with widely adopted specifications adopted by an established standards-setting body from content uploaded or distributed on the large online platform. (c) This section shall become operative on January 1, 2027. / 22757.3.2. (a) A GenAI system hosting platform shall not knowingly make available a GenAI system that does not place disclosures pursuant to Section 22757.3. (b) This section shall become operative on January 1, 2027. / (2) “Large online platform” does not include either of the following: (A) A broadband internet access service, as defined in Section 3100 of the Civil Code. (B) A telecommunications service, as defined in Section 153 of Title 47 of the United States Code. — From the definition and exclusions for large online platforms, the ban on stripping provenance data, and the obligations and start date for GenAI hosting platforms
It reaches cameras and phones too
From January 1, 2028, capture device manufacturers are covered. For any device they first produce for sale in the state on or after that date, they have to offer users the option to include a latent provenance disclosure in captured content. The disclosure carries the manufacturer's name, the device name and version number, and the time and date of creation or alteration — and it has to be embedded by default.
The definition of a capture device is broad: anything that records photographs, audio, or video, including video and still cameras, mobile phones with built-in cameras or microphones, and voice recorders. A regime for marking what AI made has circled back onto ordinary, unedited footage, and this is where the reach steps up.
The bill would also require, beginning January 1, 2028, a capture device manufacturer, with respect to any capture device the capture device manufacturer first produced for sale in the state on or after January 1, 2028, to, among other things, provide a user with the option to include a latent disclosure in content captured by the capture device that conveys certain information, including the name of the capture device manufacturer. The bill would define “capture device” to mean a device that can record photographs, audio, or video content, including, but not limited to, video and still photography cameras, mobile phones with built-in cameras or microphones, and voice recorders. / (1) Provide a user with the option to include a latent disclosure in content captured by the capture device that conveys all of the following information: (A) The name of the capture device manufacturer. (B) The name and version number of the capture device that created or altered the content. (C) The time and date of the content’s creation or alteration. (2) Embed latent disclosures in content captured by the device by default. — The first passage is the Digest on the capture device obligation, start date, and definition; the second is the statutory text listing the three disclosure items and the default-embedding requirement
The US is moving state by state
US AI regulation advances state by state, and the landing points do not line up. For another state law, see the Colorado AI Act explainer. What the California text shows is a design that skips heavy machinery like risk assessment and narrows onto a single point: provenance disclosure. For a comparison with the EU framework as a whole, see what actually starts on August 2 under the EU AI Act.
The full text sits on the California Legislature's website. If you only need specific sections out of a long bill, converting the page to Markdown first keeps section numbers and nesting intact.
Conclusion: check public accessibility in the state, then check your size
The California AI Transparency Act became operative on August 2, 2026, requiring generative AI providers with more than a million monthly users to carry provenance disclosures and publish a free AI detection tool. The starting point is not where you are established but whether anyone in the state can use your service. A business outside the US is covered once it meets the size threshold. Platforms and hosting services join in 2027, capture device manufacturers in 2028, so being outside the scope today does not mean staying outside it. Two things to check: whether your service is publicly accessible in California, and your monthly user count.
