Status at a glance
Governor Newsom signed SB 942 by Senator Josh Becker on September 19, 2024 (governor's release). It was originally set to take effect January 1, 2026. AB 853, signed October 13, 2025, moved that date to August 2, 2026 and added duties for large online platforms and capture device makers (Orrick). The provider requirements became operative on August 2, 2026 (Morgan Lewis).
Two months later the act was rewritten again. SB 1000, also by Senator Becker, was signed on September 30, 2026 as an urgency statute and took effect immediately; AB 2713 by Assemblymember Buffy Wicks, signed the same day, refines the platform duties from January 1, 2027 (Latham & Watkins; governor's release). The Secretary of State chaptered SB 1000 as Chapter 861 and AB 2713 as Chapter 856 of the Statutes of 2026; SB 942 is Chapter 291 of 2024 and AB 853 is Chapter 674 of 2025 (California Legislative Information). SB 1000 passed the Senate 33 to 1, the Assembly 72 to 0, and the Senate concurred 40 to 0. As of November 9, 2026, the provider duties in force are those of SB 942 as amended by AB 853 and SB 1000.
| Phase | Who | Duty | Date |
|---|---|---|---|
| 1 | Covered GenAI providers | Latent disclosures in image, video and audio; free verification tool | Aug. 2, 2026 |
| 1a | Covered GenAI providers (SB 1000) | No user threshold; manifest option dropped; new latent fields | Sept. 30, 2026 |
| 2 | Large online platforms; GenAI hosting platforms | Detect and display provenance data; no hosting of noncompliant systems | Jan. 1, 2027 |
| 3 | Capture device manufacturers | Offer latent disclosure in captured content | Jan. 1, 2028 |
| 4 | Assistive technology GenAI systems | Temporary exemption ends | Jan. 1, 2029 |
What is the California AI Transparency Act?
The act, codified in the Business and Professions Code starting at Section 22757, is a provenance law. It does not require AI content to carry a visible label in every case; instead it requires machine-readable data embedded in the file, plus tools that let anyone check it. When signing SB 942, the governor's office described it as requiring widely used generative AI systems to include provenance disclosures that are invisible to humans but detectable by free tools (release).
The act covers image, video and audio content only. Morgan Lewis notes that by its own terms it does not apply to AI-generated text, and that as enacted it excluded products that are exclusively non-user-generated video game, television, streaming, movie or interactive experiences. Morrison Foerster reports that SB 1000 narrows that exemption to exclusively non-user-generated video games. Our analysis of watermarking and C2PA covers how provenance standards are being used to meet duties like these.
The vocabulary matters. A latent disclosure is data present in the file but not perceptible; a manifest disclosure is a label a person can see. Provenance data is information embedded in content or its metadata about its origin and modification history, and system provenance data is the subset that is not reasonably capable of being tied to a particular user, such as the type of system that made the content (Assembly Privacy and Consumer Protection Committee analysis). SB 1000 redefines provenance data so that it must be in a format compliant or interoperable with widely adopted specifications from an established standards-setting body (same source).
Who counts as a covered provider
The core duties fall on a "covered provider." Before SB 1000, that meant a person that creates, codes or otherwise produces a generative AI system with more than 1,000,000 monthly visitors or users that is publicly accessible within California. SB 1000 deleted the user threshold (Assembly committee analysis). The act defines a GenAI system as AI that can generate derived synthetic content, including text, images, video and audio, that emulates the structure and characteristics of its training data (same source).
Since September 30, 2026, a plain test works for most businesses. You are a covered provider if all three are true:
- You create, code or otherwise produce a generative AI system.
- The system is publicly accessible within California's borders. Where the company is incorporated or headquartered is not part of the definition.
- No exemption applies, such as the exclusion for exclusively non-user-generated entertainment products, or, until January 1, 2029, the exemption for systems designed primarily to function as assistive technology for people with disabilities.
The disclosure and verification duties attach only to image, video and audio content, so a text-only chatbot has nothing to mark (Morgan Lewis).
Latham & Watkins expects SB 1000 to reach startups, open-source model providers, smaller SaaS platforms and enterprise tools that sat below the old one million mark.
What SB 942 requires of generative AI providers
As in force on August 2, 2026, the act applied to covered providers: persons that create, code or otherwise produce a generative AI system with more than one million monthly visitors or users that is publicly accessible in California (Morgan Lewis). SB 1000 deleted the one million threshold, so any person producing a publicly accessible GenAI system in California is now a covered provider (Latham; Morrison Foerster).
Latent disclosure
Providers must embed a latent disclosure, to the extent technically feasible, in image, video and audio content their system creates. It must convey the provider's name, the name and version of the system, the date and time of creation or alteration, and a unique identifier, either directly or through a link to a permanent website, and be detectable by the provider's tool. SB 1000 extends the duty to content the system alters (other than minor modifications such as cropping, resizing and format changes), changes version number to version information, and adds a field stating whether the system created or altered the content (Latham).
Verification tool
Providers must offer, at no cost, a tool that lets users check whether content was created or altered by the provider's system and outputs any system provenance data detected, with upload, URL and API access (Morgan Lewis). SB 1000 renames it a disclosure verification tool, lets providers point users to a compliant third-party tool, allows output of personal information with express consent, and lets providers limit access to prevent malicious use (Latham; Morrison Foerster).
Manifest disclosure
The August 2 version required providers to give users the option of a visible manifest disclosure. SB 1000 removed that duty, so only latent disclosures are now mandatory (Latham).
Licensees
Under the original text, a provider that learned a licensee had disabled latent disclosures had to revoke the license within 96 hours (Morgan Lewis). SB 1000 replaced revocation with a notice, remediation and reporting process: the provider has 72 hours to terminate the licensee's authorization or notify it, the licensee then has 96 hours to fix the system or stop using it and report back, and an unresolved case goes to the Attorney General (Morrison Foerster).
What SB 1000 changed
SB 1000 is the largest revision so far. It took effect on signature and rewrote the provider duties barely eight weeks after they became operative. The table compares the act as in force on August 2, 2026 (SB 942 as amended by AB 853) with the act after SB 1000, drawing on the Assembly committee analysis, Morrison Foerster and Latham & Watkins.
| Issue | SB 942 / AB 853 (Aug. 2, 2026) | After SB 1000 (Sept. 30, 2026) |
|---|---|---|
| Covered provider | GenAI system with over 1,000,000 monthly visitors or users, publicly accessible in California | Any GenAI system publicly accessible in California; no threshold |
| Content marked | Image, video, audio created by the system | Image, video, audio created or altered by the system, except by minor modification |
| Latent disclosure fields | Provider name; system name and version number; time and date; unique identifier | Same, with version information in place of version number; plus whether the system created or altered the content; from Jan. 1, 2029, whether it is assistive technology |
| Durability and standards | Permanent or extraordinarily difficult to remove, to the extent technically feasible | Also compliant or interoperable with widely recognized industry standards; provenance data redefined by reference to standards bodies |
| Manifest (visible) disclosure | Providers must offer users the option | Requirement deleted |
| Verification tool | "AI detection tool," free, upload, URL and API access | "Disclosure verification tool"; same core features; may direct users to a compliant third-party tool |
| Personal data in tool output | Tool may not output personal provenance data | May output personal information only with express consent after notice; strict limits on collecting or reusing personal information |
| Access limits on the tool | Reasonable limits to protect system security or integrity | Also limits to prevent misuse for malicious purposes |
| Licensees | Contract must preserve disclosures; revoke license within 96 hours of learning they were disabled | Notice in license terms; 72 hours to terminate or notify; licensee has 96 hours to cure and report; report failures to the Attorney General; no duty to monitor |
| Assistive technology | No special rule | Exempt until Jan. 1, 2029; $50,000 penalty per violation for a false claim |
| Entertainment exemption | Exclusively non-user-generated video game, TV, streaming, movie or interactive experiences | Narrowed to exclusively non-user-generated video games, per Morrison Foerster |
| Pending lawsuits | No provision | Suits filed before Sept. 30, 2026 cannot proceed if the conduct is lawful under the amended act |
The author's stated reason was that provenance technology had moved since 2024: the leading embedding methods were not accurately described by the original text, some marking methods could not yet survive compression, spoofing or format conversion, and visible labels risked implying that unlabeled content is authentic (Assembly committee analysis). SB 1000 did not change the dates AB 853 set for platforms, hosting platforms and capture devices (Morrison Foerster).
What AB 853 adds for platforms
AB 853 extended the act beyond model makers (Orrick):
- Large online platforms. Public-facing social media, file-sharing and mass messaging platforms and stand-alone search engines that distribute content to non-creators and exceeded 2 million unique monthly users in the prior 12 months must, from January 1, 2027, detect provenance data that complies with widely adopted standards, provide an interface showing whether system provenance data indicates the content was generated or substantially altered by AI or captured by a capture device, and let users inspect that data.
- GenAI hosting platforms. From January 1, 2027, websites and marketplaces that make GenAI systems available may not knowingly offer a system that lacks the required latent disclosures (Morgan Lewis).
- Capture device manufacturers. From January 1, 2028, makers of cameras and other recording devices first produced for sale in California on or after that date must give users the option to embed a latent disclosure with the manufacturer, device name and version, and time and date.
AB 853 also barred large online platforms from knowingly stripping standards-compliant system provenance data or digital signatures. AB 2713, signed in 2026, revises these duties: a large online platform can meet the inspection duty by displaying provenance data in its interface, linking to a first- or third-party display, or letting users download the data, and it must not knowingly strip provenance data or digital signatures from content uploaded to, distributed on or downloaded from the platform, to the extent technically feasible (Latham). It takes effect January 1, 2027.
Platforms and capture devices: what applies in 2027 and 2028
None of these duties is in force yet, and SB 1000 left their dates alone (Morrison Foerster).
January 1, 2027: large online platforms and hosting platforms
A large online platform must detect standards-compliant provenance data in content it distributes, show in its interface whether system provenance data is available, and let users inspect it. Under AB 2713 the inspection duty can be met by displaying the data in the platform's own interface, linking to a first- or third-party site or app that displays it, or letting users download it in a format that cannot easily be attached to unrelated content (Latham & Watkins). GenAI hosting platforms may not knowingly offer a system that does not place the required latent disclosures.
AB 2713 also specifies that platforms need not act on provenance data or digital signatures that do not comply with widely adopted specifications from an established standards-setting body (Legislative Counsel's digest of AB 2713). That makes the choice of standard decisive: marks that are not standards-compliant need not be read or displayed.
January 1, 2028: capture device manufacturers
The duty covers cameras, phones with built-in cameras or microphones, voice recorders and similar devices first produced for sale in California on or after January 1, 2028; devices already on sale are outside it (Orrick). Morgan Lewis reads the provision as requiring the disclosure capability by default (Morgan Lewis).
Effective dates and later phases
The dates now stand as follows. Provider duties became operative August 2, 2026 (Morgan Lewis). SB 1000's amendments took effect on signature, September 30, 2026, and do not move the AB 853 dates: January 1, 2027 for large online platforms and GenAI hosting platforms, and January 1, 2028 for capture device manufacturers (Morrison Foerster). GenAI systems designed primarily to function as assistive technology for people with disabilities are exempt until January 1, 2029, when a new latent disclosure field identifying such systems also begins (Latham).
August 2, 2026 is also the general date of application in Article 113 of the EU AI Act, Regulation (EU) 2024/1689, although the EU has since adjusted timing for some duties. For how the EU handles labeling and marking of AI content, see our page on the EU AI Act content rules.
Compliance checklist for providers
Each step below is tied to a duty in the act as amended by SB 1000. The sources are the Assembly committee analysis and the client alerts from Morrison Foerster and Latham & Watkins. It is a working list, not legal advice.
- Confirm coverage. Apply the three-part test above; with no user threshold, any publicly accessible system that outputs image, video or audio is in scope unless exempt.
- Map every output path. List each feature that creates or alters image, video or audio. Alterations count unless they are minor modifications such as color changes, sharpening, resizing, cropping, format conversion or audio denoising.
- Write the latent disclosure. Embed provider name, system name and version information, time and date, a unique identifier, and whether the system created or altered the content, directly or via a permanent link. Plan for the assistive technology field due January 1, 2029.
- Make it durable and standard. Permanent or extraordinarily difficult to remove, compatible with your tool, and compliant or interoperable with widely recognized industry standards.
- Ship the disclosure verification tool. Free, with upload, URL and API access, outputting any system provenance data detected; or direct users to a compliant third-party tool reachable from your interface.
- Limit personal data. No personal information in tool output without express consent after notice, no collection or reuse beyond what is strictly necessary, and no conditioning access on extra personal information.
- Collect feedback on the tool's efficacy and use it to improve the tool.
- Update licenses. Put the act's obligations in license terms and be ready to act within 72 hours when an identifiable licensee breaks compliance, then track its 96-hour cure and report.
- Decide on visible labels. Manifest options built for August 2 are no longer required.
- Check any assistive technology claim. A false claim carries $50,000 per violation, each day a separate violation.
Penalties and enforcement
There is no private right of action. The Attorney General, a city attorney or a county counsel may sue for a civil penalty of $5,000 per violation against a covered provider, large online platform or capture device manufacturer, and each day of violation is a separate violation; courts may also award injunctive relief, fees and costs (Morgan Lewis). SB 1000 keeps that structure and adds a $50,000 per violation penalty, until January 1, 2029, for falsely claiming that a system is primarily assistive technology (Latham).
SB 1000 also includes a safe harbor for pending suits: a civil action filed before its effective date cannot proceed if the conduct alleged does not violate the act as amended (Morrison Foerster). We have found no reported enforcement action under the act as of November 9, 2026.
The per-day structure is what makes the numbers large. As an illustration only, if a single violation continued for 30 days, the $5,000 rate and the each-day rule described by Morgan Lewis would put exposure at $150,000 for that one violation. How courts will count violations, for example per feature or per item of content, has not been tested.
The act sits alongside other California disclosure rules, including the training data law AB 2013. For the wider set of duties to label or detect AI content, see detection mandates explained.
How it compares with EU AI Act Article 50 and C2PA
California's act, Article 50 of the EU AI Act and the C2PA standard are often discussed together, but they are different kinds of instrument. Article 50(2) of Regulation (EU) 2024/1689 requires providers of systems generating synthetic audio, image, video or text to mark outputs in a machine-readable format detectable as artificially generated, and Article 50(4) requires deployers to disclose deepfakes and AI-generated text published to inform the public on matters of public interest. Breaches of Article 50 carry fines of up to EUR 15 million or 3% of worldwide annual turnover under Article 99(4). C2PA, the Coalition for Content Provenance and Authenticity, by contrast publishes a voluntary open technical standard, called Content Credentials, for recording the origin and edits of digital content.
| Feature | California AI Transparency Act | EU AI Act Article 50 | C2PA |
|---|---|---|---|
| Type | State statute | EU regulation, directly applicable | Voluntary industry standard |
| Who is bound | GenAI providers; from 2027 large online platforms and hosting platforms; from 2028 capture device makers | Providers (marking) and deployers (deepfake and public-interest text disclosure) | No one, unless adopted or required by law or contract |
| Content types | Image, video, audio | Audio, image, video and text | Media files and other digital content |
| Machine-readable marking | Latent disclosure with set fields | Machine-readable, detectable marking, as technically feasible | Signed provenance metadata (Content Credentials) |
| Visible labels | Not required since SB 1000 | Required for deployers of deepfakes and public-interest AI text | Not a labeling rule |
| Free verification tool | Required | Not required by Article 50 text | Not applicable |
| Penalties | $5,000 per violation per day | Up to EUR 15 million or 3% of turnover | None |
| Start date | Aug. 2, 2026 (providers) | Aug. 2, 2026 general application date | Not applicable |
| Reach | Systems publicly accessible in California | Providers placing systems on the EU market, wherever established, and output used in the EU | Global, voluntary |
The laws meet through standards. California's act, as summarized by the Assembly committee, refers to widely adopted specifications from an established standards-setting body rather than naming one. Marking built for Article 50 on such a standard may carry much of California's latent disclosure, but California's fixed fields and free tool are separate duties. See our page on the EU AI Act content rules and our analysis of watermarking and C2PA as a compliance floor.
Open questions and pending litigation or rulemaking
As of November 9, 2026, we have found no reported lawsuit challenging the California AI Transparency Act, no reported enforcement action under it, and no reported implementing regulations. The open questions are in the text itself:
- Version information. SB 1000 swaps "version number" for "version information" without saying what more is required (Latham & Watkins).
- Technical feasibility. Most duties apply to the extent technically feasible. The bill's author told the Assembly committee that some marking methods could not yet withstand compression, spoofing, added noise or file conversion (Assembly committee analysis). How much a provider must do before infeasibility excuses it is untested.
- The entertainment exemption. The June 2026 Assembly analysis said SB 1000 no longer changed the exemption (same source); the bill was amended again in August, and Morrison Foerster reads the enacted text as limiting it to non-user-generated video games. Streaming and film services should check the chaptered text.
- Counting violations. The act makes each day a separate violation but no court has yet applied the penalty clause.
Bottom line
If you make a generative AI system that produces images, video or audio and Californians can use it, the act applies to you now, whatever your size: embed latent provenance data with the required fields and offer a free disclosure verification tool (Morrison Foerster). Visible labels are optional since SB 1000, but license terms and licensee processes need updating. Large online platforms and model hosting sites have until January 1, 2027, and device makers until January 1, 2028. Text-only systems remain outside the marking duties.
Since September 30, 2026, any publicly accessible generative AI provider of images, video or audio must embed latent provenance data and offer a free checking tool, because SB 1000 removed the one million user threshold. Violations cost $5,000 each, with each day a separate violation, and there is no private right of action. Large online platforms and hosting platforms take on their duties January 1, 2027.
What changed
- 2026-11-09: Expanded and re-verified against bill histories and committee analyses: added chapter numbers, the covered provider test, an SB 1000 before and after table, a provider checklist, platform and device detail, an EU AI Act and C2PA comparison and open questions. Corrected the AB 2713 description (the no-stripping rule dates from AB 853; AB 2713 revises it), added SB 1000's narrowing of the entertainment exemption and the 72 and 96 hour licensee timelines.
- 2026-09-30: SB 1000 signed and effective immediately: user threshold removed, manifest option dropped, verification tool and licensee rules rewritten. AB 2713 signed, effective January 1, 2027.
- 2026-08-02: Provider requirements became operative.
- 2025-10-13: AB 853 signed: operative date moved to August 2, 2026; platform, hosting and capture device duties added.
- 2024-09-19: SB 942 signed, originally operative January 1, 2026.
Frequently asked questions
What does SB 942 require?
Providers of generative AI systems must embed latent provenance disclosures in AI-generated or altered images, video and audio, and offer a free tool that lets anyone check whether content came from their system. It does not cover text, so text-only chatbots have nothing to mark.
When did the California AI Transparency Act take effect?
The provider requirements became operative on August 2, 2026, after AB 853 delayed the original January 1, 2026 date. SB 1000 amendments took effect September 30, 2026. Platform duties start January 1, 2027 and capture device duties January 1, 2028.
What does AB 853 change?
AB 853 delayed the start date to August 2, 2026 and added duties for large online platforms and GenAI hosting platforms from January 1, 2027 and for capture device manufacturers from January 1, 2028.
Who has to comply with the AI Transparency Act?
Since SB 1000, any person producing a generative AI system that is publicly accessible in California, with no user threshold and regardless of where the company is based. Large online platforms with over 2 million monthly users, GenAI hosting platforms and capture device makers have later obligations.
What is SB 1000?
SB 1000 (Becker, Chapter 861, Statutes of 2026) is the urgency bill that rewrote the act on September 30, 2026. It removed the one million user threshold, dropped the manifest disclosure option, renamed the detection tool a disclosure verification tool, added latent disclosure fields, replaced license revocation with a notice process and created a temporary assistive technology exemption.
What are the penalties under the California AI Transparency Act?
A civil penalty of $5,000 per violation, with each day a separate violation, in suits brought by the Attorney General, a city attorney or a county counsel, plus injunctive relief, fees and costs. Falsely claiming the assistive technology exemption carries $50,000 per violation. There is no private right of action.
Sources
- California Legislative Information: SB 942 (2024)
- California Legislative Information: AB 853 (2025)
- California Legislative Information: SB 1000 (2026)
- California Legislative Information: AB 2713 (2026)
- Assembly Committee on Privacy and Consumer Protection: SB 1000 analysis (June 16, 2026 hearing)
- EUR-Lex: Regulation (EU) 2024/1689 (AI Act)
- Office of the Governor: SB 942 signing release (Sept. 19, 2024)
- Office of the Governor: AI bills signed Sept. 30, 2026
- Morrison Foerster: California Revamps AI Transparency Requirements
- Latham & Watkins: Newest Slate of AI Laws Keeps California at the Forefront
- Morgan Lewis: New California AI Disclosure Rules Become Operative
- Orrick: California delays its AI Transparency Act
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.