> ## Content Index
> Fetch the complete content index at: https://www.intercepta.ai/blog/llms.txt
> Use this file to discover other available public pages before exploring further.

# The California AI Transparency Act Is Live. Here Is the Layer It Leaves Open.
- URL: https://www.intercepta.ai/blog/california-ai-transparency-act-sb-942/
- Published: 2026-08-06T12:14:45.000Z
- Updated: 2026-08-06T12:14:45.000Z
- Description: California's AI Transparency Act (SB 942) is operative from 2 August 2026. It settles AI-content provenance, not whether that content is compliant.
- Author: Edward Sweigart
- Tags: AI Compliance, Marketing Compliance, Regulatory, California

*The California AI Transparency Act became operative on 2 August 2026\. It settles who must mark AI-generated image, video or audio content with provenance data and make a detection tool publicly available. It leaves open whether that content complies with the rules governing what it may claim.*

The Act, passed in 2024 as Senate Bill 942 (SB 942) and amended in 2025 by Assembly Bill 853 (AB 853), applies to covered providers, meaning the makers of large generative-AI systems that reach more than one million monthly users or visitors and are publicly accessible in California. From 2 August 2026, those providers must offer, at no cost to the user, a public tool that allows anyone to test whether image, video or audio content came from their system, must embed machine-readable provenance data in that content, and must give users the option to attach a visible disclosure. Obligations extend to large online platforms and generative-AI hosting platforms from 1 January 2027, and to capture-device manufacturers from 1 January 2028.

The 2 August date was chosen deliberately. Legislators aligned it with the provenance timeline in Article 50 of the European Union’s AI Act, so that content-origin obligations in California and Europe would apply on a common schedule. The alignment was built around that common date, although on the European side it has since been partially eased. A provisional agreement reached in May 2026 gives generative-AI systems already on the market before 2 August 2026 until 2 December 2026 to meet the machine-readable marking requirement under Article 50(2). The framework also continues to change on the California side. As of publication, two amending bills, SB 1000 and AB 2713, are moving through the California legislature. SB 1000 would remove the one-million-user threshold and replace the detection tool with a disclosure verification tool, while AB 2713 would tighten how large online platforms handle content provenance, including barring them from knowingly stripping provenance data or digital signatures. Both must pass by 31 August 2026 or lapse when the session ends. Covered providers should treat the enacted text as the operative baseline while tracking these changes.

![A two-track timeline aligning the California AI Transparency Act with Article 50 of the European Union AI Act. On the California track, covered providers, meaning generative-AI systems with over one million users, commence on 2 August 2026; large online platforms and hosting platforms on 1 January 2027; and capture-device manufacturers on 1 January 2028. On the European track, provenance-marking obligations apply on the same common date of 2 August 2026, with existing systems given a provisional extension to 2 December 2026. A note states that enacted phases are shown, and that the amending bills Senate Bill 1000 and Assembly Bill 2713 move through the 2026 session and are tracked separately.](https://www.intercepta.ai/blog/content/images/2026/08/California-mockup_2_timeline.png)

California's commencement dates align with Article 50 of the European Union AI Act on a common date of 2 August 2026.

## Provenance and Compliance Are Different Layers

Provenance data records how a piece of content was made. It confirms whether the content was generated or altered by a generative-AI system, and it makes the fact detectable, and visible where the user elects. The data carries no information about whether the material is accurate, whether its claims are substantiated, or whether it meets the rules that govern what may be said in a regulated market.

Consider a financial promotion produced with a generative-AI tool. If that tool is a covered provider, the mandatory provenance duties set out above apply to its output. None of that addresses whether the promotion is fair, clear and not misleading, or whether communicating it is permitted under the invitation or inducement test in section 21 of the Financial Services and Markets Act 2000\. A watermark attests to the content’s origin. The regulatory obligations attach to what it says.

![Two stacked panels comparing the same AI-generated image, video or audio asset across two layers. The upper panel, headed What the Act settles, answers whether the content is AI-made and shows three provenance duties: embedded provenance data, which is mandatory; a free public detection tool; and visible disclosure at the user's option. The lower panel, headed What it leaves open, answers whether the content is compliant and shows three obligations not addressed by the Act: fair, clear and not misleading; permission to communicate under section 21 of the Financial Services and Markets Act; and claims, risk warnings and disclosures. A dividing line reads that a watermark records origin and the obligations attach to what it says.](https://www.intercepta.ai/blog/content/images/2026/08/California-mockup_1_two_layer.png)

The same asset seen through two layers: the Act settles how it was made, not whether what it says is compliant.

## The Stakes for Compliance and Marketing

Provenance is handled upstream, by the generative-AI tool or the hosting platform. The substantive compliance obligation stays with the organisation that publishes the asset. The split extends beyond California, because many widely used generative-AI tools meet the covered-provider test, and the provenance data is present in content produced with them wherever the business is based. An organisation in London or New York that generates image, video or audio marketing assets with such a system receives them with that data already embedded. Since each covered provider must publish a free public tool that reads its own outputs, the AI origin of an asset is often detectable by an outsider, provided the embedded data has survived and they use the relevant provider’s tool. The two functions are affected differently.

![A branching diagram splitting one AI-generated marketing asset into two columns. The Compliance column, headed external detectability creates exposure, notes that AI origin is often detectable by outsiders such as regulators, competitors, journalists and claimants; that unreviewed promotions are now identifiable at scale; that detection and disclosure are not review; and that whatever goes unreviewed accumulates as compliance debt. The Marketing column, headed a label is not permission, notes that a mark or label discloses machine origin only; that it says nothing about substantiation, risk warnings or permission to publish; that a label can draw attention to an asset rather than protect it; and that the substantive obligation must still be met before publication.](https://www.intercepta.ai/blog/content/images/2026/08/California-mockup_3_stakes.png)

One detectable, unreviewed asset creates different exposure for compliance and for marketing.

For the compliance department, that outside visibility is what matters. Any such asset can be tested by a regulator, a competitor, a journalist or a claimant, without the publisher’s involvement, wherever that provenance data is intact. The consequence is exposure. A promotion that was never reviewed against the marketing rules is now identifiable as AI-generated, which invites exactly the external scrutiny under which an unreviewed gap is likely to be found, and found across many assets at once. Detection and disclosure do not amount to review. The volume of AI-generated content entering the content library still must clear the same rules it always did, and whatever goes unreviewed accumulates as [compliance debt](https://www.intercepta.ai/blog/the-definitive-guide-to-compliance-debt/) that surfaces later.

For marketing teams, the assurance a sophisticated tool seems to offer is misplaced. A model that marks its output, or offers to label it, has disclosed that the asset is machine-made. It has said nothing about whether the claim is substantiated, whether the risk warning is present, or whether the promotion may be published at all. A label can draw attention to an asset rather than protect it. The substantive obligation must still be met before publication, however the asset was produced.

## The Open Layer Is Yours to Close

California and the EU have put content-provenance obligations on a shared timetable. Each settles whether content was AI-made and leaves the open question, whether it complies, with whoever publishes it. As the volume of AI-generated marketing content grows, that obligation applies continuously rather than occasionally.

Intercepta AI validates marketing content against 1,100+ regulatory rules across the United States, United Kingdom, European Union, Canada, Australia and New Zealand. A single scan covers every jurisdiction the content reaches and returns a documented report that maps each issue to the specific regulation it references, with remediation guidance for your compliance team to review. Marketing teams validate before publication. Compliance teams hold a documented regulatory record. Every decision on every finding stays with your compliance team.

**AI built the content. AI should validate it against the rules that govern it, for your compliance team to review.**

[Your first three scans are free](https://intercepta.ai/?ref=intercepta.ai)

*No payment method required.*

## Sources

- [SB 942, California AI Transparency Act (Chapter 291, Statutes of 2024), California Legislative Information, signed 19 September 2024](https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill%5Fid=202320240SB942&ref=intercepta.ai)
- [AB 853, California AI Transparency Act (Chapter 674, Statutes of 2025), California Legislative Information, signed 13 October 2025](https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill%5Fid=202520260AB853&ref=intercepta.ai)
- [California Business and Professions Code, Division 8, Chapter 25 (commencing with Section 22757), enacted statutory text](https://leginfo.legislature.ca.gov/faces/codes%5FdisplayText.xhtml?division=8.&chapter=25.&lawCode=BPC&ref=intercepta.ai)
- [SB 1000, California AI Transparency Act (2025 to 2026 Regular Session), California Legislative Information, status verified 5 August 2026](https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill%5Fid=202520260SB1000&ref=intercepta.ai)
- [Regulation (EU) 2024/1689, EU AI Act, Article 50, transparency obligations](https://eur-lex.europa.eu/eli/reg/2024/1689/oj?ref=intercepta.ai)