California's AI Transparency Act.
California SB 942 — the AI Transparency Act — requires large generative-AI providers to make AI-generated content detectable and disclosed. Amended by AB 853, it becomes operative on August 2, 2026. Here's what it actually says, and how an independent audit helps you be ready.
At a glance
- Signed Sept 19, 2024; amended by AB 853 (Oct 13, 2025)
- Operative date now August 2, 2026 (delayed from Jan 1, 2026)
- Penalty: $5,000 per violation, per day
- Enforced by the CA Attorney General, city or county counsel
The standard, accurately
SB 942 covers large "covered providers" — those that create a generative AI system with over 1,000,000 monthly users that is publicly accessible in California. The substantive disclosure duties apply to AI-generated image, video and audio content (text is in the definition but outside the disclosure obligations).
Covered providers must offer a free, public AI-detection tool (with API access) that reports system provenance data while excluding personal data; provide an optional manifest (visible) disclosure labeling content as AI-generated; and embed a latent disclosure in covered content carrying the provider name, system name and version, a timestamp and a unique identifier — "permanent or extraordinarily difficult to remove."
There are also licensee provisions: covered providers must contractually require licensees to keep the latent-disclosure capability, and must revoke a license within 96 hours of discovering a licensee disabled it.
AB 853 pushed the effective date to August 2, 2026 and added staggered later dates — large online platforms (~2M users) from Jan 1, 2027 and capture-device makers from Jan 1, 2028. Governor Newsom invited further 2026 follow-up legislation before it takes effect, so details may still shift.
Why this lands on your desk
Most enterprises are not "covered providers" — the 1M-California-user threshold targets large model providers. But the law reaches you through vendor contracts: expect clauses requiring you not to strip watermarks or provenance, with the 96-hour revocation exposure if you do.
If you operate a large online platform (~2M users), you take on direct duties from Jan 1, 2027 — detect and display provenance data, let users inspect authenticity, and not knowingly remove provenance where technically feasible.
Even as a plain deployer, avoid pipelines that silently strip metadata or watermarks from AI-generated assets — that can breach license terms. Downstream-deployer liability is not fully settled, so a documented, good-faith posture is the safe play.
Readiness you can evidence
An audit can't issue legal certification — but it can verify concrete, checkable artifacts that map to the standard.
Disclosure-mechanism check
Verify AI-generated image/video/audio carries manifest disclosures and that latent disclosures embed the required fields (provider, system + version, timestamp, unique identifier).
Provenance / labeling review
Confirm provenance metadata (e.g. C2PA-style signatures) is embedded, survives your processing and export pipeline, and is not silently stripped downstream.
Vendor-contract posture
Review GenAI vendor agreements for the required flow-down terms and the 96-hour revocation clause, and confirm you are not disabling disclosure features.
Readiness documentation
Maintain a dated evidence trail (technical-feasibility assessments, config records, vendor representations) demonstrating good-faith readiness ahead of August 2, 2026.
Common questions
Is SB 942 in force right now?
Not yet. It was signed in September 2024 with a January 1, 2026 date, but AB 853 (October 2025) delayed the core obligations to August 2, 2026. As of 2026 it is not yet operative, and further follow-up legislation was invited before it takes effect.
Does my company have to comply if we just use AI tools?
Usually not directly — the obligations fall on "covered providers" with over 1,000,000 monthly California users. But you may inherit contractual duties from your AI vendors (keep disclosures intact), and large online platforms take on direct duties from 2027.
Does it apply to AI-generated text?
The substantive disclosure obligations apply to image, video and audio — not text — even though text appears in the statutory definition of a generative AI system.
Can an iDharma audit certify SB 942 compliance?
No — we don't issue legal certification, and no one can. We verify concrete, checkable artifacts (disclosure mechanisms, provenance durability, vendor-contract posture) and give you documented evidence of readiness mapped to the statute.
Informational only, reflecting our understanding of SB 942 and AB 853 as of 2026; not legal advice. California invited further 2026 amendments before the August 2, 2026 effective date — confirm current requirements with qualified counsel.
Turn a moving target into a documented posture
An independent audit gives you evidence you were ready — useful as rules shift and as buyers ask.
Request an AI audit