California AI Transparency Act (SB 942), explained

Timeline

1. September 19, 2024Governor Gavin Newsom signed SB 942 (Chapter 291). Its original operative date was January 1, 2026.
2. October 13, 2025AB 853 (Chapter 674) moved the operative date to August 2, 2026 and added duties for large online platforms, model hosting sites and capture-device makers.
3. August 2, 2026Duties for AI providers became operative.
4. September 30, 2026SB 1000 (Chapter 861) was signed and took effect immediately as an urgency statute. AB 2713 (Chapter 856), signed the same day, rewrote the large-platform duties.
5. January 1, 2027Large online platform and model hosting duties begin.
6. January 1, 2028Capture-device duties begin for devices first produced for sale in California on or after this date.
7. January 1, 2029The exemption for systems designed primarily as assistive technology ends.

What SB 1000 changed

SB 1000 is an urgency statute, so its changes applied from the day it was signed, September 30, 2026, less than two months after the provider duties became operative. Guides written before then describe the old rules.

Who is covered

A covered provider is a person that creates, codes or otherwise produces a generative AI system that is publicly accessible within California. Since SB 1000 there is no user, revenue or size threshold, so smaller companies with a public generative AI product are covered alongside the largest model developers.

A generative AI system is AI that can generate derived synthetic content, including text, images, video and audio, that emulates the structure and characteristics of its training data. The Act's tool and disclosure duties apply to image, video and audio content.

The disclosure verification tool

• Free to use and publicly accessible, though a provider may limit access for security, integrity or to prevent misuse.
• Lets anyone check whether image, video or audio content was created or altered by the provider's system, other than by minor modifications.
• Outputs the content's system provenance data. It may output personal information only with the affected person's express consent.
• Accepts uploads or URLs and supports an API, so it can be used without visiting the provider's website.
• A provider may instead direct users to a third-party tool that meets the same requirements, works with its latent disclosures and is clearly and conspicuously reachable from its system's interface.
• Providers must collect feedback on the tool, and may not collect, use, retain, sell or share personal information from tool users or the content they check beyond what is strictly necessary.

Latent disclosures

To the extent technically feasible, image, video and audio content created or altered by the system (other than by minor modification) must carry a latent disclosure that conveys:

• the covered provider's name;
• the system's name and version;
• the time and date of creation or alteration;
• a unique identifier;
• whether the system created or altered the content; and
• from January 1, 2029, whether the system is designed primarily as assistive technology.

The disclosure must be permanent or extraordinarily difficult to remove or tamper with, compatible with the provider's verification tool, and compliant or interoperable with widely recognized industry standards, such as the C2PA Content Credentials specification.

Licensees

Providers that license their system must tell licensees about their obligations. If a provider knows that an identifiable licensee has modified the system so it no longer complies, the provider must, within 72 hours, either terminate the licensee's authorization or notify the licensee. The licensee must then bring the system into compliance or stop using it, and report back within 96 hours. Unresolved cases are reported to the Attorney General.

Platforms, model hosts and device makers

Large online platforms, from January 1, 2027. Public-facing social media, file-sharing and mass messaging platforms and stand-alone search engines with more than 2,000,000 unique monthly users (broadband and telecommunications services are excluded). Under AB 853 as rewritten by AB 2713, they must:

• detect provenance data embedded in, attached to or associated with content they distribute;
• show, clearly and conspicuously, whether provenance data or a digital signature identifies content as AI-generated, substantially AI-altered or captured by a device;
• let users inspect the system provenance data through the interface, a link or a download; and
• not knowingly strip provenance data or digital signatures, to the extent technically feasible.

Platforms need not maintain, display or allow download of personal information in provenance data, and have no duties for data that isn't compliant or interoperable with widely adopted standards.

Model hosting sites, from January 1, 2027. A website or app that makes generative AI source code or model weights available for download may not knowingly make available a system that doesn't place the required disclosures.

Capture-device makers, from January 1, 2028. For cameras, phones with cameras or microphones, and voice recorders first produced for sale in California on or after that date, manufacturers must offer a latent disclosure of the maker, device name and version, and time and date, embedded by default, to the extent technically feasible and compliant with widely adopted specifications.

Penalties and enforcement

• A civil penalty of $5,000 per violation, with each day of violation a separate violation.
• Enforced by the Attorney General, city attorneys and county counsel, who can also recover attorney's fees and costs. The Act gives individuals no right to sue.
• Falsely claiming that a system is designed primarily as assistive technology carries $50,000 per violation, each day separate, until January 1, 2029.

Exemptions and limits

• Until January 1, 2029, the Act does not apply to generative AI systems designed primarily as assistive technology for people with disabilities.
• It does not apply to products or services that offer exclusively non-user-generated video games.
• The disclosure and tool duties cover image, video and audio, not text.
• Minor modifications, such as brightness, contrast or color changes, sharpening, resizing, cropping, format conversion or background-noise removal, don't count as altering content.

Checklist

• Check whether you produce a generative AI system that is publicly accessible in California and creates or alters images, video or audio. Size no longer matters.
• Embed latent disclosures with every required field, using a widely recognized provenance standard.
• Offer the free verification tool with an API, or link a compliant third-party tool from your interface.
• Update license terms and set up the 72-hour process for licensees that disable disclosures.
• Platforms and model hosting sites: prepare for January 1, 2027. Device makers: January 1, 2028.

For how other states and the EU handle AI disclosures, see AI transparency and disclosure laws by state, the AI disclosure tracker and our EU AI Act guide. For California's deepfake rules, see California deepfake laws.

This explains the law in plain language and is not legal advice. AI Laws by State reviews this page against the sources above and updates it when the law or its enforcement changes.