SB 942: Coverage, Disclosures, Detection Tool, and Penalties

California Senate Bill 942, the California AI Transparency Act, requires large generative AI providers whose systems are accessible in California to label AI-generated images, video, and audio, run a free public tool that detects content made by their systems, and cut off licensees who disable those labels. The requirements take effect August 2, 2026, after Assembly Bill 853 moved the original January 1, 2026 date. ¹ Violations carry civil penalties of up to $5,000 per day, enforceable by the Attorney General, city attorneys, or county counsel.

Who the Law Applies To

SB 942 reaches any person or company that creates a generative AI system meeting two conditions: the system has more than one million monthly visitors or users, and it is publicly accessible within California. ² Headquarters location is irrelevant. A tool built in another state or country still qualifies if California residents can use it.

The million-user threshold is counted globally, not just within California. ³ A provider cannot sidestep coverage by pointing to a mostly foreign user base. The statute uses “monthly visitors or users” without specifying how to measure them.

What Content Is Covered

The labeling rules apply to AI-generated images, video, audio, and combinations of those formats. ³ Text-only output is not covered. A chatbot producing a written answer falls outside the disclosure mandate; the same system producing a synthetic voice clip or AI image does not.

Section 22757.5 exempts products, services, websites, and applications that provide exclusively non-user-generated video game, television, streaming, movie, or interactive experiences. ² A visual effects studio using AI inside a feature film is not required to embed provenance data in every frame. A social media platform that lets users generate AI images is.

The Two Disclosure Requirements

Latent Disclosures (Mandatory)

Every piece of covered AI-generated content must carry a latent disclosure: technical data embedded in the file’s metadata that software can read even though a viewer cannot see it. To the extent technically feasible and reasonable, that data must convey the following, either directly or through a link to a permanent website:

The provider’s name.
The name and version of the generative AI system that produced the content.
The time and date the content was created or altered.
A unique identifier for that piece of content.

The latent disclosure must be detectable by the provider’s own detection tool, consistent with widely accepted industry standards, and permanent or extraordinarily difficult to remove to the extent technically feasible. ³ The statute does not name a specific technical framework, leaving providers room to pick a standards-compliant approach.

Manifest Disclosures (Offered as an Option)

Manifest disclosures are the visible counterpart: watermarks, overlays, or audible markers that a person can perceive. The law requires providers to give users the option to add a manifest disclosure to content they create or alter with the system. ³ This differs from the latent requirement: the hidden metadata is mandatory on every covered output, while the visible label must be offered rather than forced on every piece.

When a user turns it on, the manifest disclosure must identify the content as AI-generated, be clear and conspicuous, suit the medium, and be understandable to a reasonable person. It must also be permanent or extraordinarily difficult to remove, to the extent technically feasible. ³

The Free Detection Tool

Each covered provider must build and maintain a free, publicly accessible AI detection tool. Anyone can upload a piece of digital content and receive a response indicating whether the provider’s system created or altered it. The tool has to be available through a URL on the provider’s website and, if the provider has one, through its mobile application. ²

Providers may not collect or retain personal information from people who submit content for analysis, and they may not retain personal provenance data from uploaded content, except in narrow circumstances. ³ The tool operates as a check on content whose visible labels have been stripped: a journalist, researcher, or ordinary user can still upload the file and see whether the hidden metadata matches a known provider.

Obligations Toward Third-Party Licensees

If a covered provider licenses its AI system to another company and later discovers the licensee has modified the system so it can no longer embed the required disclosures, the provider must revoke the license within 96 hours. ³ Once the license is revoked, the licensee must stop using the system entirely.

If a licensee keeps running a revoked system, the Attorney General, a city attorney, or a county counsel can sue for injunctive relief and reasonable attorney’s fees. The $5,000 per-violation penalty that applies to covered providers does not apply to licensees in this scenario. ³ The 96-hour deadline still creates pressure for the provider, because overlooking a licensee’s modifications becomes the provider’s own compliance problem.

Penalties and Who Can Enforce

Enforcement authority is shared. The Attorney General, city attorneys, and county counsel can each bring a civil action against a covered provider. ³ A city attorney in San Francisco or Los Angeles can act independently of the state.

The civil penalty is $5,000 per violation, and each day of noncompliance counts as a separate violation. ³ A prevailing plaintiff can also recover reasonable attorney’s fees and costs. Enforcement officials may seek injunctive relief as well, including orders that force a provider to deploy a compliant detection tool, embed the required latent disclosures, or stop operating its AI system in California until it does. ²

1
California Legislative Information. California Business and Professions Code 22757
2
California Legislative Information. SB-942 California AI Transparency Act
3
LegiScan. California SB942 2023-2024 Regular Session Chaptered