California made large AI providers offer a free detection tool

The conversation

California's Governor signed Senate Bill 942, the AI Transparency Act, on 19 September 2024; the bill's own history record shows it was chaptered by the Secretary of State the same day as Chapter 291, Statutes of 2024. The law does not regulate a chatbot's conversation directly. It regulates a different kind of disclosure: whether a person looking at an image, video, or audio clip can find out, using a free tool, whether a large generative AI system made or altered it.

What the documents show

The bill's own text applies to a "covered provider," defined as a person or company that creates a generative AI system with more than 1,000,000 monthly visitors or users. A covered provider must offer, at no cost and publicly accessible, an AI detection tool letting a user assess whether content came from that provider's system, and the tool must output system-provenance data and support API access. The statute separately requires a "latent disclosure" embedded in AI-generated image, video, or audio content, conveying the provider's name, the system's name and version, and a creation timestamp, detectable by the provider's own tool; a human-visible "manifest disclosure" remains optional for the provider to offer. The text states the chapter "shall become operative on January 1, 2026."

The system boundary

SB 942 is a companion measure to, but textually separate from, California's earlier bot-identity statute, Senate Bill 1001, which requires a bot to disclose it is not a person during a commercial or election-related conversation. SB 942 does not require a chatbot to announce itself mid-conversation; it requires the system's provider to make detectable, machine-readable provenance data available for generated content and to offer a public checking tool. A builder whose product only generates conversational text, with no image, video, or audio output, sits outside SB 942's detection-tool and latent-disclosure duties even if the separate bot-identity duty still applies.

Where it fails

The statute's own operative date, 1 January 2026, means the detection-tool and disclosure duties took effect only after this entry's retrieval date, so a builder checking compliance today should verify against the current, in-force text rather than the 2024 chaptered version alone. The 1,000,000-monthly-user threshold also means a smaller provider's generative outputs carry no equivalent disclosure duty under this particular statute.

• Does our generative AI system cross the 1,000,000 monthly visitor or user threshold that makes us a covered provider?
• If we generate images, video, or audio, can our own detection tool read back the latent disclosure we are required to embed?
• Have we confirmed separately whether the state's bot-identity disclosure duty also applies to our product, since the two statutes address different conduct?

Builders should treat this summary as a starting point for legal review of the current, in-force text, not a substitute for it.