California's AI Transparency Act took effect, and it asks for a detection tool
SB 942, as amended by AB 853, applies from 2 August to generative systems with more than a million monthly visitors. Covered providers must offer a free detection tool and embed latent disclosure in what they generate.
The California AI Transparency Act became operative on 2 August. It reaches "covered providers" — developers of publicly accessible generative systems with more than one million monthly visitors — which is a threshold written to catch the large consumer products and leave everyone else alone.
Two obligations do the work. A covered provider must make an AI detection tool available at no cost, and must include latent disclosure in generated content, with manifest disclosure offered as a user option. Latent means embedded in the file where a machine can read it; manifest means visible to a person looking at it.
The free-detection-tool requirement is the unusual one. It obliges a company to ship the thing that identifies its own output — a capability most providers had treated as optional, and some as commercially awkward. It also moves the burden of provenance from the reader to the producer, which is the correct direction and the expensive one.
What the Act does not do is govern training. It is a rule about what a deployed system must disclose. Set beside the two dozen bills that moved through appropriations last week — and the training-data disclosure bill that did not — a consistent legislative preference is visible: California is regulating the output, not the input.
AB 2713 and SB 1000, both still moving, would adjust the Act's provenance-data requirements. A statute amended twice before its first birthday is a statute whose authors are still learning what they wrote.
Latham & Watkins — California Assumes Role as Lead US Regulator of AI → · Transparency Coalition — AI Legislative Update: August 14, 2026 → · Digital Democracy — SB 1000: California AI Transparency Act →