California's AI labeling law takes effect
SB 942 requires large AI providers to watermark generated content and offer a free detection tool, starting the same day the EU's own AI-labeling rule begins.
California’s AI Transparency Act, Senate Bill 942, becomes operative today, August 2. It was originally due January 1, 2026, but Assembly Bill 853, signed in October 2025, pushed the date back seven months to line up with the day the EU AI Act’s own content-transparency rule, Article 50, starts being enforced.
The law covers any generative AI provider whose system draws more than one million monthly users or visitors and is publicly accessible in California. It requires two kinds of disclosure: a visible “manifest” mark, such as a watermark, and a hidden “latent” one, metadata embedded in the file that’s built to survive basic edits. Covered providers also have to publish a free, public tool anyone can use to check whether image, video, or audio content came from their system.
What changes today
AI-content labeling in the US has mostly been voluntary until now. SB 942 makes it a standing legal requirement for large providers serving Californians. AB 853 also set up a second deadline: starting January 1, 2027, the same duties extend to platforms that host generative AI content, plus separate obligations for large online platforms and capture-device makers.
What it means for operators
If your generative AI feature reaches more than a million monthly California users, you’re covered as of today. Check whether your visible mark survives a re-encode or screenshot, and whether your free public detection tool actually works, both are explicit requirements, not just watermarking in principle. The EU’s parallel Article 50 transparency guidance landed the same day; see our guide to what changed under both rules at once for where the two actually diverge.