California AI Transparency Act Now in Effect: AI Products with 1M+ Users Must Watermark Content
One day after the EU AI Act's transparency provisions started biting, California activated its own AI content labeling law on the same date — SB 942 is now enforceable.
Three Key Takeaways
Who is covered? Any AI product with over a million monthly users in California. The California AI Transparency Act (CAITA), formed by SB 942 and its amendment AB 853, was originally set for January 1, 2026. AB 853 pushed the date to August 2, deliberately synchronizing with the EU AI Act transparency timeline. The law applies to every generative AI provider with more than one million monthly users accessible in California — OpenAI, Google, Meta, Anthropic, Midjourney, and every other major AI company is squarely in scope.
Three hard requirements: watermarks, detection, labels. Covered providers must now do three things: first, embed machine-readable latent watermarks in AI-generated images, video, and audio, with the technical standard pointing to the C2PA (Coalition for Content Provenance and Authenticity) specification — the same framework behind the EU AI Act Article 50 watermarking requirement; second, offer a free public detection tool that lets anyone check whether a piece of content was AI-generated; third, give users the option to add a visible label to AI-generated content.
Federal level: crickets. In a striking contrast, the federal government quietly blew past the August 1 deadline set by Executive Order 14409 for frontier AI regulation — neither NIST nor CISA published any formal documents, and the definition of "covered frontier model" remains unresolved. This means that for the foreseeable future, California's law is effectively the most binding AI content labeling rule on U.S. soil.
WangDou's Take
The global AI regulation timeline has finally synced up: the EU and California drew their swords on the same day, while the federal government did not even turn in the homework. The result: a single state's law becomes the de facto AI labeling standard for the entire United States. This is not the first time — California's CCPA walked the exact same path for privacy. For AI companies, compliance costs are manageable on paper. The real headache is the engineering: C2PA metadata has to survive screenshots, compression, and cross-platform resharing, yet most social platforms strip metadata on upload. The law reads well; whether it actually works in production will become clear in the first few months after August 2.