California’s law requiring companies to label AI-generated content took effect on August 2, becoming one of the first state statutes anywhere to impose a technical disclosure standard on deepfake-capable image, video, and audio generators.
What covered companies must do
Under the California AI Transparency Act — enacted as SB 942 and later amended by AB 853 — any generative AI provider with more than one million monthly users in California must offer the public a free tool that can check whether a specific image, video, or audio clip was produced by that company’s system. Covered systems must also embed a hidden “latent disclosure” carrying the provider’s name, the AI system and version used, and a creation timestamp directly into generated content, and give users the option to add a visible label as well. Companies that license their AI systems to others must cut off any licensee found disabling these disclosures within 96 hours. Violations carry civil penalties of $5,000 per day, enforceable by the California Attorney General and local prosecutors, according to the bill text.
Timed to match the EU
Lawmakers pushed the law’s original January 2026 start date to August 2 specifically to align with when the EU AI Act’s rules for high-risk systems begin taking effect, giving multinational AI companies a single global compliance window instead of staggered deadlines. Two further obligations are still to come under the amended law: generative AI hosting platforms and large online platforms must start detecting and surfacing provenance data by January 1, 2027, and makers of AI-enabled capture devices, such as cameras, must add disclosure options by January 1, 2028.
Part of a wider state patchwork
California joins a small but growing list of states writing their own AI rules ahead of any federal standard, following Colorado’s AI consumer-protection law, which took effect earlier this year. The push has been driven largely by concern over undetectable synthetic media in elections, scams, and harassment, and California’s detection-tool mandate is among the most technically prescriptive of any state AI law passed so far — building on the same content-provenance approach used by the C2PA standard.