Starting August 2, ChatGPT, Gemini, Claude and other major AI models are legally required to label everything they generate for users in California. Senate Bill 942, as amended by AB 853, took effect on a date deliberately chosen to line up with the day the European Union starts enforcing its own transparency rules for high-risk AI systems.
The rule applies to any generative AI service with more than one million users in California — which covers essentially the whole major industry: OpenAI, Google, Meta, Anthropic, xAI and Microsoft. Providers now have to do three things. Offer the public a free tool that can check where a file came from. Let users attach a visible disclosure to an AI-generated image, video or audio clip. And embed hidden metadata inside the file itself — provider name, system version, creation timestamp, a unique identifier. Companies that simply license someone else's model can't dodge the rule either; the obligation flows down to them.
The label has to be conspicuous and, in the law's own wording, "extraordinarily difficult to remove." Each violation carries a $5,000 fine, and every day a company stays out of compliance counts as a separate one, so the number climbs fast. State Senator Josh Becker, who wrote the bill, frames the urgency plainly: fake videos of politicians already cost Americans billions of dollars a year, and telling real footage from generated footage will only get harder as elections approach.
This is only the first phase. Starting January 2027, large social platforms and AI hosting services with over two million users will have to detect AI-generated content and show users where it came from. From 2028, camera and microphone manufacturers will need to build the same disclosure capability directly into recording devices.
The law's soft spot is enforcement. California has no dedicated agency scanning millions of files for hidden watermarks — the mechanism kicks in through complaints or litigation. OpenAI, Google and Anthropic already published similar technical documentation under a separate California statute, AB 2013, but none have said anything specific yet about SB 942 compliance. There's still no federal mandate requiring AI labeling, so for the other forty-nine states, disclosure remains something companies choose to do on their own.
