Some laws sound abstract — until you realize they hit almost every major AI provider. California’s AI Transparency Act (SB 942) is one of those. It went into effect on August 2nd, and it quietly changes how AI-generated content has to look in the largest US state.
What’s in it
The law targets providers of generative AI with more than a million monthly users in California. In other words, exactly the names you know: OpenAI, Google, Anthropic and the rest. Three obligations sit at its core.
First: images, video and audio must carry machine-readable provenance data — compatible with the C2PA standard, which embeds invisible origin information into content. Second: every provider has to offer a free, public tool that lets anyone check whether a piece of content came from its AI. Third: users must be given the option to add a visible AI label.
And it has teeth
This isn’t a toothless appeal. Violations cost $5,000 — per day, per instance, enforced by California’s attorney general. The law was originally meant to take effect back in January. A follow-up bill (AB 853) pushed the start date and deliberately synced it with the EU AI Act. That’s no accident: California and Brussels are moving in a similar direction on labeling.
My take
Provenance data is one of those things nobody talks about — until it’s missing. Anyone who has seen an AI image in the wild without knowing whether it’s real understands immediately why a standard like C2PA makes sense. The interesting part is enforcement: a watermark is only as good as its resistance to cropping and compression. And a detection tool is only as good as its hit rate. Still, the direction is right. Labeling is no silver bullet against deepfakes — but it’s the foundation everything else builds on. And when a market the size of California moves first, providers tend to roll the change out for everyone.
Sources: