California's AI transparency law is now live, requiring labels and free detectors

From August 2, 2026, California makes large AI providers label their own output and hand everyone a free detector.

The California AI Transparency Act (SB 942, as amended by AB 853) became operative on that date. It applies to generative AI providers with more than one million monthly users in California. Those providers must embed C2PA-compatible provenance data in the images, video and audio their systems generate, and must offer a free, publicly available AI detection tool.

What the law bans

The act also bans stripping provenance data and digital signatures from content, and goes a step further: it bans building or distributing tools designed to strip them. The target is not just the act of removing a label but the market for label-removal software.

Why this exact date

The date is not a coincidence. AB 853, signed October 13, 2025, moved the operative date to August 2, 2026 to line up with the EU AI Act’s own enforcement date. So on the same day, content provenance became a legal requirement on both sides of the Atlantic.

That synchronization matters more than either law alone. A provider serving both markets now faces the same provenance obligation in its two largest regulatory jurisdictions at once, which makes compliance the default engineering path rather than a regional add-on.

The burden shifts

What it means in practice: the burden shifts from “spot the fake” to “prove the origin”. If the system works, an unlabeled image starts to look suspicious by default instead of normal. Provenance becomes the expected state of legitimate content, and its absence becomes the signal.

The open question is durability. Will provenance marks survive one screenshot and a re-upload? The answer to that will decide whether this framework changes what people trust, or just changes what providers ship.