Starting today, August 2, 2026, a California law called the AI Transparency Act (SB 942) requires the biggest AI companies to make it easier to tell what content their tools created.
The rule applies to any generative AI system with more than one million monthly users in California — in practice, that covers ChatGPT (OpenAI), Gemini (Google), Meta AI, and other large platforms. Since companies rarely build a separate version of a product just for one state, the changes will likely show up for users everywhere, not only in California.
What's supposed to change
- A free detection tool. Covered companies now have to offer a free, public tool anyone can use to check whether an image, video, or audio clip was made or altered by their system.
- A visible "AI-generated" label option. When you create something with one of these tools, you'll be able to attach a visible marker showing it's AI-made. It's an option you choose, not something added automatically.
- A hidden digital fingerprint. Images, video, and audio from these systems also get invisible, machine-readable data baked in (the industry calls this "content provenance"). Unlike a visible label, it's built to survive even if the file gets cropped, screenshotted, or re-saved.
Companies that skip this face fines of $5,000 per violation per day, enforced by the California Attorney General or local prosecutors. The law was signed in 2024 and was originally supposed to start in January 2026, but a follow-up bill pushed the date to today, deliberately lining it up with a similar deadline in the EU.
Speaking of which: the European Union's AI Act reached a comparable milestone on this same date. Our guide on what changed under the EU AI Act covers the chatbot-disclosure side of that rule, if you're in Europe.
What this means if you're worried about scams
The good news: if you get a suspicious photo, video, or voice message, you may now be able to run it through one of these free detection tools to see if it came from a major AI platform.
The catch: the law only covers the biggest platforms — most text-only chatbots and smaller AI tools aren't required to offer anything at all. And people running scams were never going to use a mainstream, rule-following tool in the first place — they'll keep using whatever ignores the law.
So treat a detection tool as one more clue, not proof either way. Keep using the habits from our guides on spotting AI-generated photos and checking if a video call is a deepfake: if a voice or message claiming to be a family member asks for money, hang up and call them back on a number you already have saved.
More of the law phases in later. Large platforms that host AI content, such as social media apps, have to add their own detection and labeling by January 1, 2027. Camera and phone makers get until January 1, 2028 to offer an optional way to mark photos as verifiably not AI-edited.
Sources
- Bill Text - SB-942 California AI Transparency Act — California Legislative Information
- California's Ongoing AI Regulation: Key Deadlines Arriving in 2026 and Beyond — The National Law Review
- SB 942: Who It Covers, Disclosures, and Penalties — LegalClarity