Is that photo real? On October 7, Google gave everyone a new way to ask. It expanded its SynthID Detector "to everyone," so "anyone can easily check if an image, video, or audio file was made with AI from Google or our partners, including OpenAI, NVIDIA, Kakao, and soon, Apple." Google says it has watermarked "over 180 billion images and videos."
But in California, a free AI detector isn't just a nice feature. Since August 2, it's the law.
Takeaway #1: The AI Transparency Act makes AI providers hand you a detector
FROM THE STATUTE
"A covered provider shall make available a disclosure verification tool at no cost to the user"
What it means for you. This is the California AI Transparency Act, which became operative on August 2, 2026. It used to cover only AI systems with over 1 million monthly users. SB 1000, effective September 30, dropped that number. Now a "covered provider" is anyone who makes a generative AI system that's publicly accessible in California.
The tool has to do specific things. It must let a user check whether image, video or audio content "was created or altered, except by minor modification, by the covered provider's GenAI system." A crop doesn't count. It must accept an upload or a link and work through an API. It must show "system provenance data," the machine's fingerprints, but not your personal information unless you agree. A provider may instead send users to a compliant third-party tool.
The same law requires a hidden watermark, which it calls a "latent" disclosure, in AI-made or AI-altered images, video and audio, "to the extent it is technically feasible." It carries the provider's name, the system and version, the creation time, a unique ID and whether the AI created or altered the content. The visible-label option is gone.
There's a licensing hook too. If a provider learns a licensee changed its system so it no longer complies, the provider has 72 hours to cut off the license or send a notice. The licensee then has 96 hours to fix it or stop using it. Those are short fuses, and they belong in AI license agreements.
I'm not saying Google built SynthID Detector because of California. Google doesn't say that. But it's the kind of tool the law describes, and its limits match the law's. A provider's detector reads its own marks and its partners'. It can't tell you about content from a company that never watermarked.
Your move. If a detector finds no watermark, treat the file as unknown, not as proof it's real. If authenticity matters to a deal or a dispute, ask for the original file and its metadata.
Takeaway #2: Platforms join in January
FROM THE STATUTE
"A large online platform shall not, to the extent technically feasible, knowingly strip any system provenance data or digital signature from content uploaded to, distributed on, or downloaded from the large online platform."
What it means for you. Starting January 1, 2027, the duty spreads. A "large online platform" is a social media, file-sharing or messaging platform, or a search engine, with more than 2,000,000 unique monthly users. It will have to detect standards-compliant provenance data and let users see it. And it can't knowingly strip it, even from downloads.
For creators, that changes the math on AI content. The label your AI tool attached is more likely to survive the upload. If you post AI-made content as if it were real, the trail may follow it.
Notice who isn't on the hook. The duties fall on AI providers, platforms and, under the statute's penalty section, capture device makers. Not on you, the person posting. The devil is in the details, and the detail here is that this law regulates the plumbing, not the user. Other laws, like California's political-ad rule, do reach the people who publish.
Enforcement isn't private. The Attorney General, a city attorney or a county counsel can sue for $5,000 per violation, and "each day" a company is in violation counts separately.
Your move. Brands and agencies, ask your AI vendors in writing whether they're covered providers and how their watermarking works. Then make sure your own editing tools don't strip the marks you'll want later.
Sample wording (our illustration): "Vendor will not remove or alter any provenance data or watermark embedded in AI-generated deliverables."
And always remember, don't sign it till you read it, my friends.
General information, not legal advice. Reading this doesn't create an attorney-client relationship. Ask your own lawyer how California's AI laws apply to your business.
Sources
- California Business and Professions Code, Division 8, Chapter 25 (California AI Transparency Act), §§22757-22757.6 (California Legislative Information)
- Google, 'Google expands SynthID Detector for AI content', Oct 7, 2026
- SB 1000, California AI Transparency Act, Stats. 2026, ch. 861
- AB 2713, California AI Transparency Act: system provenance data, Stats. 2026, ch. 856
