The rules on labeling AI content, in one place
Marking AI output is now a legal duty in the EU and California. Here is what each rule asks, and how signed provenance can support the work.
| Law | From | What it asks |
|---|---|---|
| EU AI Act, Article 50 | 2 Aug 2026 | Mark AI output in a machine-readable way, detectable as AI-generated. |
| California AI Transparency Act | 2 Aug 2026 | Embed a latent disclosure in AI image, video and audio; offer a free check. |
| China, Deep Synthesis Provisions | 10 Jan 2023 | Label synthetic content visibly and with an implicit, machine-readable mark. |
A summary, not legal advice. Each guide gives the source and its review date.
EU AI Act, Article 50(2)
Providers of generative AI systems must ensure that outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. The duty applies from 2 August 2026. Systems placed on the market before that date have until 2 December 2026. Fines for Article 50 breaches reach EUR 15 million or 3% of worldwide annual turnover, whichever is higher, and whichever is lower for SMEs.
The final Code of Practice, published on 10 June 2026, describes a multi-layered approach for providers: signed, time-stamped metadata plus an imperceptible watermark, with a detection solution that is free to use. Following the Code is not a legal safe harbour.
California AI Transparency Act
Every generative AI provider whose system is publicly accessible in California must embed a latent disclosure in AI image, video and audio content and make a free disclosure verification tool available. SB 1000, approved on 30 September 2026, removed the earlier threshold of 1,000,000 monthly users and deleted the manifest disclosure option. The latent disclosure duty remains. The civil penalty is USD 5,000 per violation, and each day is a separate violation.
Where provenance fits
Signed C2PA Content Credentials record machine-readable origin. An invisible watermark can outlive stripped metadata. A verification API gives you a detection route for your own files. Together they support the marking and detection work these rules describe. They do not replace your disclosures, testing, terms and governance.
See which recipe fits each rule
Guides by jurisdiction
- EU AI Act Article 50Machine-readable marking and detection of AI output.
- California AI Transparency Act (SB 942)Latent disclosures and a disclosure verification tool.
- China deep synthesis provisionsLabeling of synthetic content under CAC rules.
- India IT RulesDuties on synthetic and deepfake content for intermediaries.
- UK Online Safety ActDeepfake duties for platforms.
- Australia Online Safety ActDeepfake penalties and removal powers.
- US TAKE IT DOWN ActTakedown duties for non-consensual intimate imagery.
- Texas SB 751Election deepfakes.
- Illinois BIPABiometric consent and deepfakes.
Original Pictures shows where content came from and whether it has changed since it was signed. It does not prove that a depicted event is true. Using it supports your work on the EU AI Act, the California AI Transparency Act and other AI labeling rules, and it is not by itself legal compliance with any of them. Nothing on this site is legal advice.
Map the rules you track to API calls
Book a demo and we will show your legal and engineering teams how signing, watermarking and verification fit your obligations.