Documentation menuWhat Article 50 asks
What Article 50 asks
The transparency duties of Article 50 of the EU AI Act, who they apply to, the deadlines, and which Witness feature fits each one.
This page summarises the law so you can find the right feature. It is not legal advice. Read the regulation text and get advice for your case.
The five paragraphs
| Paragraph | Who | Duty | Witness |
|---|---|---|---|
| 50(1) | Providers of AI systems that interact directly with people | Design the system so people are told they are interacting with an AI, unless that is obvious from the context | <witness-notice>, <AiNotice> |
| 50(2) | Providers of AI systems that generate synthetic audio, images, video or text | Mark the outputs in a machine-readable format so they are detectable as artificially generated or manipulated | Image XMP, text watermark, AI SDK middleware (partly, see below) |
| 50(3) | Deployers of emotion recognition or biometric categorisation | Inform the people exposed to the system | kind="emotion-recognition", kind="biometric-categorisation" |
| 50(4) | Deployers of systems that create deepfakes, or text published to inform the public on matters of public interest | Disclose that the content was artificially generated or manipulated. For text, not needed when a person reviewed it and someone holds editorial responsibility | <witness-label kind="deepfake">, labelHtml, humanReviewed |
| 50(5) | All of the above | Give the information clearly, at the latest at the first interaction or exposure, and accessibly | Notices shown before the first message, labels at the top, screen reader support |
"Provider" is whoever develops the system or has it developed and puts it on the market under their name. "Deployer" is whoever uses it in their own activity. A company that builds a chatbot on top of a model API is usually the provider of that chatbot. The rules also reach companies outside the EU, including in Switzerland, when the output is used in the EU (Art. 2(1)(c)).
Dates
- 2 August 2026: Article 50 applies.
- 2 December 2026: end of the grace period for the machine-readable marking of Art. 50(2) for generative systems that were already on the market before 2 August 2026, under the Digital Omnibus agreement. The other duties got no extra time. (CSA research note, Gibson Dunn)
- Fines for breaching Article 50 go up to EUR 15 million or 3 % of worldwide annual turnover (Art. 99(4)).
The Code of Practice
The European Commission published a voluntary Code of Practice on marking and labelling AI-generated content on 10 June 2026 (Jones Day, Lewis Silkin). In short:
- Marking should combine at least two layers, typically digitally signed metadata and an imperceptible watermark.
- A common EU icon built around the letters "AI" is foreseen for visible labels; equivalent icons are allowed.
- Labels for published text belong at the top, near the headline. Deepfake video is labelled at the start and repeatedly; audio gets a spoken disclaimer at the beginning.
Witness covers the visible label, unsigned metadata and a text watermark. It does not produce signed C2PA manifests or watermarks for images, audio or video. For full 50(2) marking of images, use a generator that embeds C2PA Content Credentials, or add a C2PA signing step; Witness keeps such files intact.