EU AI Act Article 50: what you must label, by when, and how
A practical guide for companies that create or publish AI-generated images, video, audio or text in the EU. Updated October 2026, including the changes made by Regulation (EU) 2026/1744 (the “Digital Omnibus on AI”).
- Since 2 August 2026 Article 50 applies: AI-generated content must be marked in a machine-readable way, and deep fakes must be disclosed.
- By 2 December 2026 AI systems that were already on the market before 2 August 2026 must also mark their outputs.
- Fines go up to EUR 15 million or 3 % of worldwide annual turnover, whichever is higher.
1. Who has to do what
Article 50 splits the duties between two roles. Many companies are both.
| You are | Your duty | Where |
|---|---|---|
| Provider — you build or sell an AI system that generates images, audio, video or text | Mark every output in a machine-readable format so that it is detectable as AI-generated or manipulated. | Art. 50(2) |
| Deployer — you use AI to create image, audio or video content that is a deep fake | Disclose that the content was artificially generated or manipulated. | Art. 50(4) |
| Deployer — you publish AI-generated text to inform the public on matters of public interest | Disclose that the text was artificially generated, unless it went through human review and someone holds editorial responsibility. | Art. 50(4) |
The core sentence for providers:
…shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.Regulation (EU) 2024/1689, Article 50(2)
The marking has to work in practice, not just exist on paper: providers must make their technical solutions effective, interoperable, robust and reliable
as far as technically feasible. Purely assistive editing that does not substantially alter the input is exempt.
2. The deadlines
| Date | What happens | Source |
|---|---|---|
| 2 August 2026 | Article 50 applies in full to new AI systems and to deployers. | Art. 113 |
| 2 December 2026 | Providers whose generative AI systems were placed on the market before 2 August 2026 must comply with the marking duty of Art. 50(2). | Art. 111(4), added by Reg. (EU) 2026/1744 |
The December date is a grace period for existing systems only. It does not postpone the disclosure duties of deployers.
3. When and how to inform people
…in a clear and distinguishable manner at the latest at the time of the first interaction or exposure.Regulation (EU) 2024/1689, Article 50(5)
For evidently artistic, creative, satirical or fictional works the disclosure may be limited to a note that does not spoil the work. The European Commission supports codes of practice on detecting, marking and labelling AI content (Art. 50(7)); following them is the most practical way to show good faith.
4. What happens if you ignore it
Breaching the transparency obligations of Article 50 can be fined with up to EUR 15 000 000 or, for a company, up to 3 % of its total worldwide annual turnover for the preceding financial year, whichever is higher (Art. 99(4)(g)). National market surveillance authorities enforce it.
5. A practical checklist
- List every place where your company creates or publishes AI-generated images, video, audio or text.
- For each, decide whether you act as a provider, a deployer, or both.
- Providers: embed a machine-readable label in every output. The widely read standard is the IPTC “Digital Source Type” field (for example trainedAlgorithmicMedia) in the file's metadata.
- Make the label survive the real world: metadata is often stripped when a file is uploaded to social networks, so keep a second, independent way to recognise the file.
- Deployers: add a visible notice to deep fakes and to AI-written public-interest texts, no later than first exposure.
- Keep evidence: what was labelled, when, and by whom — you may have to show it to an authority.
6. How Ankorith helps
- Machine-readable label in the file. Ankorith writes the IPTC Digital Source Type label into images, so the content is marked as AI-generated in the format platforms read.
- A label that survives social media. Each seal also stores an appearance fingerprint, so a file can be recognised even after its metadata was stripped or it was recompressed.
- A verified issuer. Seals are issued by a company with a proven domain, so a reader sees who stands behind the label.
- Evidence you can show. Every seal is written to a public, tamper-evident log signed with a quantum-resistant key and anchored in Bitcoin — a record of what you labelled and when.
Ankorith is a tool, not legal advice. Whether a particular use counts as a deep fake or as public-interest text depends on your situation; if in doubt, ask your lawyer.
Try Ankorith: verification is free · Pricing · API for developers
Sources
- Regulation (EU) 2024/1689 (Artificial Intelligence Act), Articles 50, 99(4)(g), 111(4) and 113 — EUR-Lex.
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), published in the Official Journal on 24 July 2026, in force since 27 July 2026.