Regulation (EU) 2024/1689 · Article 50
What Article 50 actually requires.
Article 50 sets four narrow transparency obligations on AI systems that interact with people or generate synthetic content. This page summarises the obligations in plain language and cites every rule back to the Regulation. It is not legal advice.
01 · Disclose AI interaction
People talking to your AI must know they are talking to AI.
Applies to providers of any AI system intended to interact directly with natural persons — chatbots, voice agents, AI-powered support. Exempts uses where the AI nature is obvious to a reasonably well-informed user.
“Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious.”
02 · Mark synthetic outputs
AI-generated content must be machine-readable as such.
Applies to providers of generative AI: text, image, audio, video. The marking must be machine-detectable — for example via content credentials, watermarks, or provenance metadata. A visible label alone is not sufficient under Art. 50(2).
“Providers of AI systems generating synthetic audio, image, video or text content shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.”
03 · Label deepfakes
Synthetic media of real people must be visibly labelled.
Applies to deployers — the party using the AI to produce the deepfake — not just the model provider. A narrow artistic-expression carve-out exists but must not impair the person’s enjoyment of their rights.
“Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated.”
04 · Disclose AI-generated text on public-interest topics
AI-written articles on matters of public interest must say so.
Applies to deployers publishing AI-generated text to inform the public about matters of public interest. Exempts editorially reviewed content where a human takes responsibility — but the burden of evidence sits with the deployer.
“Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated.”
Effective date
2 August 2026.
Article 50 applies from 2 August 2026 across the EU. The Commission’s draft guidelines were published on 8 May 2026; consultation runs through 3 June 2026.
Penalty bracket
€15M or 3% of worldwide annual turnover.
Article 99(4)(g) sets the maximum at the higher of the two. National competent authorities have discretion on lower penalties. Some commentary online cites the €35M / 7% bracket — that’s the prohibited-practice ceiling under Art. 99(3), not Art. 50 transparency.
Where ArticleProof helps
Disclosure is the easy part. Proving you disclosed is the hard part.
You can put a disclosure banner on your site in an afternoon. What an auditor or competent authority will ask for is the evidence trail: when did each disclosure render, was it acknowledged, what AI model produced the output, can you reproduce a record of any specific session. ArticleProof captures every disclosure event server-side, hashes it for integrity, and exports a deterministic PDF audit pack you can hand to a regulator.
ArticleProof organizes AI transparency records. It is not legal advice — consult an EU-qualified lawyer for guidance on your specific obligations.