Article 50 is the part of the EU AI Act that is already enforceable. Since 2 August 2026, national authorities have had fining powers over it. It was not touched by the June 2026 Omnibus — while high-risk obligations slid to December 2027, transparency did not move a day. If you ship a chatbot, a generative feature, or anything that produces synthetic media, this is the article that applies to you now.
Most teams reading about the "AI Act delay" concluded they had until 2027. For Article 50 duties, that is wrong by sixteen months.
The four duties
Article 50 imposes disclosure, not conformity assessment. There is no notified body, no CE marking, no technical file. The obligations are small — which is exactly why failing them is hard to excuse.
| Paragraph | Who it binds | What you must do |
|---|---|---|
| 50(1) — human interaction | Provider | Tell people they are dealing with an AI system, unless it is obvious to a reasonably well-informed person. Must be at first interaction. |
| 50(2) — synthetic content | Provider | Mark AI-generated audio, image, video or text in a machine-readable format so it can be detected as artificially generated. |
| 50(3) — emotion & biometric categorisation | Deployer | Inform the people exposed to the system that it is operating. GDPR duties apply on top. |
| 50(4) — deepfakes & public-interest text | Deployer | Disclose that content is artificially generated or manipulated. Narrow artistic and satirical carve-outs exist. |
Where teams actually fail this
Disclosure buried in the privacy policy. The duty is to inform the person interacting with the system, at the point of interaction. A line in a policy document nobody opens does not discharge it. The bot's greeting does.
"It's obvious it's a bot." The exemption in 50(1) is for cases obvious to a reasonably well-informed, observant and circumspect person. A widget named "Ask Ava" with a human avatar is not that. If your product design works hard to feel human, you have argued yourself out of the exemption.
Assuming the model vendor handles 50(2). Machine-readable marking of synthetic output is a provider duty. If you call OpenAI's or Anthropic's API, the marking obligation for the underlying model output generally sits with them — but if you generate, assemble or re-encode media yourself, check rather than assume. Where the provider line sits.
Emotion recognition in the workplace. 50(3) requires disclosure — but in employment and education contexts, emotion inference is prohibited outright under Article 5, not merely disclosable. Disclosure does not cure a prohibited practice.
What compliance looks like in practice
- One sentence in your chatbot's opening message: this is an AI assistant.
- A visible label on generated media in your UI, plus whatever provenance metadata your generation stack supports.
- A dated note in your compliance file recording which systems you assessed under Article 50 and what you concluded — including the ones you decided were exempt, and why.
- An Article 4 AI-literacy session minuted, since that is live too.
That is a morning's work. The reason to do it this week rather than next quarter is that this is the one part of the Act with active enforcement powers and no grace period left.
What it costs to get wrong
Article 50 breaches fall under Article 99(4): up to €15,000,000 or 3% of worldwide annual turnover, whichever is higher — or, if you are an SME or start-up, whichever is lower. The full fine structure.
Article 50 does not make your system high-risk. A chatbot with disclosure duties is a Limited Risk system, and the Chapter III obligations that everyone fears do not attach to it. If you want to know which side of that line you are on, the chatbot guide works through it.
Does Article 50 apply to your product?
The free check identifies which of your features carry disclosure duties and which are already compliant.
Start the free scanNo card, no account. Full gap report from €59. See a real report first.
Self-assessment, not legal advice. This page and the Reglynn report are a structured self-assessment based on the EU AI Act (Regulation 2024/1689, as amended by the June 2026 Omnibus). They indicate where you likely stand and what to verify with a qualified advisor. Reglynn is not a notified body and issues no certification.
How this was made. Reglynn reports are generated by an AI system and reviewed before delivery. Reglynn is a Limited Risk AI system under Article 50; we disclose AI involvement wherever it occurs, including in our chat assistant. Our own self-assessment.