EU AI Act Article 50 in numbers
Article 50 is the part of the EU AI Act that most software vendors meet first — usually inside a customer's AI questionnaire, not in a courtroom. Here is the whole article, quantified.
0 — days of delay from the Digital Omnibus
The Digital Omnibus (adopted 29 June 2026) postponed the EU AI Act's high-risk chapter. It did not delay Article 50: the transparency obligations apply from 2 August 2026. If a questionnaire answer claims "the AI Act was delayed", it is conflating chapters — the single most useful fact to get right this quarter.
5 — paragraphs, of which 4 create substantive duties
| Paragraph | Who owes it | Trigger | Duty |
|---|---|---|---|
| Art. 50(1) | Provider | AI systems interacting directly with people (chatbots, assistants, voice agents) | Tell people they are interacting with AI — unless obvious to a reasonably well-informed, observant and circumspect person |
| Art. 50(2) | Provider | Systems generating synthetic audio, image, video or text | Mark outputs machine-readably as artificially generated |
| Art. 50(3) | Deployer | Emotion recognition or biometric categorisation | Inform the people exposed |
| Art. 50(4) | Deployer | Deepfakes; AI text published to inform the public on matters of public interest | Disclose the content is artificially generated or manipulated |
| Art. 50(5) | Both | — | All of the above: clear, distinguishable, at the latest at first interaction or exposure, accessible |
2 — regulated roles, split 2 / 2 / 1
Only two roles carry Article 50 duties: providers (who develop a system or ship one under their own name) and deployers (who use one under their own authority). The duties split two to providers — 50(1) and 50(2) — two to deployers — 50(3) and 50(4) — and one transversal requirement, 50(5), to both. Which side of the line you are on decides which questionnaire answers are yours to give: who owes what under Article 50.
1 — rebranding rule that reassigns the provider role
Ship a third-party foundation model inside your product under your own brand and, for Article 50(1) purposes, you are the provider of the resulting AI system — the duty toward your users does not sit with your model supplier. It is the most commonly misread rule in Article 50, and the first thing to check before answering an AI questionnaire.
2 — extraterritorial triggers that reach non-EU vendors
Art. 2(1)(a): placing an AI system on the EU market, wherever the provider is established. Art. 2(1)(c): providers and deployers in third countries where the system's output is used in the EU. A US or UK SaaS with EU users is not automatically out of scope — details in the non-EU vendor guide.
4 — vendor-relevant exceptions written into the article
- 50(1): no duty where AI interaction is obvious from the point of view of a reasonably well-informed, observant and circumspect person, in context.
- 50(2): no marking duty to the extent the system performs an assistive function for standard editing or does not substantially alter the deployer's input or its meaning.
- 50(4): reduced disclosure for creative, satirical, artistic or fictional works.
- 50(4): no disclosure for AI-assisted public-interest text under human review with editorial responsibility.
Paragraphs 50(1) through 50(4) also each contain a law-enforcement carve-out for systems authorised by law to detect, prevent, investigate or prosecute criminal offences. Those four are excluded from the count above because they cannot apply to a commercial vendor's product — see the counting rules in Method.
1 — transition window, ending 2 December 2026
Systems already on the market when Article 50 starts applying benefit from a marking transition under 50(2) until 2 December 2026 — four months, for that duty, and that duty only.
1 — related outright prohibition disclosure cannot cure
Inferring emotions in workplace and education settings is prohibited outright by Art. 5(1)(f) — applicable since 2 February 2025 — save for medical or safety reasons. A 50(3) disclosure does not legalize a prohibited use.
7 — dates that anchor every correct questionnaire answer
| Date | What happens |
|---|---|
| 1 Aug 2024 | Regulation (EU) 2024/1689 enters into force |
| 2 Feb 2025 | Art. 4 AI literacy + Art. 5 prohibitions apply |
| 29 Jun 2026 | Digital Omnibus adopted (delayed high-risk; did not delay Article 50) |
| 2 Aug 2026 | Article 50 transparency duties apply |
| 2 Dec 2026 | End of the 50(2) marking transition for systems already on the market |
| 2 Dec 2027 | High-risk (Annex III) obligations apply — post-Omnibus |
| 2 Aug 2028 | High-risk embedded in regulated products applies — post-Omnibus |
Want the duty map for your specific feature? The free Article 50 checker walks the triggers above in six questions — no email required. And the underlying dataset is open (CC BY 4.0): use it, cite it, or send corrections.
Method
- All counts derived from the open dataset eu-ai-act-article-50 (CC BY 4.0, maintained by Deal Rescue), which was last verified against the Official Journal text of Regulation (EU) 2024/1689 on 25 July 2026.
- Counting rules: "duties" = paragraphs imposing an obligation (4 substantive + 1 on form/timing); "exceptions" = express carve-outs within Article 50's own text that can apply to commercial vendors. The law-enforcement authorisation carve-outs in 50(1)–50(4) are excluded from the headline count as inapplicable to commercial products; the Art. 5(1)(f) prohibition and Art. 2(1) scope triggers are counted separately as related provisions.
- This page is orientation data, not legal advice. Always verify against the Official Journal.