If your company uses a chatbot, generates images or text with AI, or publishes content that could pass for real, Article 50 of Regulation (EU) 2024/1689 reaches you somewhere, though rarely where people say. This page brings its paragraphs together in a single table — what each one requires, who it binds, from when, and what the deployer checks — and links the blog post that unpacks each case. It is the map; the detail is in the series.
The figure allocates the four duties of Article 50: the notice that people are interacting with an AI system and the marking of synthetic content fall to the provider; the information on emotion recognition or biometric categorisation and the disclosure of deep fakes and of text published to inform the public on matters of public interest, to the deployer.
The table
| Paragraph | What it requires | Who it binds | From when | What the deployer does |
|---|---|---|---|---|
| 50(1) | That people know they are interacting with an AI system, unless this is obvious | Provider | 2 August 2026 | Checks that the notice appears in its deployment and records it |
| 50(2) | Marking synthetic content in a machine-readable format | Provider | 2 August 2026; until 2 December 2026 for systems already on the market before 2 August (Article 111(4)) | Asks the provider whether and how it marks, and notes when the system was placed on the market |
| 50(3) | Informing the people exposed to an emotion recognition or biometric categorisation system | Deployer | 2 August 2026 | Its own obligation |
| 50(4) | Disclosing deep fakes; disclosing generated text published to inform the public on matters of public interest, unless there is human review or editorial control | Deployer | 2 August 2026 | Its own obligation |
| 50(7) | Codes of practice for marking and labelling | The Commission encourages and assesses them | — | May align with them; adherence gives no presumption of conformity |
What the deployer does on paragraphs 1 and 2 is not an Article 50 obligation: it is diligence. In the Alethexis taxonomy it is an internal methodological control (IM) — Alethexis's internal taxonomy, not the AI Act's terminology — and it is worth documenting all the same, because it is the first thing anyone asks.
Paragraph by paragraph
50(1): the provider designs the notice
The paragraph requires providers to design and develop systems intended to interact with people "in such a way that the natural persons concerned are informed that they are interacting with an AI system", unless this is obvious "from the point of view of a natural person who is reasonably well-informed, observant and circumspect". The addressee is the provider; the deployer checks that the notice is where the customer sees it. The detail is in the post on the notice the provider owes and the one on the division of roles in a chatbot.
50(2): marking, also the provider's, and the only paragraph with a transitional period
Providers of systems generating synthetic content "shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated". It is a technical mark, not a visible label. It does not apply where the system performs "an assistive function for standard editing" or does not substantially alter the input data.
It is the only paragraph with a transitional period, and the period belongs to providers alone: those whose systems "have been placed on the market before 2 August 2026" have to comply with Article 50(2) "by 2 December 2026" (Article 111(4)). The 2024 text provided for none: Regulation (EU) 2026/1744 created it. Systems launched afterwards comply from the start. The marking post goes through it in detail.
50(3): emotions and biometrics, the deployer's
Deployers of an emotion recognition or biometric categorisation system "shall inform the natural persons exposed thereto of the operation of the system" and process their data in accordance with the GDPR. Another question comes first: in the workplace, inferring a person's emotions is prohibited by Article 5(1)(f), except for medical or safety reasons, so this paragraph does not open the door that Article 5 closes.
50(4): deep fakes and text of public interest
A deep fake is, under Article 3(60), "AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful". The deployer has to disclose that it is artificial. Where it forms part of an evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing its existence "in an appropriate manner that does not hamper the display or enjoyment of the work". The EU's official icons are one way to do it, not the only one.
For text, the paragraph covers "text which is published with the purpose of informing the public on matters of public interest", and does not apply where the content "has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content". That is the editorial exemption, set out in the post on the editorial review route.
50(7): codes, with no presumption
The Commission encourages codes of practice on detection, marking and labelling, and "shall assess whether adherence to those codes of practice is adequate" for the obligations in paragraphs 2 and 4. Recital 41 of Regulation (EU) 2026/1744 narrows it: these codes "have limited legal effect, and in particular do not grant a presumption of conformity". Signing one organises the work; it does not replace it. The post on the Transparency Code covers it.
The dates, kept apart
Article 50 applies from 2 August 2026, and Regulation (EU) 2026/1744 did not postpone it: paragraphs 1, 3 and 4 have no transitional period. The post on the date that didn't move explains why, and the one on the day it started to apply, what changed that day.
On 2 December 2026 there are two different things in the calendar: the end of the marking transitional period for providers, and the start of the new Article 5(1)(ba) and (bb) prohibitions, on child sexual abuse material and non-consensual intimate content. They share the day and nothing else.
The eight posts in the series
- The "you're talking to an AI" notice is owed by the provider, not by you — 50(1).
- Who has to disclose that your chatbot is an AI? — 50(1) and the division of roles.
- Marking synthetic content: whose obligation it is, and how far the transitional period reaches — 50(2) and 111(4).
- The article that didn't move — the date.
- Article 50 transparency, in application since 2 August 2026 — what changed that day.
- Do you have to label all AI-written text? — 50(4), text.
- The EU's official icons for labelling AI content — 50(4), image, video and audio.
- The code of practice on AI content transparency — 50(7).
Who checks in Spain that these notices are where they should be: AESIA as the main market surveillance authority, with the share that falls to each sectoral authority — the map, in this article.
And in the glossary, the two definitions most asked for: transparency obligations and marking of synthetic content.
Content in accordance with Articles 3, 5, 50, 111 and 113 of Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744, and recital 41 of the latter.
This article is for informational purposes only and does not constitute legal advice.