Skip to content
disclosura

Law explained

EU AI Act Article 50, explained in plain language (2026)

Article 50 of the EU AI Act is the "transparency" article: the set of rules that says people shouldn't be deceived about when they're dealing with AI or AI-made content. It has applied since 2 August 2026 (Art. 113 of Regulation (EU) 2024/1689). Here is the whole article, duty by duty, in plain words: what each paragraph asks, the exemptions written into it, who owes which duty, the dates, the fine cap, and who enforces it where you are.

Provider or deployer? The two roles Article 50 speaks to

Every duty in Article 50 is addressed either to the provider of an AI system or to its deployer. The Regulation defines both in Article 3:

  • Provider (Art. 3(3)): "a natural or legal person, public authority, agency or other body that develops an AI system or a general-purpose AI model or that has an AI system or a general-purpose AI model developed and places it on the market or puts the AI system into service under its own name or trademark, whether for payment or free of charge". Plain example: the company that builds a chat assistant and sells or gives it away under its own name, including the maker of the general-purpose chatbot you subscribe to.
  • Deployer (Art. 3(4)): "a natural or legal person, public authority, agency or other body using an AI system under its authority except where the AI system is used in the course of a personal non-professional activity". Plain example: the online shop that installs that assistant on its website to answer customers, or the agency that publishes an image made with an AI tool for a client. Personal, non-professional use is outside the definition.

Paragraphs 1 and 2 of Article 50 bind providers; paragraphs 3 and 4 bind deployers. If you buy AI tools and use them in your business, you are a deployer of those tools, and the duties that are yours directly are the ones in 50(3) and 50(4). The free check sorts out which of them apply to you.

50(1) — Tell people they're talking to AI

If you run an AI system that interacts directly with people — a chatbot, a voice assistant — it must be designed so people are informed they're dealing with AI, unless that's already obvious to "a natural person who is reasonably well-informed, observant and circumspect." In practice: one plain sentence at the start of each conversation, close to the chat field; a small "AI" badge may stay on screen after it. That follows the Commission's guidelines (points 37 and 143), which also say what is not enough on its own (point 38): a line in the terms, a link, the word "assistant" with no "AI" in it, or a general "services on this website use AI". Who owes it: the provider designs for it — and if you built the chatbot yourself, or had it built, and run it under your own name, the Commission's guidelines (section 2.3) treat you as its provider. If you deploy someone else's chatbot on your site, you'll want the notice visible regardless — you're the face the customer sees. Full guide →

The exemptions in the text: the "obvious" case above, judged "taking into account the circumstances and the context of use"; and AI systems "authorised by law to detect, prevent, investigate or prosecute criminal offences", unless those systems are available for the public to report a criminal offence.

50(2) — Mark synthetic content machine-readably

Providers of AI that generates synthetic audio, images, video or text must ensure outputs are marked in a machine-readable way as artificially generated. Marking must be effective, interoperable, robust "as far as this is technically feasible." Marking applies since 2 August 2026; tools already on the market before that date have until 2 December 2026 (Reg. (EU) 2026/1744, the “Digital Omnibus”). If your business uses third-party AI tools, this duty belongs to your tool's maker, not to you — but it is good practice to choose tools that mark, and to avoid stripping marks when editing. Full guide →

The exemptions in the text: the duty does not apply "to the extent the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof", nor where authorised by law for criminal-law purposes. A spell-checker or a crop tool is not a synthetic-content generator.

50(3) — Emotion recognition and biometric categorisation

If you deploy these, you must inform the people exposed to them, and process their personal data in line with the GDPR (and the other data-protection acts the paragraph names). Bigger picture: several such practices are outright prohibited (Art. 5) or high-risk (Annex III). If this is you, you need a lawyer, not a widget — we say this plainly.

The exemption in the text: systems "permitted by law to detect, prevent or investigate criminal offences", with safeguards and in accordance with Union law.

50(4) — Label deep fakes and AI text on matters of public interest

Deployers must disclose (a) AI content that constitutes a "deep fake" — defined in Art. 3(60) as "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", so even benign ones — and (b) AI-generated text "published with the purpose of informing the public on matters of public interest". Deepfakes → · AI text →

The exemptions in the text: for deep fakes, where the content "forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme", disclosure is limited to a form "that does not hamper the display or enjoyment of the work". For text, no disclosure is owed 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". Both halves also exempt uses authorised by law for criminal-law purposes.

50(5) — Timing, clarity, accessibility

The information above must be provided "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure", and it must conform to the applicable accessibility requirements. Not in the terms of service, not after ten messages: at the start, where the person can see it. For a chatbot the Commission's guidelines read this as at least once at the start of each interactive session (point 143). For a deep fake it means a label people can see or hear on the content itself, without a tool and without a click; the hidden marks in the file do not replace it (point 117).

50(6) and 50(7) — What Article 50 leaves untouched, and the Code of Practice

Paragraph 6 says Article 50 does not affect the high-risk requirements of Chapter III and is "without prejudice to other transparency obligations laid down in Union or national law for deployers of AI systems" — consumer law and data-protection law still apply on top. Paragraph 7, as replaced by Reg. (EU) 2026/1744, asks the Commission to encourage codes of practice on the detection, marking and labelling of AI-generated content and to assess whether adhering to a code is adequate to ensure compliance with paragraphs 2 and 4. The final Code of Practice on Transparency of AI-generated Content was published on 10 June 2026; it is voluntary and has two parts: Section 1 for providers (marking and detection) and Section 2 for deployers (labelling deep fakes and published AI text).

The dates

  • 2 August 2026 — Article 50 applies (Art. 113, general date of application).
  • 2 December 2026 — Marking applies since 2 August 2026; tools already on the market before that date have until 2 December 2026 (Reg. (EU) 2026/1744, the “Digital Omnibus”). This is Art. 111(4), inserted by the Digital Omnibus, in force since 27 July 2026. It concerns providers of generative tools placed on the market before 2 August 2026, not the businesses using them.

A dated log of what the official sources publish, and when, is kept on the Article 50 radar.

The fine cap

Art. 99(4), point (g), covers "transparency obligations for providers and deployers pursuant to Article 50": administrative fines of up to €15,000,000 or 3% of total worldwide annual turnover, whichever is higher. Art. 99(6) adds that "in the case of SMEs, including start-ups", each fine is capped at the lower of the two; Reg. (EU) 2026/1744 inserted a paragraph 6a that extends the lower cap to "small mid-cap enterprises". Under Art. 99(1) it is the Member States that lay down the actual penalties, which "may also include administrative fines, warnings and non-monetary measures". The real numbers, in context →

Who enforces Article 50, and where to complain

Each Member State designates at least one notifying authority and at least one market surveillance authority for the Regulation (Art. 70(1)), and names one market surveillance authority as its single point of contact (Art. 70(2)). Art. 85 gives "any natural or legal person having grounds to consider that there has been an infringement" the right to submit a complaint to the relevant market surveillance authority. Who that is depends on the country:

  • Italy — the Agenzia per la cybersicurezza nazionale (ACN) is the market surveillance authority and single point of contact; the national rules on penalties and complaints are in Legislative Decree 179/2026, in force since 23 October 2026. Chi controlla l'AI Act in Italia →
  • Romania — no authority had been designated by law on 1 November 2026: a Government memorandum of 12 March 2026 proposed ANCOM as market surveillance authority and single point of contact, and ANCOM said on 24 July 2026 that it can verify and sanction only once the national implementing law is in force. Cine aplică AI Act în România →

What Article 50 is not

It is not the "high-risk" regime (that's Chapter III, phasing in through August 2028), and it's not the GPAI-model rules for AI companies. For a typical business that uses AI tools, Article 50 is the part of the AI Act you actually touch.

Questions people ask about Article 50

When does Article 50 of the AI Act apply?

Article 50 applies since 2 August 2026, the general date of application in Article 113 of Regulation (EU) 2024/1689. One transitional rule exists for machine-readable marking: Marking applies since 2 August 2026; tools already on the market before that date have until 2 December 2026 (Reg. (EU) 2026/1744, the “Digital Omnibus”). It is a deadline for the providers of those AI tools, not for the businesses that use them.

The dates, in the radar →

Where is the official text of Article 50?

On EUR-Lex, under the regulation’s permanent identifier https://eur-lex.europa.eu/eli/reg/2024/1689/oj, in all 24 official languages. The same text is kept in the Publications Office’s Cellar database (CELEX 32024R1689). The amendments of Regulation (EU) 2026/1744 are at https://eur-lex.europa.eu/eli/reg/2026/1744/oj.

Regulation (EU) 2024/1689 on EUR-Lex →

Are there official guidelines on Article 50?

Yes. The European Commission first published them on 20 July 2026, and since 12 November 2026 they are in the Official Journal (C/2026/6005), in all 24 official languages: the “Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689 (the ‘AI Act’)”. They are the Commission’s own reading of who must disclose what, with definitions, exemptions and examples. They are interpretive guidance; the binding text remains the Regulation.

The guidelines in the Official Journal (C/2026/6005) →

What is the Code of Practice on transparency of AI-generated content?

A voluntary code published in its final form on 10 June 2026, drawn up under Article 50(7). It has two parts: Section 1 for providers of AI systems that generate content (marking and detection) and Section 2 for deployers (labelling deep fakes and published AI text). Adhering to it is voluntary; the Article 50 duties themselves are not. Since Regulation (EU) 2026/1744, Article 50(7) asks the Commission, taking utmost account of the AI Board’s opinion, to assess whether adherence to such a code is adequate to ensure compliance with paragraphs 2 and 4.

The Code of Practice page →

Does Article 50(1) apply if my chatbot is obviously a bot?

Article 50(1) does not apply where it is obvious, to a natural person who is reasonably well-informed, observant and circumspect, that they are dealing with an AI system, taking into account the circumstances and the context of use. The test is the average customer’s view, not yours, and the Commission’s guidelines (section 3.2.1) read the exemption restrictively: only where almost no doubt is left, and not where its users include vulnerable people, such as children or older people, for whom the AI nature is less obvious. A one-line notice at the first interaction removes the need to rely on the exemption, which is why we recommend it even where the exemption would hold.

The “obvious AI” exemption, examined →

Does Article 50(4) apply to my blog or newsletter?

Only to AI-generated or AI-manipulated text that is published with the purpose of informing the public on matters of public interest, and not where the content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication. The Commission’s guidelines name, as text that misses that exception, a website where AI-generated articles on Union policy are posted without any deliberate human review or editorial control.

AI-written text and the public-interest rule →

What are the fines for breaking Article 50?

Article 99(4), point (g), sets the ceiling for the transparency obligations of Article 50: administrative fines of up to EUR 15 000 000 or, for an undertaking, up to 3 % of its total worldwide annual turnover for the preceding financial year, whichever is higher. Article 99(6) caps each fine at the lower of the two amounts “in the case of SMEs, including start-ups”; Regulation (EU) 2026/1744 inserted a paragraph 6a extending that lower cap to “small mid-cap enterprises” (SMCs). Member States lay down the actual penalties, which may also include warnings and non-monetary measures (Article 99(1)).

AI Act fines: the real numbers →

Disclosura is a technical implementation tool, not legal advice. For legal certainty consult a qualified lawyer.

Quotations are from the English text of Regulation (EU) 2024/1689 as amended by Reg. (EU) 2026/1744, read in the Publications Office's Cellar on 1 November 2026. The Commission's Article 50 FAQ and guidelines are the official interpretation. EU law as of 2026-11-12; national status as of 1 November 2026.

Which of these duties apply to you?

Free, 60 seconds, instant result with exact citations. No email required.

Run the free check

Sources are cited in place; official text: Regulation (EU) 2024/1689. This guide is a technical explanation, not legal advice.