EU AI Act · Article 50 · in force since 02.08.2026

Article 50 of the EU AI Act: the transparency obligations, explained

Article 50 requires you to tell people when they are dealing with AI, to mark AI-generated content so machines can detect it, and to disclose deepfakes and AI-written public-interest text. It has applied since 02.08.2026, it reaches almost every organisation using generative AI with the public, and it applies whatever the risk tier of the system.

Last reviewed · against the consolidated text and Commission guidance

02.08.2026
in force — not deferred by the Digital Omnibus
4
duties: interaction, marking, emotion recognition, deepfakes
3%
or €15m — the Art. 99 fine ceiling
The four duties

What does Article 50 require?

Two duties fall on the provider of the system, two on the deployer. Most organisations owe one or two; a company that builds and runs its own generative product can owe all four.

Art. 50(1)Providers

Tell people they are talking to an AI

A system that interacts directly with people — a chatbot, a voice assistant, an automated agent in a support queue — must be designed so the person knows it is AI, unless that is obvious to a reasonably well-informed person from the context.
Art. 50(2)Providers

Mark AI-generated content

Generative systems must mark synthetic audio, image, video and text in a machine-readable format so that it can be detected as artificially generated or manipulated — watermarks, metadata, provenance signals — as far as the state of the art allows.
Art. 50(3)Deployers

Disclose emotion recognition and biometric categorisation

Deployers of an emotion-recognition or biometric-categorisation system must inform the people exposed to it, and process their personal data under the GDPR. (Several of these uses are prohibited outright under Article 5 — check that first.)
Art. 50(4)Deployers

Disclose deepfakes and AI-written public-interest text

Deployers who generate or manipulate image, audio or video that resembles real people, places or events must disclose it. AI-generated text published to inform the public on matters of public interest must be disclosed as such, unless a human reviewed it and someone holds editorial responsibility.
Scope

Who does Article 50 apply to?

Anyone who provides or deploys AI in the EU that talks to people, generates content, reads emotions or produces deepfakes. The risk tier does not matter; the role does.

If you…You are the…You owe
build a chatbot or assistant, including on a licensed model, and offer it to peopleProviderArt. 50(1) disclosure by design
build or brand a system that generates image, audio, video or textProviderArt. 50(2) machine-readable marking
use a vendor's emotion-recognition or biometric-categorisation tool on staff or customersDeployerArt. 50(3) inform the people exposed — after an Art. 5 check
publish generated media of real people, places or events, or AI-written public-interest textDeployerArt. 50(4) visible disclosure
use a vendor's chatbot on your website under the vendor's nameDeployerThe provider designs the disclosure; you keep it switched on and visible

The provider/deployer line is the whole question. Provider or deployer walks it, and the EU AI Act guide covers the four roles.

Timeline

When did Article 50 apply?

DateWhat happened
02.08.2026Article 50 in force. The transparency duties took legal effect across every Member State. Not deferred by the Digital Omnibus.
27.07.2026Regulation (EU) 2026/1744 in force. The Digital Omnibus moved the high-risk dates and added two Article 5 prohibitions; Article 50 was left where it was.
02.12.2026Two new Article 5 prohibitions apply — AI nudification tools and AI-generated child sexual abuse material. Deepfake disclosure under Art. 50(4) does not make either of those lawful.
Enforcement

What are the penalties for breaching Article 50?

Article 50 sits in the middle band of Art. 99: up to €15m or 3% of worldwide annual turnover, whichever is higher. For SMEs and start-ups the lower of the two applies (Art. 99(6)). National market-surveillance authorities enforce it; the fine is a ceiling, weighed against the nature of the breach and what you did about it.

The work

How to comply with Article 50

Five steps, none of them a project. The hard part is finding every place AI meets a person, not writing the notice.

  1. 01Inventory every AI touchpointList the systems that talk to people, generate content, read emotions or produce media — including features inside tools you licence. Most organisations find the duties in the marketing stack before they find them in the product.
  2. 02Sort each one by paragraph50(1) and 50(2) fall on the provider of the system; 50(3) and 50(4) on the deployer. A company that builds its own chatbot on a licensed model is the provider of the chatbot.
  3. 03Write the disclosuresA short, plain notice where the interaction starts; a caption or label on deepfakes; a note on AI-written public-interest text. The generator below drafts the first one for you.
  4. 04Confirm the marking with your vendorFor 50(2), ask the model or platform provider how its output is marked and keep the answer — the duty is theirs, but you will be asked to show what you rely on.
  5. 05Record itKeep the notice text, where it appears, and when it went live. Article 50 has no registration step, so the record is what shows you met it.
Free tool

Generate your Article 50 disclosure notice

Fill in a few details and get a paste-ready AI-interaction disclosure plus the machine-readable content-marking snippet. Free, no sign-up.

Your inputs
Paste-readyArt. 50in force since 02.08.2026
You're chatting with Aria, an AI assistant. Aria is automated software, not a person — it answers EU AI Act questions. It can be wrong, so check anything important and don't treat its answers as legal advice. Type "agent" at any time to reach a human. — Disclosure shown at the start of every conversation, per EU AI Act Art. 50(1).
Machine-readable content marking (Art. 50(2))
<!-- EU AI Act Art. 50(2) content marking -->
<meta name="ai-generated" content="true">
<script type="application/ld+json">
{
  "@context": "https://veritome.eu/ns/ai-disclosure/v1",
  "aiGenerated": true,
  "aiSystem": "Aria",
  "disclosure": "EU AI Act Art. 50(1)",
  "humanHandoffKeyword": "agent"
}
</script>

Paste-ready copy from Veritome's record templates (Art. 50). A free starting point, not legal advice — start free to keep it on the matching obligation with its evidence.

Straight answers

Article 50: frequently asked questions

What does Article 50 of the EU AI Act require?

Four transparency duties. Providers must design AI that interacts with people to make that clear (Art. 50(1)) and must mark generated content in a machine-readable way (Art. 50(2)). Deployers must tell people when an emotion-recognition or biometric-categorisation system is used on them (Art. 50(3)) and must disclose deepfakes and AI-generated text published on matters of public interest (Art. 50(4)).

Who does Article 50 apply to?

Anyone providing or deploying AI in the EU that interacts with people, generates content, recognises emotions or produces deepfakes — whatever the risk tier. Paragraphs (1) and (2) fall on providers; (3) and (4) on deployers. A business that builds a chatbot on a licensed model is the provider of that chatbot.

When did Article 50 start to apply?

Article 50 has applied since 02.08.2026. The Digital Omnibus (Regulation (EU) 2026/1744, in force since 27.07.2026) deferred the high-risk obligations but did not move Article 50.

Does Article 50 apply to a customer-service chatbot?

Yes, under Art. 50(1). The person must be able to tell they are dealing with an AI system unless it is obvious from the circumstances. A one-line notice at the start of the conversation — and again when a human takes over — is the usual way to meet it.

Do I have to label AI-generated images and text?

Two different duties. The provider of the generative system must mark its output machine-readably (Art. 50(2)). The deployer must visibly disclose a deepfake — content that resembles real people, places or events — and AI-generated text published to inform the public on matters of public interest, unless a human reviewed it and takes editorial responsibility (Art. 50(4)). Ordinary marketing copy drafted with AI and reviewed by a person is not caught by 50(4).

Is Article 50 only for high-risk AI systems?

No. Article 50 applies on top of the risk tier, based on how the AI touches people. A minimal-risk chatbot owes Art. 50(1); a high-risk system that also generates content owes Article 50 as well as the high-risk chapter.

What are the penalties for breaching Article 50?

Up to €15m or 3% of worldwide annual turnover, whichever is higher — the middle band in Article 99. For SMEs and start-ups the lower of the two figures applies (Art. 99(6)).
Article 50 is one obligation of many

See everything else that applies.

Veritome maps every EU AI Act obligation that reaches your systems — transparency, oversight, documentation — and keeps the evidence with the record.