Witness
Get Started
Free ToolsPricing
Sign In
Back to Blog
July 14, 2026

Article 50 EU AI Act: Transparency Obligations Step by Step

Article 50 of the EU AI Act applies from 2 August 2026: chatbot disclosure, AI content labelling, deepfakes. The practical six-step guide for SMEs.

Kevin Miller·7 min read·Article 50 EU AI ActAI transparency obligationsAI content labellingchatbot disclosure EUdeepfake labelling

Article 50 of the EU AI Act is the obligation that hits most companies in 2026. Not the high-risk rules, which the Digital Omnibus pushed to December 2027, but the transparency obligations: they apply from 2 August 2026, and they concern every company that runs chatbots, publishes AI content, or uses deepfakes.

This guide walks through implementation in six steps. It is not legal advice, but it shows what to do and in which order.

The four duty groups at a glance

Article 50 consists of four separate obligations. Which ones apply to you depends on what your system does and which role you hold (provider or deployer):

  1. Interaction with humans (Art. 50(1)): providers of AI systems that interact directly with people (typically chatbots) must ensure those people learn they are dealing with an AI. Exception: it is obvious from the circumstances to a reasonably well-informed person.
  2. Synthetic content (Art. 50(2)): providers of AI systems that generate audio, images, video, or text must mark the outputs in a machine-readable format as artificially generated or manipulated. This is a technical requirement on the system, not just a notice.
  3. Emotion recognition and biometric categorisation (Art. 50(3)): deployers of such systems must inform the persons exposed to them. Also relevant: emotion recognition in the workplace and in educational institutions is largely prohibited under Article 5.
  4. Deepfakes and text on matters of public interest (Art. 50(4)): deployers must disclose when image, audio, or video content is a deepfake. AI-generated or manipulated text published to inform the public on matters of public interest must be labelled, unless a human carries editorial responsibility.

Softer disclosure rules apply to artistic, satirical, and fictional works, and there are exemptions for certain law-enforcement purposes.

And if your system is also a high-risk system: the transparency obligations stack on top of the high-risk obligations (Art. 50(6)); they do not replace each other.

Step 1: Inventory your AI systems

You can only label what you know about. Build a list of every AI system in the company, including the inconspicuous cases: the chat widget that came with your website builder, the AI image generation in your marketing tool, the text generator in the CMS. In practice, the list is longer than expected, because AI features ship inside standard software.

Step 2: Clarify your role

Provider or deployer? The answer decides which paragraphs of Article 50 apply to you. Rule of thumb: whoever develops an AI system or places it on the market under their own name is a provider. Whoever uses someone else's system under their own authority is a deployer. Being both at once is possible, for example when you build a purchased model into your own product under your own brand.

Step 3: Map each system to a duty group

Assign every system from step 1 to one or more of the four duty groups, or establish that none applies. Internal use without publication generally does not trigger a labelling duty. An internally used text draft is a different thing from a published AI-written article.

This is exactly the step our free Article 50 checker automates: seven questions, three minutes, no sign-up. At the end you know which duty group applies to you.

Step 4: Implement disclosure

For chatbots and interactive systems this means a clear notice, at the latest at the first interaction. What works in practice: a label right in the chat window ("You are chatting with an AI assistant") rather than a hidden sentence in the terms and conditions. The information must be clear and distinguishable; when in doubt, make it more visible, not more subtle.

The same principle applies to deepfakes and label-required text: disclosure belongs on the content itself, not in a legal notice three clicks away.

Step 5: Verify machine-readable marking

The technical marking obligation for synthetic content primarily hits providers of the generating systems. As a deployer you should still check: does the tool your marketing team uses to generate images mark the outputs in a machine-readable way (for example via metadata or watermarking techniques)? Ask your tool vendor in writing. The answer belongs in your documentation, because that is exactly what enterprise customers will ask about in supplier audits.

Step 6: Document

The underrated step. Record the system inventory, the role mapping, the duty-group mapping, the measures implemented, and the vendor statements. Not only in case a regulator asks. More and more tenders and procurement questionnaires include the question "How do you comply with the transparency obligations under Article 50?". If you can send a document, you are ahead of everyone who has to start writing one. For exactly this purpose we offer our Article 50 template pack with ready-made notice texts, inventory templates, and an implementation checklist.

Deadlines and penalties

  • 2 August 2026: Article 50 becomes applicable.
  • 2 December 2026: end of the transitional window for existing systems. It covers only the content-marking and disclosure duties under Art. 50(2) and (4) (synthetic content, deepfakes) and applies exclusively to AI systems placed on the market before 2 August 2026. Chatbot disclosure (paragraph 1) and the emotion-recognition notice (paragraph 3) are not deferred, and new systems must comply from day one.
  • Penalty framework: infringements of Article 50 can be fined with up to EUR 15 million or 3% of worldwide annual turnover, with SMEs paying the lower of the two. The free penalty calculator shows the ceilings for your revenue. For context: no fine has been issued under the EU AI Act so far. Right now, the practical pressure comes from customers and tenders, not from authorities.

The bottom line

Article 50 is doable. It is not high-risk documentation with a conformity assessment. At its core it means: know where AI is involved, disclose, label, document. A company with one chatbot and AI images in marketing gets the analysis done in a morning.

The first step is free: start the Article 50 checker and know within three minutes whether and how you are affected.

Check if the EU AI Act applies to you

Free classification in 3 minutes. No signup required.

Get Started