EU AI Act Article 50: what the Commission's May 2026 transparency guidance means for your website

What is the European Commission's draft about?

On 8 May 2026, the European Commission published draft guidelines on the transparency obligations in the AI Act (Regulation (EU) 2024/1689), Article 50. The guidelines are open for consultation until 3 June 2026 and are themselves non-binding — they are the Commission's first consolidated interpretation of Article 50. The transparency obligations in Article 50, however, become applicable on 2 August 2026 regardless of how the final guidance is worded. Here is what they mean — especially if you run a website.

What Article 50 of the AI Act requires

Article 50 is about transparency: people should know when they are dealing with AI. The obligations fall on different parties depending on the situation — sometimes the party that provides the AI system (the provider), sometimes the party that uses it (the deployer). Three parts matter for most websites:

• Tell people it's an AI. If you have a chatbot or AI assistant on your site, visitors must be informed that they are interacting with an AI system — not a human. Under the regulation the formal obligation sits with the provider of the AI system, but in practice it is you as the website owner who needs to make sure the disclosure is actually visible to visitors — whether the chat is built in-house or bought from an external supplier. There is an exception: if it is already obvious to a reasonably well-informed and observant person that they are dealing with an AI, no separate disclosure is needed. But for a chatbot that mimics a human agent it is rarely obvious — so the disclosure obligation applies.
• Label AI-generated content. Providers of generative AI must mark the system's output in a machine-readable format so it can be recognised as artificially generated or manipulated. That obligation rests with the party that provides the generative system — not with you when you simply use a finished tool.
• Label deepfakes. Anyone who uses AI to generate or manipulate images, audio or video depicting real people or events (deepfakes) must disclose that the content is artificially generated or manipulated. That duty lies with the party publishing the content (the deployer). Where the content is part of an evidently artistic, creative, satirical or fictional work, the requirement is lighter — a disclosure that does not spoil the experience of the work is enough. Law-enforcement authorities also have specific exemptions.

Article 50 also covers emotion-recognition and biometric-categorisation systems: there, the party using the system (the deployer) must inform the people exposed to it. This rarely affects ordinary websites.

Timeline: when does Article 50 start to apply?

• 8 May 2026 — the Commission publishes the draft guidelines.
• 3 June 2026 — the consultation on the draft closes.
• 2 August 2026 — the transparency obligations in Article 50 become applicable.

A provisional agreement between the Council and the Parliament on 7 May 2026 — a so-called "Digital Omnibus" — could give generative AI systems already on the market a transition period until 2 December 2026. This has not yet been finally adopted and may change, so treat it as a possible exception in the making, not as settled law. It does not alter the fact that Article 50 fundamentally becomes applicable on 2 August 2026.

Do I have to tell visitors they're chatting with an AI?

For most websites it is the first point that matters: if you have a chatbot or AI assistant, visitors must be told it is an AI. A clear line in the chat window is usually enough — but it actually has to be there, and the information must be given no later than the first interaction. It is easy to miss when a chat widget is added by the marketing team or an external supplier without anyone thinking about the disclosure duty. If you also use generative AI to create images, audio or video depicting real people or events and publish them, the deepfake disclosure above applies on top.

Quick answers to common questions

• Do I have to label my chatbot? If a visitor might think they are talking to a human: yes. If it is already obvious that it is an AI, no separate disclosure is needed — but it is rarely obvious for a chat that mimics a human agent.
• Does it apply if the chat comes from an external supplier? The formal responsibility for the system itself sits with the provider, but you as the website owner still need to ensure the disclosure is actually shown to your visitors.
• When does Article 50 start to apply? On 2 August 2026. The Commission's guidance is a draft and may change after the consultation, but the date the obligations themselves apply is fixed.

How CompliantHQ helps

The AI Act will be the next compliance area in CompliantHQ. Our AI Act module is on the way — it will detect AI chats and assistants on your website and flag cases where the visitor isn't told they are talking to an AI (Article 50.1). It is not live yet. What we already check today, in the same scan, is cookies and tracking, accessibility, and your policies and terms.

Read the Commission's official page on the draft and the consultation here: Consultation on draft guidelines on transparency obligations under the AI Act.

Want to know where your website stands today? Run a free scan to see your current findings across cookies, accessibility and policies — and get a head start before the AI Act.