AI Act Article 50 for product teams: what compliant AI disclosure looks like
Gideon Awolesi ·
Article 50 of the AI Act has applied since 2 August 2026. If your product has a chatbot, an AI assistant, or anything that generates text, images, audio, or video, it applies to you, including if your company is outside the EU and the output is used in the EU. Law firms have explained the legal text well, so this guide focuses on what compliant disclosure actually looks like in a product.
01The four duties in plain English
| Article | Who | What you have to do |
|---|---|---|
| 50(1) | Providers of AI systems that interact with people | Design the system so people are told they're interacting with AI, unless that's obvious to a reasonably well-informed, observant person. |
| 50(2) | Providers of systems that generate synthetic audio, images, video, or text | Mark the output in a machine-readable format so it can be detected as AI-generated or manipulated. |
| 50(3) | Deployers of emotion recognition or biometric categorisation | Tell the people exposed to the system that it's operating. |
| 50(4) | Deployers of deepfakes, and of AI-generated text published to inform the public on matters of public interest | Disclose that the content was artificially generated or manipulated. |
A provider builds the AI system, and a deployer uses it in their own product or service. A lot of product teams are both, because they build features on top of a model and then show them to their own users.
02Dates and fines
- 2 August 2026: Article 50 applies.
- 2 December 2026: providers of systems already on the market before 2 August 2026 must comply with the machine-readable marking duty in 50(2). This grace period was added by the Digital Omnibus on AI and covers marking only.
- Fines for breaching Article 50 go up to €15 million or 3% of worldwide annual turnover, whichever is higher. For SMEs, it's whichever is lower.
The Commission's FAQ also confirms you don't have to go back and label content generated before 2 August 2026.
03When and how you have to disclose
Article 50(5) says the information must be given "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure," and it has to meet accessibility requirements. The Commission's guidance, published on 20 July 2026, adds a few points that matter for design:
- People must be told from the start of the first interaction.
- The "it's obvious" exception is read narrowly, so don't rely on it because your product has a robot avatar.
- If you deploy someone else's model, you can't rely on the provider's machine-readable marks to cover your own disclosure duty.
04What compliant disclosure looks like in a product
Chat and assistants
Put the disclosure in the chat itself, at the top of the conversation and next to the input, where people look before they type. A line like "You're chatting with an AI assistant. It can make mistakes, so check important answers" does two jobs: it meets the duty, and it sets expectations. A notice buried in your terms of service or a settings page is unlikely to count as "clear and distinguishable".
Generated text, images, and video
Label the content where people see it. For images and video, the EU has published three optional icons: Basic, Fully AI-Generated, and Partially AI-Modified. Using them is optional, but labelling isn't. The Commission's guidance on placement says the label should be visible at first exposure, not hidden by overlays, embedded in the content where possible, and still visible after the content is shared or downloaded.
Accessibility
The disclosure has to be accessible too. Give icons alt text or ARIA labels, use plain language with no abbreviations other than "AI", and keep brief on-screen notices up long enough for people with cognitive or processing difficulties to read them.
Agents that act for the user
When an AI agent sends an email or books something on someone's behalf, the people on the other end are interacting with AI too. Design the message so it's clear an AI sent it, and who it's acting for.
05What's out of scope
- AI that only works in the background, or machine-to-machine, with no direct interaction with people.
- For the marking duty, systems that only perform an assistive function for standard editing, or don't substantially change the input.
- For deepfakes in evidently artistic, satirical, or fictional work, the duty is lighter: you disclose that such content exists in a way that doesn't spoil the work.
- For AI-generated text on matters of public interest, the label isn't needed if the text has had human review or editorial control and someone holds editorial responsibility for it.
06The Code of Practice
The final Code of Practice on marking and labelling AI-generated content was published on 10 June 2026. Signing it is voluntary. Once the Commission and the AI Board assess it positively, signatories can use it to show they comply. Section 1 covers marking for providers, and section 2 covers labelling for deployers.
07A quick check for your product
- List every place your product uses AI that people interact with or whose output they see.
- For each one, check if a reasonably well-informed person would know it's AI from the first moment. If there's any doubt, disclose.
- Check the disclosure is in the interface itself, at first interaction, and accessible.
- If you generate images, audio, video, or text, confirm your marking is in place by 2 December 2026.
- Test the wording with real users, so it informs them without making them distrust a feature that works.
If you want a second pair of eyes, send us one AI flow and one question for our free AI audit, and you'll get a recorded walkthrough and a one-page fix list within five business days.