Article
AI Act Article 50: when you have to say it was made with AI
Article 50 of the EU AI Act is the transparency article, and it has applied since 2 August 2026. It does not ask a company to label everything AI touches. It names four situations, says who owes the information in each, and says how and when it must be given.
The four situations
| Use | What is owed | Unless |
|---|---|---|
| A system meant to interact directly with people, such as a chatbot | It is designed so those people know they are dealing with an AI system | That is obvious to a reasonably well-informed and observant person in the circumstances |
| Emotion recognition or biometric categorisation | The deployer informs the people exposed to it | - |
| Image, audio or video that is a deep fake | The deployer discloses that it was generated or manipulated | - |
| AI-generated text published to inform the public on matters of public interest | The deployer discloses it | The text went through human review and someone holds editorial responsibility |
The information must be given clearly and distinguishably, at the latest at the first interaction or exposure. The verified wording and its sources.
What counts as a deep fake
The regulation defines a deep fake 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. A clearly stylised illustration is not that. A synthetic video of a real colleague announcing a policy is. The definition.
The exception most companies care about
The text rule turns on something a company controls. AI-generated text published to inform the public on matters of public interest does not need a disclosure if it went through human review or editorial control and a person or company holds editorial responsibility for publishing it.
The Commission's guidelines on Article 50, which are not binding, set the bar for that review:
- it is a deliberate examination of the substance by someone with relevant knowledge and professional judgement, and fact-checking is the minimum
- spell-checking, grammar correction, the mere existence of an editorial policy, automated review or a cursory approval do not count
- a substantive AI change made after the review voids the exception
The same guidelines place a company's advertising and product descriptions outside the public-interest text rule, unless they make claims about matters such as health, consumer safety or sustainability. A fixed out-of-office reply is not an AI system at all.
What it costs to get wrong
Breaches of the Article 50 transparency duties are subject to administrative fines of up to EUR 15 000 000 or 3% of total worldwide annual turnover, whichever is higher. For SMEs, including start-ups, it is whichever is lower. Member states set the rules, which can also include warnings and non-monetary measures. The penalty provisions.
Where this fits
Disclosure is easy once you know which of your uses fall into which row of the table above, and that is an inventory question. The AI Act File starts with that inventory and keeps the decision for each use in writing, so that the next new feature reopens the right page instead of nobody's.