Article 50 applies: which of your people need to recognise it

Article 50 of the EU AI Act has applied since 2 August 2026. Most organisations handled that as a legal question: which paragraph applies, are we the provider or the deployer, what does the guidance say.

That work matters. It is also not where compliance fails. It fails at the moment a marketer publishes a generated image without a label, or a service agent lets a customer talk to a bot that calls itself a digital assistant. Neither person did anything reckless. Nobody told them the rule existed.

Four duties, and the people behind them

Article 50 splits four transparency duties across two roles. Translate them into who in your organisation actually acts.

Someone has to know a chatbot must say it is AI. That is a product decision, so it lands with product and engineering. But it is service teams who notice when a bot has been rebranded to something friendlier and the disclosure quietly disappeared.

Someone has to mark generated audio, image, video or text in a machine-readable format. That is technical, and it sits with whoever ships the generative feature.

Someone has to inform people when emotion recognition or biometric categorisation is used on them. That is process, and it usually lands with operations or HR.

Someone has to disclose deepfakes and certain public-interest text. That is publication, which means marketing and communications, every single time they post.

The pattern is that three of the four are executed by people who do not read regulations. If your literacy programme only reached compliance and legal, the duty is documented and unmet.

What each role needs to be able to do

Not to know the article number. To recognise a situation.

A marketer needs to notice that the image they just generated for a campaign is a person who does not exist, and that this is the category the law is about. A service lead needs to notice that "digital assistant" describes a function and not the nature of a system. A content editor needs to know when a human taking editorial responsibility after substantive review removes the labelling duty for text, and when it does not.

That is role-based literacy, and it is measurable. You can put a marketer in a scenario with four generated assets and see which ones they flag.