Transparency obligations under Article 50 of the EU AI Act

What does Article 50 require?

Paragraph	Who	What
(1)	Providers of AI systems that interact with people	Design them so that people are informed they are interacting with an AI, unless this is obvious
(2)	Providers of systems that generate text, images, audio or video	Mark outputs in a machine-readable way as artificially generated
(3)	Deployers of emotion recognition or biometric categorisation	Inform the people affected
(4)	Deployers who publish deepfakes or AI texts	Disclose that content has been artificially generated or manipulated

The information must be clear and distinguishable, given at the latest at the first interaction, and meet accessibility requirements (paragraph 5).

Since when does the obligation apply?

Since 2 August 2026. The Digital Omnibus (Regulation (EU) 2026/1744) has not postponed this. It only gives providers of systems already on the market before 2 August 2026 until 2 December 2026 for the machine-readable marking under paragraph 2 (Article 111(4)).

When do texts have to be labelled?

Paragraph 4 covers texts published to inform the public on matters of public interest, such as press releases, public notices or articles on your website. The obligation does not apply if

• a person has reviewed or edited the text and
• a person or organisation holds editorial responsibility for the publication.

In practice: an AI draft that a specialist reads, corrects and approves generally does not have to be labelled if your organisation takes responsibility for the publication. Record who reviews and approves. Internal texts and letters to individual people are not covered.

For deepfakes, meaning images, audio or video showing realistic-looking people, places or events, this editorial exception does not exist. For evidently artistic or satirical works, a notice that does not spoil the enjoyment of the work is sufficient.

What applies to chatbots?

A chatbot on your website, in your citizen portal or in patient information must make clear that an AI is answering. Under paragraph 1, the obligation falls on the provider of the system. Anyone who puts a chatbot into service under their own name for citizens, patients or customers can be a provider themselves, even if it is built on a platform. More on this in the guide Do our own AI apps make us a provider?

An internal assistant that staff deliberately open as an AI tool is generally obviously an AI. A short notice in the interface and in the training under Article 4 does no harm, though.

What help is available?

• Code of Practice: in June 2026 the Commission published a voluntary Code of Practice on marking and labelling. It has one part for providers (marking) and one for deployers (labelling). Not signing it does not in itself breach the Regulation, but the obligations must then be met in another way.
• Guidelines: the Commission has published guidelines on Article 50 that explain terms and exceptions and assign the obligations along the supply chain.
• Icons: the Commission provides freely usable icons for labelling.

How does this work with basebox?

basebox is a platform, not a high-risk system. Which obligations you have depends on how you use it. basebox labels AI answers: anyone using basebox is evidently chatting with an AI platform, and a notice to this effect appears below the input and answer window. This covers the information under paragraph 1. If you publish an AI text or deploy a chatbot for citizens or patients, check paragraphs 2 and 4 separately for your case.

Every answer shows its sources with the quoted passage; the specialist checks against the source and is responsible for how the answer is used. This supports the editorial review that paragraph 4 requires. Your administrator sets usage policies and system instructions centrally; they apply to all apps in your organisation. More under Security.

Not legal advice: as of 11 October 2026. The Commission’s guidelines and Code of Practice are being updated, and the classification of individual cases has not yet been settled by the courts.