Article 50: AI transparency, explained

Article 50 is the transparency tier of the Act. It does not ban anything, and it does not impose the heavy duties of high-risk systems. It asks one thing: that people are not misled about when they are dealing with AI, or with content AI produced.

In practice it covers four situations: systems that interact with people, AI-generated or manipulated content, emotion recognition and biometric categorisation systems, and deepfakes and certain AI-generated text published to inform the public.

A system that interacts directly with people, such as a customer-service chatbot or a voice assistant, generally has to make clear that the person is dealing with AI rather than a human.

There is a common-sense limit: you do not need a notice where it is already obvious to a reasonably well-informed person from the context. If in doubt, disclose. A short, plain line is enough.

Where you publish AI-generated or AI-manipulated image, audio, or video that resembles real people, places, or events, a deepfake, you generally have to disclose that it was artificially generated or manipulated.

Providers of the generating systems are separately expected to mark their output as artificial in a machine-readable way where technically feasible. As a deployer, your duty is the clear disclosure to the people who see the content.

If you operate a system that recognises emotions or categorises people using biometric data, you generally have to inform the people exposed to it.

Tread carefully here. Some emotion recognition, such as in the workplace, and some biometric categorisation are not merely transparency matters but are banned outright under Article 5. The transparency duty applies to the uses that remain allowed.

The duties have limits. Disclosure is not required where it is obvious from the context, and there are narrow carve-outs, including for some law enforcement uses and for clearly artistic, creative, or satirical work, where the disclosure is made in a way that does not spoil the piece.

These exceptions are narrower than they sound. For ordinary business uses, the safe assumption is that disclosure applies.

The Article 50 transparency obligations apply from 2 August 2026. The core high-risk obligations were postponed to 2 December 2027 by the 2026 Digital Omnibus, so transparency is now the nearer of the two. That gives most organisations a clear runway to get their notices and labelling in place.

Add a clear notice to any customer-facing chatbot or assistant, and adopt a consistent labelling practice for AI-generated content you publish. Decide who owns this, and keep a simple record of where and how you disclose.

None of it needs to be heavy. For most businesses, a consistent line of text and an internal note of the practice is the bulk of it.

Veillo generates an Article 50 transparency disclosure tailored to the limited-risk systems in your register, so the wording matches the tools you actually run. As with every generated document, it is a starting point to review for your situation.