Article 50 in practice: transparency for customer-facing AI

Article 50 of the EU AI Act sets out transparency obligations for certain AI systems. It is short, but it is often summarised as "label everything made with AI". That is not what it says. This guide walks through the text in practical terms, so your team can tell which questions are relevant to your system and which are not.

It is educational material, not legal advice. Where it quotes the Act or the European Commission, the source is linked at the end with the date we checked it.

Start with the role: provider or deployer

Article 50 places some obligations on providers of AI systems and others on deployers. The AI Act defines both terms, and the Commission's FAQ opens with two questions: who is a provider, and who is a deployer.

In practice, the same company can be both. A business that builds a chat assistant on a third-party model and offers it to customers under its own name may be the provider of that assistant as well as its deployer. That is exactly the kind of assumption a review should write down and check with legal counsel, rather than decide silently.

The four situations Article 50 covers

1. AI systems that interact directly with people (providers)

Article 50(1) asks providers to ensure that AI systems intended to interact directly with natural persons are designed so that people "are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use."

For a customer-facing product, this usually means chat and voice assistants. The practical questions are when the user is told, where, and in what words. There is also an exception for certain systems authorised by law to detect, prevent, investigate or prosecute criminal offences.

2. Synthetic audio, image, video or text (providers)

Article 50(2) asks providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text to ensure that outputs "are marked in a machine-readable format and detectable as artificially generated or manipulated."

This is a technical obligation on the provider, not a visible label for the user. It does not apply "to the extent the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof", or where authorised by law for criminal-offence purposes.

3. Emotion recognition and biometric categorisation (deployers)

Article 50(3) asks deployers of an emotion recognition system or a biometric categorisation system to inform the natural persons exposed to it, and to process personal data in line with EU data protection law. Many customer-facing products do not use these systems. If yours does, this is a narrow and specific case to look at closely.

4. Deepfakes and text on matters of public interest (deployers)

Article 50(4) has two parts.

• Deepfakes. Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake must disclose that the content has been artificially generated or manipulated. For evidently artistic, creative, satirical or fictional works, the obligation is limited to disclosing the existence of such content "in an appropriate manner that does not hamper the display or enjoyment of the work."
• Text. "Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated." This does not apply where the AI-generated content "has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content."

The text rule is narrow. It concerns text published with the purpose of informing the public on matters of public interest. AI-written product descriptions, for example, are generally not covered by it.

How the information must be given

Article 50(5) applies to all four situations: the information "shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure" and "shall conform to the applicable accessibility requirements."

A useful practical test for a user-facing flow: would a typical user notice the information at that moment, on every channel where they meet the system? A statement that exists only in a help article or in the terms of service is worth examining against that wording.

Timing

According to the European Commission, "Article 50 of the AI Act applies as from 2 August 2026."

The Commission describes one limited grace period: "A limited grace period is envisaged only for AI systems placed on the market before 2 August 2026 and only as regards the marking and detection obligation for AI-generated content (Article 50(2) of the AI Act). Providers of such systems must comply with those obligations only as from 2 December 2026."

The FAQ also states: "Content generated prior to 2 August 2026 does not need to be labelled retroactively."

Guidelines and the Code of Practice

The Commission has published guidelines on transparency obligations for providers and deployers of certain AI systems. It describes them as helping "providers and deployers of AI systems and competent authorities in ensuring compliance with the transparency obligations under Article 50 of the AI Act."

There is also a Code of Practice on Transparency of AI-generated content. The Commission's FAQ describes it as "a voluntary practical tool to help providers and deployers" and states that "the code has been assessed as adequate by both the Commission and the AI Board."

On enforcement, the FAQ states that "compliance with the rules will mainly be enforced by national competent market surveillance authorities."

Common misunderstandings

• "Every AI feature needs a label." No. Whether an obligation is relevant depends on the system, the role and the use case, and there are exceptions.
• "Any AI-written text must be disclosed." The Article 50(4) text rule covers text published with the purpose of informing the public on matters of public interest, and has an exception for human review or editorial control with editorial responsibility.
• "Marking and labelling are the same thing." Article 50(2) machine-readable marking is a provider obligation for outputs. Disclosure to people under Article 50(1), (3) and (4) is about what users are told.
• "If we are not the model developer, nothing applies to us." Your role depends on what you do with the system. Write the assumption down and check it.

A practical starting checklist

This is an illustrative starting point for a single customer-facing system, not a complete or authoritative list.

1. Describe the system, the flow and the channels where users meet it.
2. Write down your role assumption (provider, deployer or both) and who will confirm it.
3. List which of the four situations above could be relevant, and which are clearly not, with a reason.
4. Walk the flow as a user and capture evidence of what is shown, where and when.
5. Compare what you captured with the wording of Article 50(5): clear, distinguishable, at the latest at the first interaction or exposure, accessible.
6. Agree prioritised actions and owners, and plan a re-check after release.

If you would like help with steps 3 to 6 for one system, that is what the Article 50 Review is for. You can also look at the sample report to see the output.