Article 50 of the AI Act: transparency obligations for the use of AI

Article 50 of the AI Act: transparency obligations for the use of AI

From 2 August 2026, the transparency obligations under Article 50 of the AI Act will apply. On 20 July 2026, the European Commission published non-binding guidelines clarifying when chatbots, AI-generated content and deepfakes must be labelled. This series summarises the main points.

Part 1 — What transparency obligations does Article 50 of the AI Act contain?

No general disclosure obligation

Organisations increasingly use artificial intelligence in customer contact, communications, reporting, articles, images and videos. Does this mean that every professional use of AI must now be disclosed?

No. The AI Act does not impose a general obligation to disclose every use of AI. From 2 August 2026, however, specific transparency obligations will apply to certain AI systems and to certain forms of AI-generated or manipulated content.

The Commission guidelines of 20 July 2026 are not legally binding — only the Court of Justice of the European Union can provide an authoritative interpretation — but they do clarify who is responsible, which obligations apply and which exceptions may be available.

What does Article 50 regulate?

Article 50 contains four separate transparency obligations, each aimed at a different type of AI system or output. A fifth paragraph sets out how and when the relevant information must be provided.

Provision	Responsible party	Subject
Article 50(1)	Provider	Direct interaction between an AI system and natural persons
Article 50(2)	Provider	Technical marking and detectability of AI-generated content
Article 50(3)	Deployer	Emotion recognition and biometric categorisation
Article 50(4)	Deployer	Deepfakes and certain publicly available AI-generated texts
Article 50(5)	Provider and deployer	Form, timing and accessibility of the information

The guidelines follow this structure and explain, paragraph by paragraph, how the obligations should be applied in practice.

Classification matters, not the use of AI as such

An internal memo drafted with the assistance of AI does not automatically fall under the same rules as:

• a chatbot that communicates directly with customers;

• an AI system that generates synthetic images;

• an emotion recognition or biometric categorisation system;

• a deepfake;

• a publicly available AI-generated article on a matter of public interest.

The obligations may also apply cumulatively. An AI system that communicates directly with individuals and generates images during that interaction may fall under both Article 50(1) and Article 50(2). If the images are subsequently published as deepfakes, Article 50(4) may also become relevant.

Article 50(1) — Direct interaction with AI

Providers of AI systems that communicate directly with natural persons — such as chatbots, telephone-based AI assistants and AI-powered customer service tools — must design their systems so that it is clear that the user is interacting with AI.

The user must be informed no later than at the start of the interaction. A separate disclosure is not required only where, in the specific circumstances and context of use, it is already evident that the interaction is with an AI system.

Article 50(2) — Technical marking of AI-generated content

Providers of AI systems that generate or manipulate text, images, audio or video must ensure that the output is:

• marked in a machine-readable format; and

• detectable as artificially generated or manipulated.

This obligation does not apply to standard editing functions or situations in which the input, or its meaning, is not substantially altered.

The obligation rests with the provider of the AI system. An organisation that merely uses an existing generative AI tool is therefore not automatically required to develop its own technical marking or detection system.

Article 50(3) — Emotion recognition and biometric categorisation

Deployers of emotion recognition or biometric categorisation systems must inform the individuals exposed to those systems.

This obligation applies alongside the GDPR and other data protection rules. Providing information does not, by itself, make the use of the system lawful.

Article 50(4) — Deepfakes and certain AI-generated texts

For deepfakes, it must be made clear, either visibly or audibly, that the content has been artificially generated or manipulated.

For AI-generated or AI-edited text, the disclosure obligation applies only where the text is published for the purpose of informing the public about matters of public interest.

An exception applies where the text has undergone substantive human review or editorial control and a natural or legal person assumes editorial responsibility for the publication. This exception is discussed in more detail in Part 2 of this series.

Article 50(5) — Clear, timely and accessible information

The information required under Article 50(1) to (4) must:

• be clear;

• be distinguishable from other information;

• be provided no later than at the first interaction or exposure;

• comply with the applicable accessibility requirements, including those relevant to persons with disabilities.

According to the guidelines, the characteristics of the relevant audience should also be taken into account, for example where children, older persons or other potentially vulnerable groups may reasonably form part of the audience.

A general reference in a privacy notice or AI policy is not sufficient on its own. The disclosure must form part of the actual interaction or be presented with the content itself.

Provider or deployer?

The allocation of responsibilities depends on the organisation’s role.

• A provider develops or has an AI system developed and places it on the market or puts it into service under its own name or trademark. Article 50(1) and Article 50(2) apply to providers.

• A deployer uses an AI system under its own responsibility and determines how and for what purpose it is used. Article 50(3) and Article 50(4) apply to deployers.

• Article 50(5) applies to both providers and deployers.

An organisation may perform more than one role, for example where it develops a chatbot and also uses that chatbot in its own customer service.

Transitional period: only for technical marking

Article 50 applies from 2 August 2026. There is no general postponement.

A limited transitional period until 2 December 2026 applies only to the technical marking and detectability obligation under Article 50(2) for certain generative AI systems that were already on the market or in use.

For systems that are both interactive and generative, such as a chatbot that also generates images, the transitional period applies only to the technical marking obligation under Article 50(2). The direct interaction disclosure obligation under Article 50 (1) must already be met by 2 August 2026.

Content generated or manipulated before 2 August 2026 does not have to be labelled retroactively, unless text created before that date is first published after 2 August 2026.

What should organisations assess now?

Organisations should consider the following questions:

• Does an AI system communicate directly with natural persons?

• Does the system generate or manipulate text, images, audio or video?

• Is emotion recognition or biometric categorisation used?

• Are deepfakes or AI-generated texts on matters of public interest published?

• Is the organisation acting as a provider, a deployer, or both?

• Is the required information clear, timely and accessible?

Enforcement and fines

Member States are responsible for establishing the rules for enforcement of the AI Act. For infringements of Article 50, the maximum fine is EUR 15 million or 3% of worldwide annual turnover. For SMEs, the lower of the two maximum amounts applies.

Conclusion to Part 1

Article 50 does not require every use of AI to be disclosed. The first question is therefore not: Has AI been used?

The relevant question is: Which part of Article 50 applies, what role does the organisation have, and which transparency obligation follows from that classification?

In Part 2, we will examine AI-generated texts, human review and editorial responsibility in more detail.

DUFINCO helps organisations identify their AI applications, determine the correct role under the AI Act and assess which transparency obligations apply. Questions about Article 50 or preparing for 2 August 2026? Contact us at info@dufinco.nl.