Article 50 of the EU AI Act (Regulation (EU) 2024/1689) imposes transparency obligations on providers and deployers of AI systems that interact directly with people or produce AI-generated content. The provisions apply from 2 August 2026 and cover virtually every organisation operating a chatbot, generative AI tool, or synthetic media service through a digital channel. Unlike the requirements for high-risk AI systems, these obligations have not been deferred by the AI Omnibus agreement, with one exception: the machine-readable marking obligation under paragraph 2 benefits from a transitional period until 2 December 2026 for systems already on the market before August 2026.

Four obligations in brief

Article 50 contains four distinct obligations, each targeting a different type of system or deployment context:

Paragraph 1 applies to providers of AI systems designed to interact directly with natural persons, such as chatbots and voice assistants. They must design and develop their systems so that users are informed they are interacting with an AI system, unless this is obvious to a reasonably well-informed, observant, and circumspect person given the circumstances and context of use.

Paragraph 2 requires providers of generative AI systems (including general-purpose AI systems producing synthetic audio, images, video or text) to mark outputs in a machine-readable format so that they are detectable as artificially generated or manipulated. The technical solutions employed must be effective, interoperable, robust, and reliable as far as technically feasible.

Paragraph 3 requires deployers of emotion recognition or biometric categorisation systems to inform the individuals exposed to the system of its operation, and to process personal data in compliance with the GDPR.

Paragraph 4 requires deployers using an AI system to generate or manipulate deepfake content to disclose that the content has been artificially created or manipulated. A comparable disclosure obligation applies to AI-generated or AI-manipulated text published with the purpose of informing the public on matters of public interest, unless the content has undergone a process of substantive human review and an identifiable person or organisation bears editorial responsibility for its publication.

What counts as a deepfake?

The Regulation defines a deepfake as image, audio, or video content that bears a significant resemblance to existing persons, objects, places, entities, or events, and would falsely appear to a person to be authentic or truthful. The disclosure obligation applies regardless of any intent to deceive: synthetic media produced without fraudulent purpose falls within scope once this definition is met. For evidently artistic, creative, satirical, or fictional works, a more limited obligation applies: it is sufficient to disclose the existence of the generated or manipulated content in a manner that does not hamper the display or enjoyment of the work.

When and how to inform

Article 50(5) requires that the relevant information be provided to the individuals concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure. The information must conform to applicable accessibility requirements. A reference buried in terms and conditions or displayed in small print at the bottom of a page does not satisfy this requirement. For chatbots, this means a visible notice in the opening message or at the point of first contact; for AI-generated images or video, a visible label attached to the content itself.

Exemptions for law enforcement and editorial responsibility

The obligations in Article 50 do not apply where the use is authorised by law for the detection, prevention, investigation, or prosecution of criminal offences, subject to appropriate safeguards for the rights and freedoms of third parties. Systems made available to the public for the purpose of reporting a criminal offence remain subject to the disclosure obligation in paragraph 1.

For AI-generated text on matters of public interest, an exemption applies where the content has undergone a process of substantive human review and an identifiable person or organisation bears editorial responsibility. Spell-checking or a cursory editorial sign-off is insufficient; the review must be genuine and must specifically address the substance of the text.

Machine-readable marking and the Code of Practice

Paragraph 2 requires that outputs of generative AI systems be marked in a machine-readable format. The European Commission's AI Office is facilitating the development of a voluntary Code of Practice to assist providers and deployers in meeting this obligation in practice. A second draft of the Code was published in March 2026; the final version is expected in June 2026, ahead of the 2 August 2026 application date. The Code is voluntary, but is likely to be used by supervisory authorities and courts as a reference when assessing compliance. Elements under development include a standardised EU label for AI-generated content, a distinction between fully AI-generated and AI-assisted content, and technical standards for watermarking and metadata.

Enforcement and fines

Non-compliance with the transparency obligations of Article 50 may result in an administrative fine of up to €15 million or, if higher, 3% of total worldwide annual turnover. In the Netherlands, enforcement of the AI Act will be assigned to designated national supervisory authorities. In April 2026, the Dutch government opened a legislative proposal for public consultation to establish the national supervisory framework.

What organisations can do now

Organisations deploying AI through digital channels can take the following steps to prepare for 2 August 2026:

  • Map which AI systems fall within the scope of Article 50: chatbots, generative tools, emotion recognition systems, and deepfake generators.
  • Determine for each system whether a provider obligation (paragraphs 1 or 2) or a deployer obligation (paragraphs 3 or 4) applies.
  • Design visible, accessible notices and labels that are displayed no later than at the point of first contact.
  • Assess whether AI-generated text on matters of public interest has undergone substantive human review before relying on the editorial responsibility exemption.
  • Monitor the final Code of Practice and any guidelines published by the European Commission once available.