Transparency obligations for certain AI systems

Article 50 of the EU AI Act establishes transparency obligations that apply regardless of an AI system's risk classification. They target five distinct obligations across specific functional scenarios: AI-to-human dialogue, synthetic content generation, emotion recognition or biometric categorisation, deepfake production, and the generation of AI text on matters of public interest. Each obligation falls on a specified actor — providers or deployers — and each carries its own carve-outs 32024R1689 Article 50@2024-06-13.

Who is subject to these obligations?

The obligations in Article 50 apply to providers and deployers whose systems fall into specified functional categories — not to all AI systems, and not only to systems classified as high-risk under Chapter III. This makes Article 50 cross-cutting: compliance is required even where a system sits outside the high-risk classification 32024R1689 Article 50@2024-06-13.

1. Disclosure of AI interaction (paragraph 1)

Providers of AI systems designed to interact directly with natural persons must ensure those persons are informed they are interacting with an AI system. The obligation applies unless the AI nature of the interaction 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" 32024R1689 Article 50@2024-06-13.

A carve-out exists for systems authorised by law to detect, prevent, investigate, or prosecute criminal offences, subject to appropriate safeguards for the rights and freedoms of third parties — with a sub-exception that restores the disclosure duty where such systems are available to the public for reporting crimes 32024R1689 Article 50@2024-06-13.

2. Machine-readable marking of synthetic content (paragraph 2)

Providers of AI systems — including general-purpose AI systems — that generate synthetic audio, image, video, or text must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated 32024R1689 Article 50@2024-06-13. Technical solutions must be effective, interoperable, robust, and reliable as far as technically feasible, accounting for the specificities and limitations of different content types, implementation costs, and the generally acknowledged state of the art, as may be reflected in relevant technical standards 32024R1689 Article 50@2024-06-13.

Two exemptions apply: to the extent 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; and systems authorised by law for criminal enforcement purposes 32024R1689 Article 50@2024-06-13.

3. Emotion recognition and biometric categorisation notice (paragraph 3)

Here the obligation falls on deployers, not providers. Deployers operating emotion recognition or biometric categorisation systems must inform the natural persons exposed to such systems of the system's operation, and must process personal data in accordance with applicable EU data protection law 32024R1689 Article 50@2024-06-13. An exemption applies for systems permitted by law to detect, prevent, or investigate criminal offences (note: unlike paragraphs 1, 2, and 4, the scope here does not extend to prosecute), subject to appropriate safeguards for the rights and freedoms of third parties and in accordance with Union law 32024R1689 Article 50@2024-06-13.

4. Deepfake disclosure and AI-generated public-interest text (paragraph 4)

Deployers of systems that generate or manipulate image, audio, or video content constituting a deepfake must disclose that the content has been artificially generated or manipulated 32024R1689 Article 50@2024-06-13. The criminal-enforcement exemption applies. Where the content forms part of an evidently artistic, creative, satirical, fictional, or analogous work or programme, the transparency obligation is reduced to disclosing the existence of such generated or manipulated content in an appropriate manner that does not hamper display or enjoyment of the work 32024R1689 Article 50@2024-06-13.

Deployers of systems generating or manipulating text published with the purpose of informing the public on matters of public interest must likewise disclose that the text has been artificially generated or manipulated 32024R1689 Article 50@2024-06-13. This obligation does not apply where the use is authorised for criminal enforcement, or where the AI-generated content has undergone human review or editorial control and a natural or legal person holds editorial responsibility for the publication 32024R1689 Article 50@2024-06-13.

5. Timing and accessibility (paragraph 5)

All information required under paragraphs 1–4 must be provided in a clear and distinguishable manner at the latest at the time of first interaction or exposure, and must conform to applicable accessibility requirements 32024R1689 Article 50@2024-06-13.

6. Relationship to other obligations (paragraph 6)

Article 50's transparency requirements are without prejudice to the obligations in Chapter III and do not displace other transparency obligations under Union or national law applicable to deployers 32024R1689 Article 50@2024-06-13.

7. Codes of practice and implementing acts (paragraph 7)

The AI Office is charged with encouraging and facilitating codes of practice at Union level to support implementation of detection and labelling obligations. The Commission may adopt implementing acts to approve such codes in accordance with the procedure laid down in Article 56(6), or — where a code is deemed inadequate — to specify common rules via the examination procedure laid down in Article 98(2) 32024R1689 Article 50@2024-06-13.

This block connects to your organisation's own AI governance policy. In the full product it shows, cited to your policy, how YOUR organisation implements the regulation above — private to your organisation. (Demo placeholder.)

The cases below are labeled illustrative context only; they are not regulatory authority.

Synthetic content obligations entering force (EU, July 2026): Italian media reported that as of 2 August, EU AI Act transparency obligations — including technical marking requirements for synthetic audio, image, video, and text — began applying in practice, with the European Commission issuing guidelines on machine-readable marking for synthetic content CASE-1. The timing corresponds to Article 50's entry into application and confirms that the paragraph 2 marking obligation is operationally live, not prospective.

State-attributed deepfake of Indian Prime Minister Modi (India, July 2026): A deepfake video of Prime Minister Modi circulated virally on social media. Indian government sources attributed its spread to accounts described as running Pakistani propaganda and initiated fact-checking and takedown efforts CASE-2. The incident is a direct illustration of the scenario Article 50 paragraph 4 addresses: manipulated video content disseminated without disclosure, intended to mislead the public about a sitting head of government.

Deepfake of Indian Minister Piyush Goyal deployed in protest context (India, July 2026): A digitally altered video of Commerce Minister Piyush Goyal surfaced showing him appearing to make threatening remarks against students involved in nationwide protests over examination policy. Government sources confirmed the video was fabricated and reported social media links were removed CASE-3. This case sits at the intersection of two paragraph 4 duties: the deepfake disclosure obligation (manipulated video) and the public-interest text dimension (content plainly intended to inform — or misinform — the public on a matter of active national political controversy), which is exactly the category Article 50 reaches with its separate AI-generated text disclosure duty.