Skip to main content

Article 50: Transparency & Disclosure Requirements

Article 50 of the EU AI Act: who must disclose chatbots, mark AI-generated content and label deep fakes, since when, with exceptions and fines.

Providers and deployers must disclose AI use (Article 50). This covers systems that interact with people, generate synthetic content, recognise emotions or make deep fakes.

Since 2 August 2026, the duties apply to AI systems in the EU, whatever their risk class. Providers of generative AI systems placed on the market earlier have until 2 December 2026 to mark their output. Fines reach EUR 15 million or 3% of worldwide annual turnover, whichever is higher (Article 99). For SMEs and small mid-caps, the cap is the lower amount. Next step: find which of the four duties covers your system.

Article 50(1) to (4) set four duties: two for providers, two for deployers

Providers: tell people they are dealing with an AI system

AI systems intended to interact directly with people, such as chatbots, must be designed so that people know they are dealing with an AI system. This is not needed when it is obvious to a reasonably well-informed, observant and circumspect person, given the circumstances and context of use. AI systems that the law authorises to detect, prevent, investigate or prosecute criminal offences are exempt, with safeguards for the rights of third parties. The exemption does not cover systems the public can use to report a criminal offence (Article 50(1)).

Providers: mark generated content so machines can detect it

Providers of AI systems that generate synthetic audio, image, video or text, general-purpose AI systems included, must mark the output in a machine-readable format. The marking must make the output detectable as artificially generated or manipulated. The technical solution must be effective, interoperable, robust and reliable as far as technically feasible, taking costs and the state of the art into account. The duty does not apply to systems that only assist with standard editing or do not substantially change the input data or its meaning. It also does not apply where the law authorises the use to detect, prevent, investigate or prosecute criminal offences (Article 50(2)).

Deployers: inform people exposed to emotion recognition or biometric categorisation

An emotion recognition system identifies or infers emotions or intentions from biometric data (Article 3(39)). A biometric categorisation system assigns people to specific categories based on biometric data (Article 3(40)). A system is not one if the categorisation is ancillary to another commercial service and strictly necessary for objective technical reasons. Deployers of these systems must inform the people exposed to them that the system is operating. They must process personal data under the GDPR (Regulation (EU) 2016/679), Regulation (EU) 2018/1725 and Directive (EU) 2016/680, as applicable. Systems that the law permits for detecting, preventing or investigating criminal offences are exempt, with safeguards for third parties (Article 50(3)). Two bans sit beside this duty. Emotion recognition at work and in education is banned, except for medical or safety reasons (Article 5(1)(f)). Biometric categorisation that infers race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation is banned, with limited exceptions (Article 5(1)(g)).

Deployers: disclose deep fakes and AI-generated text on public-interest matters

A deep fake is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful (Article 3(60)). Deployers must disclose that such content is artificially generated or manipulated. For an evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing that such content is present. The disclosure must not hamper the display or enjoyment of the work. Deployers must also disclose AI-generated text published to inform the public on matters of public interest. This does not apply if humans reviewed or edited the text and a person holds editorial responsibility for the publication. Uses authorised by law to detect, prevent, investigate or prosecute criminal offences are exempt (Article 50(4)).

Disclosures must be clear and reach people at first contact

The information must reach the people concerned in a clear and distinguishable manner, at the latest at the time of their first interaction or exposure. It must meet the applicable accessibility requirements (Article 50(5)). Article 50 does not replace the requirements for high-risk systems or other transparency rules in Union or national law (Article 50(6)).

The Commission encourages codes of practice on detecting, marking and labelling AI-generated content. If it finds a code inadequate, it may adopt an implementing act with common rules (Article 50(7)).

The duties apply since 2 August 2026; marking has a grace period to 2 December 2026

Article 50 applies from 2 August 2026 (Article 113). Providers of generative AI systems placed on the market before that date must take the necessary steps to comply with Article 50(2) by 2 December 2026. This grace period comes from Article 111(4), inserted by Regulation (EU) 2026/1744.

The grace period covers only the marking duty in Article 50(2) and only systems placed on the market before 2 August 2026. The duties in Article 50(1), (3) and (4) have no grace period.

Some systems are outside Article 50

The Act does not apply to AI systems where and in so far as they are used exclusively for military, defence or national security purposes. The same holds for placing them on the market or putting them into service for those purposes (Article 2(3)). It does not apply to AI systems or models specifically developed and put into service for the sole purpose of scientific research and development (Article 2(6)). It does not apply to research, testing or development of AI systems or models before they are placed on the market or put into service (Article 2(8)). Testing in real-world conditions is not covered by that exclusion. It does not bind deployers who are natural persons using AI systems in the course of a purely personal non-professional activity (Article 2(10)). For AI systems that are high-risk under Article 6(1) and relate to Annex I Section B products, the Act applies only in part (Article 2(2)). Only Article 6(1), Article 60a and Articles 102 to 112 apply. Articles 57 to 59 apply only in so far as the high-risk requirements are integrated in that legislation. Free and open-source AI systems stay subject to Article 50 (Article 2(12)).

Fines reach EUR 15 million or 3% of worldwide turnover

Breaches of Article 50 can be fined up to EUR 15 million or, for an undertaking, up to 3% of its total worldwide annual turnover, whichever is higher. The legal basis is Article 99(4)(g). For SMEs, including start-ups, and for small mid-caps (companies that have outgrown the SME definition, Article 3(14b)) the cap is the lower amount (Article 99(6) and (6a)). A small mid-cap enterprise is defined in Article 3(14b) by reference to point 2 of the Annex to Recommendation (EU) 2025/1099, which sets upper size limits.

For two kinds of AI systems, the AI Office alone supervises and enforces. These are systems that a provider builds on its own general-purpose AI model, and systems of very large online platforms and search engines. Exceptions to the first kind are listed in Article 75(1)(a). The AI Office is competent for deployers only if they are also the provider.

What Scanara does for Article 50

A guided Article 50 assessment

Scanara walks you through the four duties for your system and records your answers. You decide which duties apply, with counsel where the stakes warrant.

Rule-based code checks

Scanara’s code scan applies rules for disclosure mechanisms in chat interfaces and generative AI calls. It detects standard framework patterns and can miss non-standard code.

Disclosure document templates

Scanara generates Article 50 disclosure documents from templates, partly pre-filled from your answers. You complete them and put the disclosures into your product.

Common questions: internal chatbots, “obvious” and watermarks

Does Article 50 apply to internal chatbots used only by employees?

Yes, if the chatbot is an AI system intended to interact directly with natural persons, employees included. The disclosure is not needed when it is obvious to a reasonably well-informed, observant and circumspect person, given the circumstances and context of use (Article 50(1)). A clear label in the interface helps, but the test is what is obvious to that person.

What counts as “obvious from the circumstances and context of use”?

The Act sets one test (Article 50(1)). It must be obvious from the point of view of a reasonably well-informed, observant and circumspect natural person, taking into account the circumstances and the context of use. It lists no examples. How market surveillance authorities and courts apply the test is open, so disclose when in doubt.

Must machine-readable marking survive editing or compression?

Article 50(2) requires solutions that are effective, interoperable, robust and reliable as far as technically feasible. It takes into account the limits of different content types, costs and the state of the art. It names no technique and no survival test, so document which technique you chose and how you tested it.

Must a system placed on the market before 2 August 2026 be marked?

Yes, but later: by 2 December 2026 (Article 111(4)). This applies to providers of generative AI systems placed on the market before 2 August 2026, who must comply with Article 50(2). Systems placed on the market from 2 August 2026 must comply when placed.

Read the detail behind related duties

Scanara prepares evidence from your repositories

Scanara reads your connected GitHub repositories and runs rule-based scans for standard AI and machine-learning patterns. It can miss non-standard code. Separately, it generates draft documentation from templates and your assessment answers as evidence for your compliance work. You make the classification decisions.