EU AI Act Article 50: Transparency for Chatbots and Deepfakes

Transparency Obligations for AI Systems

Article 50 of Regulation (EU) 2024/1689 addresses transparency requirements for AI systems that present specific risks to individuals even when they are not classified as high-risk. These obligations apply to a broad range of AI applications that are increasingly prevalent in daily life, including chatbots, deepfake generators, emotion recognition systems, and general-purpose AI-generated content. The provision reflects the principle that individuals have a right to know when they are interacting with or affected by AI systems.

Unlike the requirements of Chapter III, Section 2, which apply specifically to high-risk AI systems, Article 50 operates independently of risk classification. A chatbot or deepfake generator need not be classified as high-risk to trigger Article 50 obligations. This broader scope ensures that transparency protections extend to AI applications that may not pose the severe risks associated with high-risk systems but that nonetheless affect individuals in ways that warrant disclosure.

The transparency obligations under Article 50 became applicable on 2 August 2025, alongside the provisions on general-purpose AI models. This earlier application date compared to the high-risk system requirements (2 August 2026) reflects the urgency of addressing transparency concerns in areas where AI deployment is already widespread.

Chatbot Disclosure Requirements

Article 50(1) establishes that providers must ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the circumstances and the context of use. This obligation applies to the provider of the AI system and captures a wide range of conversational AI applications including text-based chatbots, voice assistants, and virtual agents.

The disclosure must be made in a clear, timely, and intelligible manner. It should be provided at the outset of the interaction or at the latest at the point where the individual might reasonably believe they are communicating with a human being. The form of disclosure should be appropriate to the context: a text-based chatbot might display a notice in the chat interface, while a voice-based system might include an audio announcement.

The exception for situations where the AI nature of the system is obvious from the circumstances and context of use provides necessary flexibility. Where the design and presentation of the system make it self-evident that the user is interacting with an AI (for example, a clearly labelled automated customer service system on a website), additional disclosure may not be required. However, providers should exercise caution in relying on this exception, as user expectations and perceptions can vary significantly.

Deepfake Labelling Requirements

Article 50(4) addresses the growing concern around deepfakes by requiring that deployers of AI systems that generate or manipulate image, audio, or video content constituting a deepfake disclose that the content has been artificially generated or manipulated. The term deepfake is defined in Article 3(60) as AI-generated or manipulated image, audio, or video content that resembles existing persons, objects, places, entities, or events and would falsely appear to a person to be authentic or truthful.

The disclosure must be made in a manner that is clear and distinguishable, including where applicable through the use of machine-readable markers. This dual requirement serves two purposes: human-readable disclosures inform individuals who encounter the content directly, while machine-readable markers enable platforms, media organisations, and other intermediaries to detect and flag deepfake content at scale.

The labelling requirement applies to the deployer of the AI system rather than the provider. This allocation of responsibility recognises that the same AI tool may be used for both legitimate and problematic purposes, and that the deployer is in the best position to ensure appropriate labelling of the specific content they generate.

Emotion Recognition and Biometric Categorisation Disclosure

Article 50(3) imposes specific transparency obligations on deployers of emotion recognition systems and biometric categorisation systems. Deployers must inform the natural persons exposed to such systems of their operation and must process personal data in accordance with Regulation (EU) 2016/679, Directive (EU) 2016/680, and Regulation (EU) 2018/1725 as applicable.

This requirement addresses the particular intrusiveness of systems that analyse individuals' emotional states or categorise them based on biometric data. Such systems may operate in ways that are not apparent to the affected individuals, making disclosure obligations especially important. The information must be provided to the specific natural persons who are exposed to the system, not merely published as a general notice.

It is important to note that certain uses of emotion recognition and biometric categorisation systems are prohibited entirely under Article 5 of the Regulation. For example, emotion recognition in the workplace and in educational institutions is prohibited except for medical or safety reasons. The Article 50 transparency requirement applies only to uses that are not already prohibited under Article 5.

AI-Generated Content Marking

Article 50(2) requires providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video, or text content to ensure that the outputs of the AI system are marked in a machine-readable format and are detectable as artificially generated or manipulated. This obligation rests on the provider rather than the deployer and applies at the technical level of the AI system itself.

The machine-readable marking requirement is intended to create a technical infrastructure for content provenance. By embedding markers in AI-generated content at the point of creation, the provision enables downstream detection by platforms, media organisations, and verification services. The specific technical standards for such marking are to be developed through harmonised standards and the AI Office's guidelines.

Article 50(2) acknowledges that the technical marking must not compromise the functionality or quality of the output and must be as far as possible technically viable and effective. This pragmatic approach recognises that certain marking techniques may not be feasible for all types of content or may degrade content quality in ways that undermine the utility of the AI system.

Exceptions and Special Provisions

Article 50 includes several important exceptions and qualifications. Article 50(4) provides an exception to the deepfake disclosure requirement where the use is authorised by law for the purpose of detection, prevention, investigation, or prosecution of criminal offences. This exception enables law enforcement and security services to use deepfake detection and generation tools without compromising their operational effectiveness.

A further exception applies where content forms part of an evidently artistic, creative, satirical, fictional, or analogous work or programme, provided that appropriate safeguards for the rights and freedoms of third parties are in place. This exception protects creative freedom and recognises that AI-generated content is increasingly used in legitimate artistic and entertainment contexts where mandatory disclosure might be impractical or artistically counterproductive.

For AI-generated or manipulated text that is published for the purpose of informing the public on matters of public interest, the deployer must disclose that the content is AI-generated, but this disclosure may take into account the editorial standards and processes applicable to the publication. This provision balances transparency with press freedom and editorial autonomy.

Member States may also adopt additional rules on the implementation of Article 50 obligations, provided these are consistent with Union law. The European Commission is empowered to adopt implementing acts specifying the detailed arrangements for the implementation of Article 50, particularly regarding the technical specifications for machine-readable marking and disclosure formats.

This article is for informational purposes only and does not constitute legal advice. Regulatory requirements change frequently — verify current rules with official sources. Built by Sawai Gyoseishoshi Office, Hiroshima, Japan.