Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act — C(2026) 5054 final, Annex

(32) Article 50(1) AI Act obliges providers of AI systems directly interacting with natural persons to ensure that their AI systems (including AI agents and GPAI systems) are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system.

(33) While the mechanism to inform should be embedded in the design and development of the system and in the way the system interacts with the natural persons, notification should be provided during the operation of the AI system and at the latest at the time of the first interaction with the natural person (as required in Article 50(5) AI Act, see Section 7 below). Providers are responsible for ensuring that, after the AI system’s placement on the market or putting into service, natural persons interacting with the AI system are effectively and consistently informed throughout the lifecycle of the AI system, as placed on the market or put into service by that provider.2.2

(34) Format of disclosure: Article 50(1) AI Act does not specify any particular disclosure technique. Providers can therefore choose any appropriate technique, so long as two requirements are met: (i) the requirements set out in Article 50(5) AI Act are fulfilled regarding the provision of clear and distinguishable information at the latest at the time of the first interaction (see Section 7 below); and (ii) the characteristics of natural persons belonging to vulnerable groups due to their age (e.g. children or the elderly) or disability are taken into account to the extent that the AI system is intended or reasonably likely to interact with natural persons belonging to those groups (e.g. including when the AI system is reasonably likely to interact with the general public) (19). The information must also be provided in an accessible format for natural persons with disabilities, as specified in Article 50(5) and Recital 132 AI Act (see Section 7). Where the system may interact with children, those notifications must be child-friendly, age-appropriate, easy-to- understand and easily accessible to all children, including those with disabilities or additional accessibility needs. Where available and appropriate, providers are

(19) See Recital 132 AI Act. This does not require that the system is targeted at those vulnerable groups. If a system is intended to interact with any member of the public, then members of the vulnerable groups are included as well and should be considered by providers when implementing the notification measures.

recommended to use standardised disclosure practices and labels to ensure effectiveness and consistency of the notifications and improved user awareness.

(35) Substance of the notification: Article 50(1) AI Act requires providers of interactive AI systems to design and develop them in a manner that all natural persons interacting with such AI systems are explicitly informed about the artificial, non-human nature of the interacting counterpart. If it is reasonably foreseeable to the provider that the AI system may interact with natural persons belonging to the above-mentioned vulnerable groups, then the information should be adapted accordingly and provided in an accessible format, as required under applicable Union law to also ensure the full understanding of those natural persons (20).

(36) The notifications should provide information on the artificial origin of the interaction in ways that are appropriate to the context in which the interaction takes place. This should be done in a manner that ensures that the persons concerned are properly and effectively informed, avoiding any risks of deception or manipulation before and throughout the interaction, while ensuring feasible and proportionate disclosures. Examples of providing the information include notifications in writing (e.g., a chatbot that starts a conversation by mentioning that it is based on AI technology), visual means (e.g. an email generated by an AI agent sent to a natural person that features an AI label at the top), auditory means (e.g. a voice assistant that says at the beginning of a session that it is powered by AI), disclosure of AI identifiers, and credentials (e.g. AI agents that disclose their AI identity to the extent feasible in a verifiable manner (21).

(37) Format of the notifications: The design of the AI system with multimodal and accessible disclosure approaches is recommended as a best practice. Such notifications may combine different techniques to reinforce user understanding, where appropriate and depending on the context and the target and foreseeable audience groups and the available resources of the providers, in particular Small and Medium Sized Enterprises (SMEs). Examples of appropriate techniques may include:

- Textual disclosure (User Interface (UI)-based): Prominent, plain-language labels or banners (e.g. “You are interacting with an AI system”) and first-turn greetings in chatbots that may be combined, as appropriate, with persistent badges or labels visible throughout the interaction as proportionate to the specific context and risks. Furthermore, it is recommended to position accompanying disclosures close to the interaction interface (e.g. near the input/output field) and use simplified wording, particularly for users with lower digital literacy and children.

- Auditory disclosure: In voice-based or telephony contexts, explicit spoken statements at the beginning of the interaction (e.g. “This is an AI-powered assistant”) combined, as appropriate, with periodic reminders in longer interactions, in particular in case of interruptions or a change of the role of the AI system during a user journey. Distinct audio cues (e.g. tones or earcons) may support recognition, particularly for visually impaired users, but are not considered sufficient by themselves.

(20) See Section 7.3 for applicable accessibility requirements.

(21) For example, electronic attestations of attributes as established under Regulation (EU) No 910/2014 and as made available in the EU Digital Identity Wallets established under the same Regulation, as well as the proposed European Business Wallets, can provide secure and efficient means of identifying AI agents. The Wallets can store and manage electronic attestations that verify the AI agent's identity, attributes, and authorisations, thereby enabling seamless and trustworthy disclosure to natural persons.

- Visual/graphical cues: Design of the AI system with persistent icons, watermarks, coloured frames or recognisable “AI” symbols to complement textual disclosures. The use of standardised visual indicators, including across services of a provider, is recommended, since such indicators can significantly reduce cognitive burden and increase recognisability.

- Multi-modal combinations: Combining text, audio and visual cues (e.g. a chatbot displaying a label, while also providing a first-turn textual disclosure) to ensure accessibility and reinforce clarity across different user groups.

(38) Certain techniques are not necessary and, when used alone, insufficient for effectively fulfilling the transparency obligation in Article 50(1) and (5) AI Act, notably:

- Disclosures contained only in terms and conditions, URLs, or documentation (such disclosures may complement, though not replace, in-context disclosure);

- Machine-readable markings (e.g. metadata or watermarks) that are not perceivable by users at the point of interaction. This does not affect the appropriateness of such marks to comply with Article 50(2) AI Act (see Section 4);

- Unclear or ambiguous signals (e.g. generic references to “assistant”) or human-like representations that may mislead users;

- Generalised disclosures that are not sufficiently specific to the AI system’s outputs and interactions (e.g. on a platform providing a variety of services or individual inputs, a general disclosure like “Services on this website use AI” is insufficient);

- Technical or capability-based descriptions: statements solely referring to underlying technologies (e.g. “this system uses LLMs”) without explaining the function or implications of the system for the user and its artificial origin.

(39) Disclosures should be proportionate and calibrated to the different contexts and target audiences of the AI system to remain effective, user-friendly and avoid, to the extent, possible disruptions of the user experience or habituation effects (so-called “banner blindness”).

(40) A single, prominent notification before the first interaction of the AI system with a particular natural person is likely to suffice in most instances to meet the obligation in Article 50(1) AI Act. However, in certain riskier contexts, periodic reminders and context-aware disclosures are likely to be necessary to ensure continued user awareness. This is in particular the case in the context of AI systems interacting with vulnerable persons (e.g. children, the elderly, persons with disabilities and those with lower literacy), sustained and evolving interactions in sensitive or immersive contexts (e.g. where users may express or experience emotional distress or vulnerability or be at risk of addiction- like behaviours), where there is an increased risk of users being misled (e.g. financial advice, insurance, legal assistance, health advice, and complaints handling), or of forming emotional attachments or dependencies (e.g. AI companions), as well as when there is a need to ensure awareness and control over the actions of AI agents. Providers must also design and develop the AI system to ensure disclosure in all situations where the AI system is being asked questions relating to its nature or to the origin of the interaction, or where it can be reasonably assumed from the exchanges with the natural person that that person is likely to be misled or confused about the AI origin of the interaction.