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

(42) To be able to rely on the first exception, providers will need to assess and demonstrate: i) the obvious artificial nature of the interaction to ii) a natural person who must be reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use.

(43) This standard draws on established EU consumer protection law regarding the notion of ‘average consumer’ and should be interpreted consistently, while taking into account the specific and varied contexts and uses of AI systems directly interacting with natural persons and the objectives pursued by Article 50(1) AI Act.

(44) Conceiving such ‘an average’ natural person comprises two steps. First, providers should consider the audience with whom the AI system is intended and reasonably likely to interact. Second, the provider should examine how reasonably well-informed, observant and circumspect an average member of that audience is, taking into account the circumstances and the context of use. Several factors should be considered when performing this assessment:

- Providers should consider the target audience of the AI system, as well as a broader potential audience that is reasonably foreseeable. This will especially be the case if the general public can easily access the AI system or if the system is operating in virtual or physical public spaces and capable of interacting with natural persons present in those spaces. The audience that may be exposed to an AI system does not always equals the target audience.

- Providers should consider the potential diverse composition of the target and reasonably foreseeable audience in line with Recital 132 AI Act. Where persons with disabilities, elderly people or minors are likely to be part of that audience, the expected levels of information, observance and circumspection of an average member of such an audience will be lower as compared to an average member of an audience that does not include those categories of persons. By contrast, if the interactive AI system is only available to a professional or specialised target audience, then this could positively impact the expected levels of knowledge, observance and circumspection and an average member of that audience may be considered to be better informed and more observant and circumspect.

- As part of the broader circumstances and context of use, providers should consider the level of digital and AI literacy of the target and reasonably foreseeable audience and relevant social, cultural, and linguistic factors.

(45) Obviousness: Upon establishing the hypothetical reasonably well-informed, observant and circumspect natural person, providers should evaluate whether it will be obvious to such a person that they are directly interacting with an AI system. The exception should be interpreted restrictively given that it deprives natural persons from the protection and the right to be informed with clear and distinguishable disclosures that they are interacting with AI systems. The general awareness of consumers and other natural

persons that AI systems (including chatbots and agents) exist does not imply that they recognise them in interactions. Several factors may be relevant and considered as part of the assessment whether the interaction is obvious:

- The anticipated nature of the interaction: In instances where the AI system interacts physically with natural persons, pertinent elements may include the visibility of mechanical components (which would increase obviousness) or the degree to which the AI system authentically replicates its non-artificial equivalent (which would decrease obviousness). For interactions that are visual or auditory in nature, elements such as writing or speech patterns, vocal tone (robot voice vs. genuine human- sounding voice), user interface design (e.g. a profile picture related to a chatbot that displays a human), and the capability for advanced personalised interaction may impact this assessment.

- The composition of the target and foreseeable audience: If the interactive AI system is only available to a professional or specialised audience, without potential exposure to vulnerable groups, this could increase the possible obviousness of the artificial nature of the interaction to an average member of the respective audience. By contrast, if the AI system may be accessed by the general public or consumers that include also vulnerable persons (e.g. such as children, the elderly, persons with cognitive, physical or mental disabilities, or persons with lower AI literacy) for whom the interaction is less obvious, the exception cannot be relied upon in order to protect those vulnerable users efficiently.

- Consideration should also be given to the fact that it is becoming increasingly hard for natural persons to know whether an interaction with an AI system or an actual human being is taking place. Due to the use of avatars and agents, the general anonymity of interactions online, and the capabilities of AI systems to write and interact in a human-like manner, the “obviousness” exception should be limited to cases where there is almost no doubt left about the nature of the interaction for an average person from the targeted and reasonably foreseeable audience. Examples of obviousness for which the transparency obligation under Article 50(1) AI Act does not apply:

- AI-powered code assistance and code review chatbots available only to professional developers who by virtue of their expertise and the context of use have no reasonable expectation of interacting with a human and can easily recognise that suggestions are generated by an AI system.

- An internal employee-facing assistant for properly trained staff who is AI literate and aware that they are using AI systems for internal organisational purposes (e.g. HR, legal, procurement, compliance or IT support).

- Interactive AI systems intended only to be used by properly trained health professionals to support medical diagnosis and suggest related treatments.

- Ambient-AI embedded in home appliances with functionalities constrained to the home appliance assistance and operation.

- Interactions with AI-enabled Non-Playable Characters (NPCs) in a single-player video game where the nature of the video game makes it clear that no other natural person can participate

and interact with the user and the AI nature of the interaction is also obvious for all users (including vulnerable persons). Examples that do not fulfil the obviousness-exception and for which the transparency obligation under Article 50(1) AI Act applies:

- An AI-powered robotic companion pet, looking highly similar to its natural equivalent and designed to mimic typical human-pet interaction, creating ambiguity as to whether the interaction is with an AI system or not.

- AI systems embedded in immersive environments (e.g. virtual or augmented reality) using realistic avatars or voices resembling humans, where users and particularly children, the elderly, or persons with disabilities may not be able to distinguish easily between human and AI interaction.

- AI chatbots embedded in online platforms or assistance support tools (helpdesks) whereby users directly interact and receive AI outputs (e.g. replies to queries or other AI-generated content) they may perceive as human-generated.