“Hello, I’m Maria from the customer service team. How can I help you?”
But… is Maria really a person?
Customer service chatbots, virtual assistants, product images, marketing campaigns and AI-generated texts are already part of many companies’ day-to-day operations. Yet users cannot always tell when they are interacting with AI or when content has been artificially generated.
This is precisely why Article 50 of the AI Act, applicable since 2 August 2026, lays down different transparency, information and marking obligations for certain AI systems. The European Commission has published guidelines on how these obligations apply, clarifying who must provide information, when they must do so and, above all, how to comply in practice.
The answer depends on the specific use of AI:
Chatbots and virtual assistants: duty to inform
When an AI system interacts directly with natural persons, the provider must ensure that they know they are interacting with AI. In practice, this may take the form of an initial message, banner, label or oral notice such as: “You are interacting with an AI assistant”.
The obligation does not apply where it is obvious to a reasonably well-informed person that they are interacting with an AI system.
AI-generated or manipulated content: machine-readable marking
Providers of systems that generate or manipulate images, video, audio or text must ensure that their outputs are marked in a machine-readable format and are detectable as artificially generated or manipulated.
The solution may involve metadata, watermarks, fingerprinting or other machine-readable marking techniques, together with mechanisms that allow the marking to be detected.
Emotion recognition and biometric categorisation systems: prior information
Here, the obligation falls on the person using or deploying the system. Individuals exposed to it must be informed about its operation before or at the time they become subject to it, for example through a sign, pop-up or on-screen notice.
Deepfakes and public-interest texts: visible labelling
Where deepfakes are generated or manipulated, or AI-generated texts are published to inform the public about matters of public interest, it must be visibly disclosed that the content has been artificially generated or manipulated.
For texts, there is a particularly relevant exception for companies and media organisations: labelling is not required where the content has undergone genuine human review or editorial control and a person assumes editorial responsibility for it. Importantly, a simple spelling correction or superficial approval is not sufficient.
What Article 50 of the AI Act means for businesses
Businesses must consider not only whether they may use AI, but also how they must disclose that they are using it.
Depending on the system and its purpose, the appropriate measure may be a disclaimer, a visible label, metadata or technical marking embedded in the content itself.
If your company uses AI systems and needs advice on which transparency obligations apply or how to implement them, contact the ACROSS LEGAL team.
Written by: Inés Arocas
↳ Discover more content in the section Insight.
↳ Do you want to stay up today about the sector related news? Follow us on LinkedIn.
| Article | Type of AI system/output | Transparency obligation | Exception or special regime | Practical implementation | Definitions/key issues |
| Art. 50(1) | AI system that interacts directly with natural persons. | Providers must develop and design the AI system in such a way that the natural persons concerned are informed that they are interacting with an AI system. | Exceptions: (1) where the artificial nature of the interaction is obvious; or (2) where the system is authorised by law to detect, prevent, investigate or prosecute criminal offences, unless it is available to the public for reporting a criminal offence. | Visible notice/disclaimer: disclose at the start of the interaction that the user is interacting with AI. Example: “You are interacting with an AI assistant”. In chatbots, this may be shown through an initial message, banner or label; voice assistants may use an oral notice. | Direct interaction: a two-way exchange in real time or near real time. Reasonably well-informed, observant and circumspect natural person: refers to the concept of the ‘average consumer’. |
| Art. 50(2) | AI systems that generate or manipulate synthetic image, video, audio or text content. | Providers must ensure that the AI system’s outputs are marked in a machine-readable format and are detectable as artificially generated or manipulated, using effective, interoperable, robust and reliable technical solutions. | Exceptions where: (1) the AI system performs an assistive function for standard editing or does not substantially alter the input data or its semantics; or (2) the AI system is authorised by law to detect, prevent, investigate or prosecute criminal offences. | Technical content marking: incorporate a machine-readable mark, for example through metadata, a watermark, fingerprinting or cryptographic techniques, together with a mechanism that allows the mark to be detected. | Synthetic content: content perceptible by natural persons, in the form of text, images, audio or video, generated or manipulated by an AI system. It also includes multimodal content, 3D content and virtual, augmented or mixed-reality content. |
| Art. 50(3) | AI systems for emotion recognition or biometric categorisation. | Deployers must inform the natural persons exposed to the system about the operation of the AI system. | Exception where the AI system is permitted by law to detect, prevent or investigate criminal offences. | Visible notice/disclaimer: provide information before or at the time the person becomes exposed to the system. Examples include a sign at the entrance to a space, an in-app pop-up or an on-screen notice. | This obligation applies regardless of whether the system falls within the scope of high-risk AI systems. Biometric identification: automated recognition of physical, physiological or behavioural human features, such as the face, eye movement, body shape, voice, intonation, gait, posture, heart rate, blood pressure, odour or keystroke characteristics, for the purpose of establishing a person’s identity. |
| Art. 50(4) | AI systems that generate or manipulate deepfakes or texts published to inform the public about matters of public interest. | Deployers must disclose that the content has been artificially generated or manipulated. | Exceptions where: (1) the AI system is authorised by law to detect, prevent, investigate or prosecute criminal offences; or (2) the published text has undergone human review or editorial control and is subject to editorial responsibility. A specific transparency regime applies to deepfakes forming part of artistic, creative, fictional, satirical or analogous works or programmes. | Labelling/disclaimer alongside the content: visibly indicate that the content has been generated or manipulated using AI. EU icons for labelling AI-generated content | Human review: deliberate examination of content by one or more persons with the appropriate competence and professional judgement. Human review does not include grammatical or spelling correction, superficial approval without substantive review, or the mere existence of an editorial policy. Deepfake: AI-generated or manipulated image, audio or video content resembling real persons, objects, places, entities or events that could falsely appear authentic or truthful to a person. |
