AI Act Article 50: transparency obligations for chatbots and generated content
·
Updated on ·
By 
Aymeric Guyard · 6 min read
In brief
- Article 50 imposes transparency obligations for limited-risk systems: informing the user that they are interacting with an AI, labelling generated content (text, image, sound, video), and identifying deepfakes. These obligations affect a very large number of companies and are already applicable, independently of the postponement of high risk to 2027.
The four obligations under Article 50
- Interaction with an AI: when a system interacts with a person (chatbot, voice assistant), they must be informed that they are communicating with a machine, unless this is obvious.
- Generated or manipulated content: content produced by AI (text, image, audio, video) must be marked in a machine-readable way as artificially generated.
- Deepfakes: content that imitates real persons, objects or events must be clearly identified as artificial.
- Public information texts: texts generated by AI published to inform on matters of public interest must be disclosed, except where subject to human editorial oversight.
How to apply it in practice
On a chatbot: display a clear notice from the outset — 'You are communicating with a virtual assistant.' The information must be visible, not hidden in the terms and conditions. If the assistant can transfer to a human, indicate this.
On generated content: two levels — a visible notice ('Image generated by AI') and, where technically feasible, machine-readable marking (watermarking, metadata). Technical marking standards are taking shape: follow their evolution.
On deepfakes: any synthetic reproduction of a real person must be disclosed as artificial. The issue goes beyond compliance: reputation and liability in case of deceptive use.
Exceptions
Certain exceptions exist: where interaction with the AI is obvious, or where generated content is subject to human editorial control with editorial responsibility. But these exceptions are to be interpreted strictly: when in doubt, it is better to inform. Best practice: integrate transparency from the design stage (transparency by design) rather than adding it afterwards. A native notice is better received than a tacked-on warning.
Take action
Bring your chatbots and AI content into compliance with Article 50: our experts support you with Data Comply One.
Discover the AI Act software →Test my compliance for free