Transparency obligations for chatbots, deepfakes, and emotion recognition (Article 50)
In short
Article 50 is a transparency floor that applies even to systems that are otherwise low-risk , it stacks on top of any risk tier a system already sits in, rather than replacing the high-risk obligations for systems that also qualify as high-risk. It covers three distinct disclosure duties.
Disclosing that you're talking to an AI (Art. 50(1))
Providers of systems intended to interact directly with people must design them so those people are informed they are interacting with an AI system, unless this would be obvious to a reasonably well-informed person from the circumstances. The disclosure has to happen at the latest at the time of first interaction, not buried in terms of service discovered later. A narrow exception exists for authorised law-enforcement use in crime detection, prevention, and investigation, but it does not apply where the system is publicly available for reporting a crime.
Labelling AI-generated content (Art. 50(2))
Providers of systems that generate synthetic audio, image, video, or text content must mark the output, in a machine-readable format, as artificially generated or manipulated, as effectively as is technically feasible, and before the content is distributed or made public, not as an afterthought if challenged. Exceptions apply for purely assistive editing functions, content that does not substantially alter the original, and authorised law-enforcement use, each of which is treated differently under Art. 50(2).
Informing people about emotion recognition and biometric categorisation (Art. 50(3))
Deployers of emotion recognition or biometric categorisation systems must inform the people exposed to the system that it is in operation, and must process any personal data involved in accordance with GDPR, the Act treats this as an explicit requirement alongside the disclosure duty, not a separate, optional concern. A narrow law-enforcement exception exists here too, but it requires appropriate safeguards, not just a law-enforcement label.
How this interacts with a high-risk classification
A system can trigger both Article 50 and the full high-risk obligation set at once, a recruitment chatbot, for example, may need to disclose it is an AI system under Art. 50(1) while separately carrying the Article 9-15 and Article 26 obligations because it also qualifies as high-risk under Annex III. Meeting the Article 50 disclosure does not satisfy, or reduce, any high-risk duties the same system independently carries.
Because “obvious from the circumstances” and “substantially alter the original” are judgment calls rather than bright lines, borderline cases (a highly realistic voice assistant, a lightly-edited photo) are worth confirming rather than assuming the disclosure duty does not apply.
Frequently asked questions
- Does a chatbot have to tell users it is an AI?
- Yes. Under Article 50(1), providers of systems that interact directly with people must design them so people are informed they are interacting with an AI, at the latest at first interaction, unless that would be obvious to a reasonably well-informed person. A narrow law-enforcement exception applies.
- Does AI-generated content have to be labelled?
- Yes. Under Article 50(2), providers of systems generating synthetic audio, image, video, or text must mark the output as artificially generated or manipulated in a machine-readable format, as effectively as technically feasible, before it is distributed, with exceptions for purely assistive editing, non-substantial alterations, and authorised law enforcement.
- Do transparency duties replace high-risk obligations?
- No. Article 50 stacks on top of any risk tier. A recruitment chatbot may need to disclose it is an AI under Article 50(1) while also carrying the full Articles 9–15 and Article 26 obligations because it qualifies as high-risk under Annex III, meeting the Article 50 disclosure does not reduce those duties.
Related guides
Not sure where your company stands?
Our free assessment gives you an indicative result in minutes (free and anonymous) no account needed.
Start the free check