From 2 August 2026, anyone who deploys a system for emotion recognition or biometric categorisation must inform the natural persons exposed to it. This follows from Article 50(3) of the AI Act. But transparency is not the first question you should ask here. Many uses of emotion recognition and biometric categorisation are in fact already prohibited under Article 5, which has been in force and enforceable since 2 February 2025. If your use is prohibited, a tidy disclosure does not make it permitted after all. So test the prohibition first, and only then turn to the disclosure duty.
This page accompanies our practical overview of Article 50. Below you will find what Article 50(3) requires, who carries the duty and why the prohibition under Article 5 takes precedence.
What exactly does Article 50(3) require?
Article 50(3) requires deployers of a system for emotion recognition or biometric categorisation to inform the natural persons exposed to it about the operation of the system. In addition, they must process the personal data in accordance with the GDPR and other applicable law.
The disclosure must be given at first exposure, and must be clear and accessible. So it is not a formality somewhere in a privacy statement, but information that actually reaches the person concerned at the moment the system is applied to them.
Who must inform?
The duty sits with the deployer: the organisation that uses the system under its own responsibility. Not the person being scanned, and under this provision not the provider of the system either.
| Question | Answer |
|---|---|
| Who carries the duty? | The deployer who uses the system |
| For which systems? | Emotion recognition and biometric categorisation |
| When to inform? | At first exposure, clearly and accessibly |
| From when? | 2 August 2026 |
Note the link with the GDPR: biometric data are special categories of personal data. The transparency duty under the AI Act therefore comes on top of the requirements the GDPR already sets for processing that data.
Why the prohibition under Article 5 takes precedence
This is the point organisations most often miss. Emotion recognition in the workplace and in education is a prohibited practice under Article 5, except for medical or safety reasons. That prohibition has been in force and enforceable since 2 February 2025. An employer that infers the mood of employees from facial expressions or voice, or a school that does the same with pupils, falls under that prohibition. Transparency changes nothing here: informing does not make a prohibited practice permitted.
A prohibition also applies to certain categories of biometric categorisation. Inferring race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation is prohibited under Article 5. Other forms of biometric categorisation may be high-risk under Annex III; that is a separate dimension alongside the transparency duty, not a replacement for it.
Always test Article 5 first. If your use falls under the prohibition, the disclosure duty under Article 50(3) is no longer relevant: the use is simply not allowed. Breaching the prohibition also carries a higher level of fines than a transparency breach.
Which exception applies to the disclosure duty?
There is an exception to Article 50(3) for systems permitted by law to detect, prevent or investigate criminal offences, with the corresponding safeguards. Outside that domain, the disclosure duty applies in full.
This exception is narrow and tied to a legal basis. Anyone relying on it must be able to show that the use is permitted by law and that the safeguards are observed.
The assumptions behind emotion recognition
There is a substantive reason to be strict here. According to a report by the Dutch Data Protection Authority, emotion recognition rests on contested assumptions, carries high error rates and brings discrimination risk. The objective presentation of subjective and unreliable data is misleading: a system that presents an emotion as fact creates an appearance of certainty that the technology does not deliver.
That is precisely why the legislator prohibited these uses in the workplace and in education, and why transparency elsewhere is not enough to remove the underlying risk.
What should you do now?
Map where it is used
Map where in your organisation emotion recognition or biometric categorisation is used or being considered. Think of HR, customer contact, security and educational settings.
Test Article 5 first
Build disclosure and follow the GDPR
For the uses that remain: build the disclosure in at first exposure, clearly and accessibly, and bring the processing in line with the GDPR. Keep evidence of the disclosure and the legal basis.
Organisations that keep to this order avoid the classic mistake: building a tidy disclosure for a use that is not allowed at all.
Frequently asked questions about emotion recognition and biometric categorisation
Practical questions about the disclosure duty under Article 50(3) and the prohibition under Article 5 of the AI Act.