Emotion recognition
Emotion-recognition systems infer a person's emotions from biometric data. The EU AI Act prohibits their use in the workplace and in education (Article 5); elsewhere they are generally high-risk under Annex III.
The Act defines an emotion-recognition system as one for the purpose of identifying or inferring emotions or intentions of natural persons on the basis of their biometric data. Article 5 prohibits placing on the market, putting into service or using such systems in the areas of workplace and education institutions, except where intended for medical or safety reasons — detecting fatigue in a pilot, for example. Where not prohibited, emotion recognition is listed in Annex III(1) as high-risk, and Article 50 adds a deployer duty to inform the people exposed to the system about its operation and to process their personal data in line with data-protection law. Physical states such as pain or fatigue are treated differently from emotions in the definition.
- Art. 3(39)
- Art. 5(1)(f)
- Art. 50(3)
- Annex III(1)
Emotion recognition: why it matters
Emotion recognition is the clearest case in the Act of the same technology carrying three different legal statuses depending on where it points: prohibited at work and in schools, high-risk in most other settings, and outside the definition entirely when it detects physical states rather than emotions. That makes the deployment context — not the model — the compliance question, and it makes vendor claims about "sentiment" or "engagement" features worth reading very carefully before a contract is signed.
Emotion recognition in practice
A contact-centre platform offers real-time agent-coaching that infers agent frustration from voice. Pointed at employees this is emotion inference in the workplace and prohibited. The same vendor's feature that infers *customer* sentiment from a call is not caught by the workplace ban, but it is high-risk biometric territory and triggers the Article 50 duty to inform the customers exposed to it. One product, two features, two entirely different outcomes.
Emotion recognition: the mistake people make
Assuming text-based sentiment analysis is caught. The definition turns on inferring emotions from *biometric* data — voice, face, physiological signals. Analysing the words in a written review is not emotion recognition under the Act, though it may raise other issues; conversely, relabelling a facial-analysis feature as "engagement scoring" changes nothing.
Emotion recognition: common questions
- Can we use emotion recognition on employees?
- No. Article 5 prohibits emotion-inference systems in the workplace, with a narrow exception for systems intended for medical or safety reasons such as fatigue detection in safety-critical roles.
- Is emotion recognition allowed on customers?
- It is not prohibited outside workplace and education contexts, but it is generally high-risk under Annex III(1) and the deployer must inform the people exposed to the system under Article 50(3), in addition to complying with data-protection law.
- Does text sentiment analysis count as emotion recognition?
- Not under the Act's definition, which requires inference of emotions or intentions on the basis of biometric data. Text analysis is outside that definition, though other obligations may still apply to the system it sits in.
Emotion recognition: terms that come up alongside it
Run the free 2-minute EU AI Act checker — no sign-up.