Article

Emotion recognition at work is banned under the EU AI Act: what that covers

Some AI features in hiring and workforce tools claim to read engagement, stress or confidence from a face or a voice. In the workplace, the EU AI Act puts that use among its prohibited practices, and the prohibition has applied since 2 February 2025.

What is prohibited

Article 5(1)(f) prohibits placing on the market, putting into service for this purpose, or using AI systems to infer the emotions of a natural person in the workplace or in education institutions, except where the system is intended for medical or safety reasons. The verified wording.

An emotion recognition system, in the regulation's definition, identifies or infers emotions or intentions of people on the basis of their biometric data. That is the part that decides most cases: the inference is drawn from the face, the voice or other biometric signals.

Where common features fall
Feature Why it matters
Call centre analytics that flag an agent's stress from their voice Infers an employee's emotions from biometric data at work: inside the prohibition
Video interview scoring that rates enthusiasm or confidence from facial expressions Infers emotions from biometric data; see the warning below on candidates
Fatigue detection for a driver or machine operator May fall under the safety exception, depending on its intended purpose
Sentiment analysis of what a customer wrote in an email Drawn from text, not biometric data, and not about the worker

The neighbouring prohibitions

The same article also prohibits biometric categorisation that deduces race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation, and social scoring that leads to detrimental or disproportionate treatment. Prohibited practices.

What it costs

Prohibited practices carry the highest fines in the regulation: up to EUR 35 000 000 or 7% of worldwide annual turnover, whichever is higher, and for SMEs and start-ups whichever is lower. The fines in one table.

What to do this week

Ask each supplier of an interview, monitoring or contact centre tool one written question: does any feature infer emotions, intentions or states such as stress or engagement from face, voice or other biometric data, and can it be switched off? Keep the answer. Then switch it off, or record the medical or safety purpose that justifies it.

Where this fits

The AI Act at Work includes the supplier letter that asks exactly this kind of question and grows into a vendor questionnaire, and the register where the answer and your decision are kept.