Emotion recognition at work: the ban that already applies

Short answer: Since 2 February 2025 it is prohibited to use AI to infer the emotions of workers or students — in the workplace and in education. That hits camera analysis of facial expressions, voice analysis for stress or engagement, and "sentiment" scoring during work or job interviews. There is one narrow exception: medical or safety purposes. The ban sits in Article 5, the heaviest category of the AI Act.

What exactly is prohibited

The ban covers AI systems that infer emotions or intentions of a person in the context of work or education. Concrete examples that fall under it:

• facial analysis during a (video) job interview to score "enthusiasm", "honesty" or "fit";
• voice or tone analysis in a call centre to gauge staff mood;
• cameras estimating engagement or fatigue on the work floor or in the classroom.

It is about inferring an inner state. Merely detecting a physical expression without attaching an emotion to it strictly falls outside — but the line is thin.

The narrow exception

Emotion recognition is permitted only for medical or safety purposes — for example, a system detecting a bus driver's drowsiness to prevent accidents. The exception is deliberately tight: wellbeing monitoring that in fact measures performance or productivity falls under the ban, per the Commission guidelines.

Why this is sharp for HR

Many modern recruitment and monitoring tools promise exactly what is banned: "soft skills from video", "engagement scores", "stress detection". An employer deploying such a feature breaches the AI Act's hardest provision — with the highest fine band (up to EUR 35 million or 7% of worldwide turnover). And unlike high-risk, no classification has to be established: the ban is directly enforceable.

What to do

• Screen your recruitment and monitoring tools for emotion, sentiment or "soft skill" analysis from image or sound.
• Switch those features off or replace the tool — even if the supplier offers them as optional.
• Watch the proxy: "engagement" or "wellbeing" scores that in fact infer emotions are caught too.
• In doubt? Treat it as prohibited until proven otherwise; the 4 February 2025 guidelines are the test.

Biometrics and emotion AI are the most tempting and the most risky HR applications. There is no "if done carefully" here — the rule is: don't.