Your AI may not read your staff's emotions

Using AI to read the emotions of your own staff is banned under the EU AI Act. That covers contact-centre agents and warehouse teams too.

Article 5(1)(f) bans AI that infers how people feel at work. The ban has applied since 2 February 2025. There is one narrow exception: medical or safety reasons.

The key test is biometric data. Article 3(39) defines emotion recognition as inferring emotions from biometric data, such as voice, face or body movement. The Commission's guidelines say that reading emotion from written text alone is not covered. Spotting a physical state, such as fatigue or pain, is not emotion recognition either (Recital 18).

Customers are treated differently. Reading shoppers' emotions is not banned. But the deployer must tell people the system is running (Article 50(3)). Such systems are also high-risk under Annex III point 1(c).

What to do now:

1. Check your contact-centre and workforce tools. Look for voice analytics that score agent "sentiment" or "stress", or webcam tools that measure engagement. 2. One tool can do both jobs. Scoring the customer's tone of voice is allowed with notice. Scoring the agent's tone is banned. Switch off the agent-facing feature, or confirm it works on text only. 3. Where you analyse customer emotions, add a clear notice. Plan for high-risk duties as well.

Get this wrong and the top fine tier applies: up to €35 million or 7% of worldwide annual turnover (Article 99(3)).