Monitoring employees with AI: what is allowed and what isn't?

Short answer: AI for monitoring employees is not allowed without limits. One form is banned entirely: AI that infers emotions in the workplace (Art. 5 AI Act, since 2 February 2025). Other forms — performance or behaviour monitoring — can be high-risk (Annex III, workforce management) and in any case fall under the GDPR, which requires a valid legal basis, transparency and proportionality. Continuous, intrusive monitoring is legally vulnerable.

What is banned

AI systems that infer employees' emotional state — through camera, voice or behaviour analysis — are banned in the workplace. It does not matter how "helpful" it is meant to be; only a narrow exception applies for medical or safety purposes. See prohibited AI practices.

When monitoring is high-risk

AI that monitors and evaluates workers' performance or behaviour falls under workforce management in Annex III and can therefore be high-risk. Then human oversight (Art. 14), informing the workers concerned (Art. 26) and data-quality requirements apply, among others. The more the system weighs in decisions about people, the heavier the regime.

The GDPR sets the limits

Even without a high-risk qualification, the GDPR applies. Monitoring requires a valid legal basis (employee consent is usually not freely given, so weak), transparency in advance, data minimisation and a proportionality test: does the purpose outweigh the privacy intrusion? A Data Protection Impact Assessment (DPIA) is often mandatory for systematic monitoring. In many countries the works council also has consultation or co-determination rights.

What to do

• Start with whether it is banned (emotion recognition) — if so, don't.
• Classify the monitoring: high-risk or not (see high-risk obligations).
• Test proportionality and record a DPIA.
• Inform employees transparently and involve worker representatives.
• Limit what you collect and how long you keep it.

Monitoring is where employers most quickly go wrong: the technology can do a lot, but the law draws sharp lines. Start with the purpose and proportionality, not with what is technically possible.