Emotion analysis in interviews and surveillance: what has been banned since February 2025

What Article 5 says

Among the practices prohibited since 2 February 2025 is emotion recognition in the workplace and education. The target is AI systems designed to infer a person's emotions in the workplace or in an educational establishment. Two strict exceptions: uses for medical or safety purposes (for example, detecting an operator's drowsiness to prevent an accident).

HR uses concerned:

• Analysis of a candidate's emotions or 'stress' during a video interview.
• Assessment of 'sincerity' or 'motivation' by reading micro-expressions.
• Monitoring of employees' emotional state (frustration, engagement) at the workplace.
• Measurement of attention or mood during training.

Be careful not to confuse: analysing the content of a candidate's answers (what they say) is not emotion recognition (how they say it, their supposed emotional state). It is this second use that is prohibited. See the 4 risk levels of the AI Act.

What you must do

• Audit your HR tools: does a module analyse emotions, affect or psychological state from voice or face?
• If yes, cease this use: the prohibition is immediate and sanctionable at the maximum level (up to €35m or 7% of global turnover).
• Require from your vendors written confirmation of the absence of emotion recognition.
• Document your verification, as proof of due diligence.

Note: beyond the AI Act, these uses raise fundamental problems with regard to GDPR (biometric data, legal basis, proportionality) and the principle of non-discrimination. The AI Act prohibition simply closes a door that was already widely challenged. See also AI and recruitment.