Emotion Inference in Workplaces and Educational Settings

The sixth prohibited practice covers AI systems that infer emotions or emotional states from video, audio or biometric data and then use those inferences to make decisions about people in workplaces and educational settings.

Emotion recognition or emotion inference AI claims to identify a person’s emotional state from facial expressions, voice tone, body language, posture, or heart rate data. Some of these systems use video analysis to measure facial movements and micro-expressions; others use physiological sensors to measure heart rate variability or skin conductance. The premise behind these systems is that emotions correlate reliably with performance, engagement, attention or suitability for a specific role.

The EU prohibits this practice in two specific contexts: the workplace and education. The concerns are substantial. Emotion inference is scientifically unreliable even in controlled laboratory settings. It undermines human dignity by treating emotional state as a measurable technical parameter to be harvested and judged. It creates risks of discrimination and manipulation. It assumes that emotional states are observable and meaningful when in fact they vary between individuals and cultures.

The science of emotion recognition remains contested and unsettled. There is no universal code for facial expressions that holds across cultures. The same facial expression can mean different things in different cultural contexts. A person’s facial expression in a video call does not reliably indicate their emotional state, cognitive capability, or commitment to work. Facial expressions can be controlled or suppressed. The inference is often based on culturally specific assumptions that systematically misclassify people who do not fit Western European stereotypes. People with autism, Parkinson’s disease, Bell’s palsy and other conditions may show different facial expressions than the model expects, leading to incorrect inferences.

Using emotion inference to make decisions about hiring, performance management, promotion or discipline is prohibited. An employer cannot use an AI system that infers a job candidate’s emotional stability from their interview video and relies on that inference to make hiring decisions. An employer cannot use an AI system that monitors employee emotions during work through CCTV facial analysis and flags those with low detected engagement for management attention. A school cannot use an AI system that analyses students’ facial expressions during lessons to determine who is learning effectively or who lacks attention.

There is a narrow exception for emotion inference systems used in security and law enforcement contexts. Law enforcement can use emotion detection for specific criminal investigations if they have legal authority and the detection is targeted to specific individuals suspected of offences. This exception does not extend to workplace or education use.

The exception for education is limited to security screening at school entrances. A school can use emotion or biometric systems to screen for prohibited items, weapons or other security threats at building entrances. This is a targeted security exception, not a performance or engagement exception.

For most organisations, the rule is clear: do not use emotion inference in employment decisions, performance management, promotion decisions or student assessment. Do not use it in any decision that affects access to work or education. The technology is not scientifically reliable enough and the harm from errors is too large. Emotions are private aspects of human experience and should not become data to be mined and used against a person.

Examples of prohibited emotion inference in employment include: an HR system that monitors CCTV video during work and flags employees who show low facial engagement during working hours, triggering disciplinary discussions; a recruiting company that uses AI to analyse candidate interview videos for emotion and engagement signals, then uses these signals as part of hiring decisions; a student assessment system that monitors classroom CCTV and flags students who appear disengaged or distracted, using that data in academic evaluation; a call centre monitoring system that infers customer service representative engagement from facial expressions in video calls and uses that to make performance ratings or disciplinary decisions; an online education platform that monitors student video during lessons and infers attention levels to automatically adjust content difficulty or grade performance.

In contrast, an employer can monitor actual productivity (time spent on tasks, output produced), can conduct personal interviews and make hiring decisions based on interaction quality, or can assess student learning through assignment completion and exam performance. The line is clear: monitor behaviour and output, not emotion.