Article 5(1)(f) prohibits the placing on the market, the putting into service or the use of AI systems to infer emotions of a natural person in the areas of workplace and education institutions, except where the use is intended for medical or safety reasons. Emotion inference means inferring emotions from biometric data, and the prohibition is tied to the setting and to biometric inference, not to every assessment of mood or sentiment. The medical or safety exception is narrow and must be tested on the intended reason, the setting, and the safeguards, not on a general claim that the system improves wellbeing or safety.
What is emotion inference in this article?
Emotion inference in the Act is tied to biometric data. It means inferring emotions or intentions from biometric data, such as facial expressions, voice characteristics, or other physiological signals, to conclude how a person feels. A survey that asks a person how they feel, a text analysis that assesses sentiment in written words, or a manager's own observation does not meet that description merely because it concerns mood. The record should describe the biometric input, the inference performed, and the emotion or intention output, because the prohibition is about the biometric inference step. A system that measures fatigue, attention or stress from biometric signals needs the same analysis, and should not be relabelled as wellness or engagement monitoring to avoid it.
Which settings does point (f) cover?
Point (f) covers the workplace and education institutions. The workplace includes employment settings where a worker's emotions would be inferred by or for the employer, including remote and hybrid work where biometric signals are captured through a device. Education institutions include schools, colleges, universities and other institutions where learners' emotions would be inferred in an educational setting. The setting is assessed by where and for whom the inference is performed, not by the supplier's sector label. A system sold as a general tool that is deployed in a workplace or school to infer emotions is within the setting description, even if the same tool could be used elsewhere.
For the broader regulatory framework, see prohibited practices under Article 5.
How narrow is the medical or safety exception?
The exception covers use intended for medical or safety reasons. Medical reasons include clinical purposes where emotion inference from biometric data supports diagnosis, treatment or patient safety under medical supervision. Safety reasons include detecting a condition that creates an immediate safety risk, such as a driver or operator whose state creates danger to themselves or others. The exception is tied to intention and to reason: the use must be intended for that medical or safety purpose, and the design, deployment and safeguards should reflect that purpose. A general wellbeing, productivity or engagement purpose does not become medical or safety merely because it is described as beneficial. The record should state the specific medical or safety reason, the professional or safety context, and the limits placed on the inference.
What about emotion inference outside work and education?
Outside the workplace and education settings, point (f) does not prohibit emotion inference in the same way, but other provisions can apply. Annex III includes biometric identification and categorisation uses that can be high-risk, Article 50 transparency duties can apply to emotion recognition and biometric categorisation systems, and data protection law governs biometric data with strict conditions. A system used in retail, entertainment or customer service to infer emotions from biometric data should therefore still be assessed, even though point (f) may not be the provision that governs it. The setting analysis should be done first, and the other provisions checked second, so that a system is not cleared on the setting alone.
What should employers and schools record?
An employer or education institution considering any biometric emotion tool should record the setting, the biometric data collected, the emotions inferred, the purpose, the medical or safety reason relied on if any, the consent and consultation position under employment and education law, and the data protection assessment. Worker and learner representatives, where the applicable law requires information or consultation, should be involved before deployment. Where the exception is relied on, the record should show the medical or safety professional involved, the risk being addressed, and the safeguards that limit the inference to that purpose. A tool adopted for productivity or classroom management without a medical or safety reason should not proceed on the basis of point (f).
Continue with biometric categorisation prohibition.
What does this guide not conclude?
This guide explains the scope of point (f) and its exception in outline. It does not decide whether any particular tool infers emotions from biometric data, whether a setting is a workplace or education institution, or whether a medical or safety reason is made out. Those questions need the biometric input, the inference, the setting, and the purpose evidence, read against the consolidated text and the applicable employment, education and data protection law. Tools that infer emotions of workers or learners should be confirmed with a competent authority or legal adviser before deployment, because the affected people are often in a position of dependence on the institution deploying the system.
Frequently asked questions
Is sentiment analysis of text prohibited by point (f)?
Not by point (f) as described, because the prohibition concerns inferring emotions from biometric data in workplace and education settings. Text sentiment analysis can still engage data protection law, employment law and other AI Act provisions, which must be checked separately.
Can fatigue detection for drivers rely on the safety exception?
It can be considered under the safety exception where the use is genuinely intended for safety reasons, such as preventing danger from a fatigued operator, and the design and safeguards reflect that purpose. The exception must be evidenced, not assumed from the label fatigue detection.
Does point (f) cover students at home using school software?
The setting analysis looks at the education institution context and the purpose of the inference, not only at the physical location of the device. A school-deployed tool that infers a learner's emotions from biometric data in an educational setting needs the point (f) analysis even where the learner is at home.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689 ↗AI Act Service Desk timeline ↗European Commission regulatory framework for AI ↗