Biometric categorisation and emotion inference must be assessed function by function. Article 5 prohibits, in outline, biometric categorisation systems that infer sensitive attributes such as race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation, subject to defined exceptions for lawful labelling or filtering, and it prohibits emotion inference in the workplace and in education, except for medical or safety reasons. Annex III point 1 lists biometric categorisation and emotion recognition systems as high-risk uses. Article 50 also requires deployers to inform people exposed to emotion recognition or biometric categorisation systems. A system that infers an emotion or a category from text, voice or behaviour must be tested against the biometric definition, not assumed to be outside it.
What is biometric categorisation?
Biometric categorisation is the assignment of a person to a category on the basis of biometric data, such as facial, voice or other physiological or behavioural characteristics. The category can be a sensitive attribute, such as race or ethnic origin, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation, or it can be a non-sensitive category, such as age range or a demographic group used for a lawful purpose. The distinction matters because Article 5 addresses categorisation that infers sensitive attributes, while Annex III addresses categorisation more broadly, and the legal treatment differs between the two. A team should therefore record what data are used, what category is inferred, how the inference is made and what decision follows from it. A system described only as analysing faces or voices cannot be assessed, since the same input can support categorisation, identification, emotion inference or verification, each with a different legal treatment.
What does Article 5 prohibit for categorisation?
Article 5 prohibits, in outline, the placing on the market, putting into service or use of biometric categorisation systems that categorise natural persons based on their biometric data to infer sensitive attributes, including race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation. The prohibition is subject to defined exceptions, including labelling or filtering of lawfully acquired biometric datasets and categorisation of biometric data in law enforcement, which must be checked in the consolidated text. A team should test the system against the prohibition's elements: whether biometric data are used, whether a sensitive attribute is inferred, and whether an exception applies on the facts. A system that categorises by a non-sensitive attribute is not within that prohibition, but it can still be high-risk under Annex III and can still carry transparency and data protection duties. The prohibition screen and the classification must therefore be recorded separately, with the reasoning for each.
For the broader regulatory framework, see healthcare and biometrics under the AI Act.
What does Article 5 say about emotion inference?
Article 5 prohibits, in outline, the placing on the market, putting into service or use of AI systems to infer the emotions of a natural person in the areas of workplace and education institutions, except where the system is intended to be used for medical or safety reasons. Emotion inference means inferring emotions or intentions from biometric data, such as facial expressions or voice patterns, and the definition should be checked in the consolidated text, since it turns on biometric data rather than on any assessment of sentiment. A system that analyses the sentiment of written text, without biometric data, is assessed differently from a system that infers emotion from a face or voice, and the two should not be conflated. A workplace or education deployer should therefore record the input data, the inference made and the purpose, and should test the medical or safety exception on its own facts, with qualified advice, rather than assuming that a wellbeing or safety label brings the system within the exception.
How does Annex III classify categorisation and emotion recognition?
Annex III point 1 lists biometric categorisation systems and emotion recognition systems as high-risk use cases, subject to the Article 6 conditions and the profiling limit. A system that falls within the listing is high-risk under Article 6(2), and its provider and deployer duties follow the Chapter III framework from 2 December 2027, while prohibition screening under Article 5 and transparency duties under Article 50 apply on their own dates. The classification must consider the function actually performed: a system that categorises people by age for a lawful purpose, a system that infers emotion for a medical purpose and a system that infers emotion for performance management are different cases, even where the technology is similar. The record should cite the Annex point, the function, the purpose and the Article 6 reasoning, and should be revisited if the function, the population or the decision that follows from the inference changes.
What transparency duties apply to people exposed to these systems?
Article 50 requires deployers of emotion recognition systems and biometric categorisation systems to inform the natural persons exposed to the system that they are exposed to such a system, subject to the exceptions in the provision, including where Union or national law authorises the use for detecting, preventing or investigating criminal offences. The information must be provided in a clear and accessible manner, at the appropriate time, and must allow the person to understand that they are exposed and, where relevant, to exercise their rights. That duty is distinct from the prohibition and classification questions: a system can be lawful and high-risk, and still require a notice to the people exposed to it. A deployer should therefore plan the notice as part of deployment, including its wording, its timing, its accessibility and the route for questions or challenges, and should record that the notice was given. A notice that is hidden, generic or given after the exposure has occurred does not satisfy the provision.
Continue with biometric identification systems.
What does this guide not decide?
This guide does not decide whether any categorisation or emotion inference system is prohibited, high-risk or subject to a transparency duty, and it does not determine whether an exception applies on given facts. Those questions depend on the input data, the inference, the category or emotion concerned, the setting and the purpose, and they require the consolidated definitions in Article 3, the text of Articles 5 and 50 and Annex III to be applied to the specific system. The guide also does not assess the accuracy or validity of emotion inference, which is a scientific and technical question that affects the risk assessment. Where a categorisation or emotion inference system is to be used, confirm the prohibition, classification, transparency and data protection position with a competent authority or legal adviser before use.
Frequently asked questions
Is emotion inference from text covered by the Article 5 prohibition?
The prohibition concerns inferring emotions from biometric data in the workplace and education. Sentiment analysis of written text without biometric data is assessed differently and must be tested against the definitions and provisions that actually apply to it.
Does the medical exception cover any health-related use?
No. The exception for medical or safety reasons must be tested on the facts of the system and its intended purpose, with qualified advice. A wellbeing or safety label alone does not bring a system within the exception.
Must people be told they are exposed to categorisation or emotion recognition?
Yes, as a general rule. Deployers must inform exposed persons under Article 50, subject to the exceptions in the provision, including the criminal-offence exception. The notice must be clear, accessible and timely.
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, as amended ↗AI Act Service Desk implementation timeline ↗