Emotion-reading robots and the AI Act: where the ban applies and where it does not

No, not in general. The EU AI Act bans AI that infers people's emotions only in workplaces and education institutions, with exceptions for medical or safety reasons. A robot reading the moods of visitors at a trade fair is not covered by the ban. It is still regulated, though: it must tell people what it is doing, will count as high-risk, and must not analyse the staff working around it.

This article explains the scope of the prohibition, how the European Commission interprets it, and what it means for companion-style robots that greet, chat with and react to people at events. It is general information, not legal advice.

What the AI Act actually prohibits

Article 5(1)(f) of Regulation (EU) 2024/1689 prohibits placing on the market, putting into service for this purpose, or using AI systems that infer the emotions of a natural person "in the areas of workplace and education institutions". The only exception is where the system is intended for medical or safety reasons. The consolidated text of Article 5 on the Commission's AI Act Service Desk shows the wording, which reflects amendments made by the Digital Omnibus on AI.

The prohibitions in Chapter II have applied since 2 February 2025, according to Article 113. They are not a future deadline. Breaching any Article 5 prohibition can lead to fines of up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher, under Article 99(3). For SMEs, the lower of the two figures applies.

The Act defines an emotion recognition system in Article 3(39) as an AI system that identifies or infers emotions or intentions on the basis of biometric data. Biometric data covers physical, physiological or behavioural characteristics, such as facial images.

How the Commission reads the ban

The Commission first published its guidelines on prohibited AI practices on 4 February 2025. They are non-binding: only the Court of Justice of the EU can give an authoritative interpretation. They are still the clearest indication of how regulators are likely to approach the rules. The formally adopted English text, reference C(2025) 5052 final, devotes a full chapter to emotion recognition.

According to the guidelines, the ban applies only when four conditions are met at the same time:

1. An AI system is placed on the market, put into service or used.
2. The system identifies or infers emotions.
3. This happens in a workplace or an education institution.
4. The use is not for medical or safety reasons.

The ban applies to both providers and deployers. The guidelines also state that it applies to a deployer even if the provider's terms of use exclude that use. For events, this means the organiser or exhibitor operating the robot cannot rely on the supplier's contract alone.

Smiles are not emotions, but "happy" is

The guidelines draw a line that matters for robots. Detecting a readily apparent expression, gesture or movement is not emotion recognition. Concluding what someone feels from it is. They give examples: noticing that a person is smiling is not emotion recognition, but concluding that the person is happy is.

Several things fall outside the definition entirely:

• Physical states such as pain or fatigue. Recital 18 gives the example of fatigue detection for pilots and drivers.
• Text-based sentiment analysis, because written text is not biometric data. Inferring emotions from facial expressions, voice, posture or typing patterns does count.
• Crowd-level mood analysis that does not infer the emotions of a specific person, such as measuring general noise or mood levels at a venue.

The concept of emotion is to be read broadly. The guidelines warn that relabelling an emotion as an "attitude" does not avoid the ban.

Why event visitors are not covered, and staff are

The guidelines state plainly that emotion recognition used in a commercial context to address customers does not fall under Article 5(1)(f). One example they give is a call centre tracking its customers' anger or impatience, which is not prohibited. A robot at a brand activation that reacts to how visitors seem to feel is, on this reading, outside the ban.

The complication is that "workplace" is interpreted broadly. The guidelines say it covers any physical or virtual space where people carry out work, from offices and warehouses to publicly accessible places such as shops, stadiums and museums, as well as temporary or mobile work sites. This applies whatever the person's status: employee, contractor, trainee or volunteer.

An exhibition hall or event venue is therefore a workplace for the booth staff, promoters, hosts and technicians present. A robot that infers the emotions of those people falls within the prohibition. In a related example, the guidelines say that a supermarket using cameras to spot suspicious customers is not prohibited "when it is ensured that no employees are being tracked". They also say that a supermarket tracking its employees' emotions is prohibited.

In practice, a robot that runs emotion inference on every face in front of it, including the brand ambassador standing next to it, may cross the line. Whether a system can reliably exclude staff is a question to put to the supplier before the event.

Schools and STEM events

The guidelines say "education institutions" should be read broadly. The term covers public and private institutions, all ages, vocational schools and continuing training, whether teaching is online, in person or blended. Examples of prohibited uses include an institution using emotion recognition to infer students' interest and attention, and using it during admission tests.

The guidelines also note that an emotion-sensing language app used outside an education institution is not prohibited, but becomes prohibited if the institution requires students to use it. A robot brought into a school or university that infers pupils' emotions is therefore in a sensitive area. The safest assumption is to switch emotion inference off. Our guide to robots at school and STEM outreach events covers the wider planning.

Outside the ban does not mean unregulated

The guidelines state that emotion recognition systems outside the workplace and education are considered high-risk. Annex III, point 1(c) lists AI systems intended for emotion recognition, to the extent their use is permitted under Union or national law. Under the consolidated Article 113, the high-risk rules for Annex III systems apply from 2 December 2027. This date follows the Digital Omnibus changes explained in our AI Act timeline for robots.

One obligation already applies. Article 50(3) requires deployers of an emotion recognition system to inform the people exposed to it. Article 50(5) requires this to be done clearly and accessibly, at the latest at the first interaction or exposure. Article 50 falls under the Act's general application date of 2 August 2026.

Two other limits remain:

• Manipulation. The guidelines say emotion recognition in commercial settings may still fall under the bans on harmful manipulation and exploitation in Article 5(1)(a) and (b). Their example contrasts a transparent system that analyses customers' emotions to help them with a hidden one that times higher-priced offers for moments when consumers are more likely to buy.
• Data protection. The guidelines note that GDPR and consumer law continue to apply. A robot camera processing visitors' faces needs a lawful basis whatever the AI Act says. See our explainer on what EU regulators say about robot cameras and video.

The scientific doubts behind the rules

The reason for the strict approach is set out in Recital 44, which the guidelines quote. It refers to serious concerns about the scientific basis of emotion recognition, because the way people express emotions differs between cultures, situations and even within one individual. It names limited reliability, lack of specificity and limited generalisability as key shortcomings.

For event teams, this is also a practical point. A robot that "reads" a visitor's mood may simply get it wrong, especially across a varied international audience. Features marketed as emotion-aware should be judged on what they actually infer, not on the label.

What this means for event organisers

• Ask exactly what the robot infers. Face detection, smile detection and gesture tracking are different from emotion inference. Put the question in writing when you brief a robot supplier.
• Keep staff out of scope. If emotion features are on, make sure booth staff, hosts and operators are not being analysed. Otherwise, switch the feature off.
• Treat schools as a no-go for emotion inference. In education institutions, the ban applies to pupils, students and teachers alike.
• Tell visitors. Since 2 August 2026, people exposed to emotion recognition must be informed clearly at first contact. Signage and on-screen notices should be planned alongside the camera and privacy arrangements.
• Expect stricter rules from December 2027. Emotion recognition outside the ban becomes subject to high-risk requirements, which suppliers will need to address.

Frequently asked questions

Is it legal for a robot to read visitors' emotions at a trade fair in the EU?

The AI Act's emotion recognition ban in Article 5(1)(f) is limited to workplaces and education institutions, and the Commission's guidelines say emotion recognition aimed at customers in a commercial context is outside it. The system is still treated as high-risk, visitors must be informed under Article 50(3), and data protection law applies. Staff working at the venue must not be analysed, because the venue is their workplace.

Does detecting a smile count as emotion recognition under the AI Act?

No, not on its own. The Commission's guidelines, drawing on Recital 18, say that detecting readily apparent expressions such as a smile or a frown is not emotion recognition. It becomes emotion recognition when the system concludes what the person feels, for example that they are happy or angry. Detecting physical states such as fatigue or pain is also outside the definition.

Can a robot use emotion recognition at a school STEM event?

This is a sensitive area. The ban covers education institutions broadly, including private schools, all ages and vocational training, and the guidelines list inferring students' interest and attention as a prohibited use. The medical and safety exception is interpreted narrowly and does not cover general wellbeing. Turning emotion inference off for school visits avoids the question, and a legal adviser can assess specific cases.

When did the AI Act emotion recognition ban start to apply?

The prohibitions in Chapter II of the AI Act, including Article 5(1)(f), have applied since 2 February 2025. The transparency duty to inform people exposed to emotion recognition in Article 50(3) applies from 2 August 2026. According to the consolidated text after the Digital Omnibus, the high-risk requirements for Annex III systems, including emotion recognition, apply from 2 December 2027.

How this article was made: written by the humanoid4rent editorial team with AI assistance; every factual statement was checked against the sources listed below on 2026-10-31. Spotted an error? Write to info@humanoid4rent.com.