Emotion Recognition and Biometric Categorisation Notices
Article 50 extends beyond chatbots to systems that infer emotions or recognise biometric characteristics. This lesson covers the transparency duties when AI systems analyse emotional states from video, voice, facial expressions or physiological signals, and when systems recognise or infer protected characteristics from biometric data. These disclosures are among the most sensitive because they involve personal data and psychological or identity-based inference.
Emotion recognition AI systems are deployed in workplaces (performance monitoring), educational settings (assessing student engagement), healthcare (detecting patient distress) and retail environments (measuring customer satisfaction). Under Article 50, a person being analysed by these systems must be informed. The notice must be given before the analysis occurs, not after the emotion recognition system has already collected and processed data. In a workplace setting, this means the employer must notify employees that emotion recognition is in use, what emotions are being inferred, and how the results will be used.
Notice timing and format
A notice placed on a wall in a warehouse, stating that emotion recognition cameras are in operation, satisfies the notice requirement if the notice is clear and appears before a person enters the area. A notice in an employment contract satisfies the requirement if signed by the employee. A video placed on the employer’s intranet explaining the system satisfies the requirement if employees are required to watch it as part of onboarding. The critical distinction is that the person must have a reasonable opportunity to become aware of the system before they are analysed.
Biometric categorisation
Biometric categorisation systems infer protected characteristics (such as race, ethnicity, gender, age group, sexual orientation) from facial or other biometric analysis. These systems are heavily restricted under Article 5 (prohibited practices) when used by public authorities. Under Article 50, even systems that are not outright banned must disclose that they are in use. A retail store using facial analysis to infer age and gender for targeted advertising must disclose this to customers, even though age and gender inferences might not be prohibited in all contexts.
The notice should explain what biometric characteristics are being inferred and how that information will be used. A vague statement such as “We use biometric analysis” is less clear than “Our system infers your approximate age and gender from facial analysis to personalise advertising”. The purpose matters because it affects the risk to the individual. A system inferring age for age-restriction compliance (preventing minors from viewing certain content) carries different risks than a system inferring physical characteristics for discriminatory targeting.

Medical and healthcare exceptions
Healthcare settings have a partial exemption. A medical device that uses AI to recognise patient distress from facial expressions, for the purpose of alerting a clinician to a patient’s wellbeing, must still disclose that the system is in use, but the disclosure can be streamlined. The patient is in the care of a clinician and the system is designed to improve care, not to surveil or manipulate. The disclosure can state “Your vital signs and wellbeing are monitored by a clinical decision support system”. This is less stringent than the disclosure required in a workplace surveillance context, but it is still required.
Consent versus notice
Article 50 requires notice, not consent. A person must be told that emotion recognition or biometric analysis is in use, but they cannot be required to consent before being informed. This is a key distinction from other regulations like GDPR, which can require consent for personal data processing. Under Article 50, the information right is unconditional. A person has the right to know, even if they cannot opt out. However, organisations should consider whether additional regulatory requirements (such as employment law or GDPR) require consent or consultation in addition to the Article 50 notice.
Individual versus group notice
Where a system analyses multiple people (such as a group video conference or a crowd in a retail store), a group notice is acceptable. A notice to all employees that emotion recognition is in use satisfies the requirement for all employees. Individual notice to each person is not required. However, if a person can be identified by the system and the purpose includes influencing decisions affecting that individual (such as performance evaluation), individual notice may be required under additional regulations like employment law.
In-House Corporate AI Strategy & Governance Clinics
Bespoke executive clinics led by Dr. Stylianos Kampakis for leadership teams, boards, and division directors.
