Point 1 of Annex III AI Act includes the following use cases:
Remote biometric identification (RBI) systems;
AI systems intended to be used for biometric categorisation, according to sensitive or protected attributes or characteristics based on the inference of those attributes or characteristics; and
AI systems intended to be used for emotion recognition.
The aforementioned use cases are classified as high-risk given their potential intrusiveness on the rights and freedoms of the persons concerned by the systems , including because of the systems’ reliance on biometric data and their sensitive nature. As explained in Section 2.6 above, the fact that an AI system is classified as a high-risk AI system under these use cases should not be understood to mean that the use of the system is lawful under other acts of Union law or under national law compatible with Union law.
i. Biometric data
RBI systems involve the processing of ‘biometric data’. Article 3(34) AI Act defines ‘biometric data’ as ‘ personal data resulting from specific technical processing relating to the physical, physiological or behavioural characteristics of a natural person, such as facial images or dactyloscopic data’.
Personal characteristics from which biometric data may be extracted are physical, physiological or behavioural attributes of a natural person . Physiological and physical biometrics employ bodily, structural, and relatively static attributes of a person, such as their fingerprints, the pattern of their iris, contours of their face, or the geometry of veins in their hands. Some attributes are microscopic in nature, but still exhibit biological and chemical structures that can be acquired and identified e.g., DNA and odour9. Behavioural biometrics monitor the distinctive characteristics of movements, gestures, and motor-skills of natural persons as they perform a task or series of tasks. This means that human movements, such as walking (gait analysis) or finger contact with a keyboard (keystrokes), are captured and analysed. Behavioural biometrics encompass a variety of attributes that exhibit both voluntary and involuntary repeated motions and associated rhythmic timings or pressures of body features ranging from signatures, gait, voice, click patterns, and keystrokes , through to eye tracking and heartbeats10 , electroencephalography (EEG)11, or electrocardiograms (ECG)12. The biometric input may relate to one attribute (e.g., facial images) or multiple attributes (e.g., facial information combined with electroencephalogram (EEG)). Biometric data referring to characteristics of natural persons can not , as such, concern deceased individuals, dead bodies, human remains of deceased people or traces thereof. For example, the following AI systems do not fall within the use case of point 1 of Annex III:
- an AI system comparing DNA samples from a corpse;
- an AI system comparing facial image from human remains;
- an AI system comparing facial images, DNA, or fingerprints from a known deceased individual, coming from official register, including police records.
AI systems intended to be used to search for objects that do not make use of biometric data to track the movement of a suspect , e.g. searches based on number plates, specific clothes, weapons, background, etc., will not be classified as high-risk pursuant to point 1(a) of Annex III AI Act.
Examples with regard to the element of biometric data:
- An AI system inferring emotions of an individual from key stroke (way of typing), facial expressions, body postures or movements is based on biometric data.
- An AI system following a person with a red scarf or a blue jacket is not based on biometric data.
- An AI system used in supermarkets to detect whether a product is being put into a basket , cart or into a bag is not based on biometric data when it only performs object recognition. The same is true for a system used at train stations or airports that detects suspicious objects.
- An AI system inferring emotions from typed text (content, sentiment analyses or stylometry) to define the style or the tone of a certain article is not based on biometric data.
The AI Act ’s definition of biometric data is broad and includes any biometric data used for biometric identification, biometric categorisation, emotion recognition or other purposes. Contrary to the definition of biometric data in the EU data protection law, including Regulation (EU) 2016/679 (‘GDPR’), Directive (EU) 2016/680 (‘LED’) and Regulation (EU) 2018/1725 (‘EUDPR’), the definition of biometric data in the AI Act does not include the wording ‘which allow or confirm the unique identification’ (the functional use of biometric data)since it aims to , cover AI systems that use biometric data not only in cases where the AI system identifies an individual or confirms its identity as specified in Recital 14 AI Act.
ii. Interplay with prohibited practices
The use cases listed in point 1 of Annex III should be considered in the light of the AI practices prohibited by Article 5(1 )(f), (g), (h) and (2) to ( 5) AI Act. That is because , in specific instances where all the conditions for one or more prohibit ed practices under Article 5 AI Act are met, the placing on the market, putting into service and use of the AI system is prohibited. It will then be unnecessary to consider whether those systems should be classified as high-risk. Most AI systems that fall under an exception from the prohibitions provided in Article 5 AI Act will qualify as high-risk.
For example, emotion recognition systems, where they do not fulfil the conditions for the prohibition of Article 5(1)(f) AI Act, will fall within the use case of point 1(c) of Annex III AI Act and be classified as high-risk pursuant to Article 6(2),insofar as the conditions under Article 6(3) AI Act do not apply. In particular, emotion recognition systems outside the areas of workplace and education institutions, as well as such systems used in the areas of workplace and education institutions, when used for medical or safety reasons, are not prohibited, but qualify as high-risk. Similarly, biometric categorisation systems that do not fulfil the conditions for the prohibition of Article 5(1)(g) AI Act will fall within the use case of point 1(b) of Annex III and be classified as high-risk pursuant to Article 6(2) AI Act , where they are intended to be used for biometric categorisation according to sensitive attributes or characteristics protected under Article 9(1) GDPR (insofar as the conditions of Article 6(3) AI Act do not apply). In contrast, RBI systems that fall within the use case of point 1(a) of Annex III AI Act may in some cases be prohibited if they fulfil the conditions for the prohibition under Article 5(1)(h) AI Act, i.e. they are being used in real-time, at publicly accessible places , for law enforcement purposes , and without one of the exceptions applying. Conversely, RBI systems that fall under one of the exceptions of Article 5(1)(h) AI Act13 will be classified as high-risk under Article 6(2) AI Act , as w ill RBI systems that are not used for law enforcement purposes or not in real time.
- Physiologicaland Behavioural Biometrics –BiometricsInstitute, available online: https://www.biometricsinstitute.org/physiological-and-behavioural-biometrics/.
- Physiological and Behavioural Biometrics –BiometricsInstitute; available online: https://www.biometricsinstitute.org/physiological-and-behavioural-biometrics/.
- See EDPS, TechDispatch 1/2024 –Neurodata, 3.6.2024, in which the use of brain data and related technology is discussed, as well as the legal implication, including the proposition of new ‘neurorights’, including mental privacy and integrity,In S. O’Sullivan, H. Chneiweiss, A. Pierucci and K. Rommelfanger, Neurotechnologiesand Human Rights Framework: Do we need new Human Rights? , available online: https://rm.coe.int/rapporteur-report-neurotechnology-final-e-2758-7638-0166-1/1680a90ae0Report, OECD and CoE, 9.11.2021, p.33 , a state of the art and legal aspects of neurotech is discussed.
- See Hasnul et al., 2021, Electrocardiogram-Based Emotion Recognition Systems and Their Applications in Healthcare, available online: https://www.mdpi.com/1424-8220/21/15/5015.
- The exceptions setout in Article 5(1)(h)(i)-(iii) AI Act aim to allow the use of certain AI and investigative tools for law enforcementpurposes: (i) the targeted search of victims of three specific serious crimes and missing persons [protection]; (ii) the prevention of imminent threats to life or physical safety or a genuine threat of terrorist attacks [prevention]; and (iii) the localisation or identification of suspects and offenders of certain serious crimes as listed in Annex II AI Act [prosecution/investigation].