Safeguards and conditions for the exceptions (article 5(2)- (7) AI Act)

The use of real-time RBI systems for one of the objectives listed in Article 5(1)(h)(i) to (iii) AI Act is subject to certain safeguards and conditions, which are detailed in Article 5(2) to Article 5(7) AI Act.

First, the use of real-time RBI systems in publicly accessible spaces for law enforcement purposes is only allowed to ‘confirm the identity of the specifically targeted individual.’ This first condition aims to balance the seriousness of the situation and the harm resulting from not using the system with the impact of the technology on individuals’ rights and freedoms. It aims to avoid mass surveillance by targeting an individual for the deployment of real-time RBI. As a consequence, the deployment of a real-time RBI system in publicly accessible spaces for law enforcement purposes should only be authorized for targeted individuals.

The use of the expression ‘confirming the identity’, as opposed to ‘identification’, is meant as an additional safeguard for the fundamental rights limiting the risk of indiscriminate surveillance and implies that the identification of an individual within the meaning of Article 5(1)(h) AI Act must be targeted. That expression should be understood as meaning that the use of real-time RBI may only be initiated to search for specific individuals for which the law enforcement authorities have reasons to believe or are informed that they are victims of the crimes listed in Article 5(1)(h)(i) AI Act or are involved in one of the scenarios described in Article 5(1)(h)(ii) or Article 5(1)(h)(iii) AI Act. This means, in practice, a comparison of the data collected real-time with the data contained in the reference database. As regards the use of real-time RBI system in the scenarios described in Article 5(1)(h)(ii) AI Act and for conducting a criminal investigation within the meaning of 5(1)(h)(iii) AI Act, law enforcement authorities do not necessarily need to know the identity of the individuals they are searching for before using the system. If they have factual indications and information about a planned terrorist attack by a terrorist group (without knowing who will execute the plan) at a specific time and place, the RBI system may be used to identify the offender from the terrorist group, provided the law enforcement authorities have constituted a reference database containing the biometric data of the individuals forming part of the terrorist group. In all three scenarios described in Article 5(1)(h)(i) to (iii) AI Act, ‘confirming the identity’ may also include the localisation of the person in question.

Second, before using the system, the nature of the situation giving rise to the possible use, in particular, the seriousness, probability and scale of the harm for natural persons, society and law enforcement purposes that would be caused if the system were not used, should be assessed against the consequences of the use of the system on the rights and freedoms of the persons concerned, in particular, the seriousness, probability and scale of those consequences. This should include evaluating whether less intrusive alternative solutions are available to the law enforcement authorities or entities acting on their behalf.

For example, law enforcement authorities are prohibited from using real-time facial recognition systems in the street based on general security, crime prevention and over- crowding concerns, since that would involve the constant monitoring and surveillance of all persons, it is not limited in time, and it would therefore not meet the criteria for the exception from the prohibition laid down in Article 5(1)(h) AI Act.

The ‘seriousness’ criterion, applied here in connection to the possible harm and consequences, implies a variation in degrees of interference with the fundamental rights at stake, which is linked to the principle of proportionality. Concerning the interferences with fundamental rights, some interferences are viewed as more serious than others.

The ‘scale’ criterion refers, in particular, to the number and categories of persons affected by the interference (including children and vulnerable or marginalised persons).

Finally, the ‘probability’ is the likelihood that an event will occur.

The assessment of the seriousness, scale and probability of the harm and consequences should all be part of the Fundamental Rights Impact Assessment that the law enforcement authority is obliged to complete (see below). That assessment will be concluded on a case-by-case basis.

Third, the real-time use of RBI should be clearly limited in terms of geographic scope, duration, and the targeted person. This is to ensure that the RBI system is only used when strictly necessary.

Concerning the geographic restriction, it may cover one or several geographical areas based on ‘objective and non-discriminatory factors’. In the case of biometric identification this implies that the geographic restriction applies to a clearly delineated boundary for which there are indications that the event will take place. Such a delineation should -under normal circumstances- not comprise an entire city or country, but should be more targeted.

Another safeguard relates to the personal scope of the measure, i.e. defining the categories of persons concerned. This would exclude the untargeted, indiscriminate identification of persons, without further indications of an incident.

Finally, the time limit is a period limited to what is strictly necessary, but which may be extended in case of need in accordance with the applicable rules. The use of real- time RBI systems therefore cannot be for an indefinite or vague period of time. The period needs to be determined in light of the concrete indications that lead to the use of RBI systems.

Fourth, before deployment, the law enforcement authority deploying the real-time RBI system must have conducted a Fundamental Right Impact Assessment (FRIA) and registered the system in the EU database (except in a duly justified case). 10... Fundamental Rights Impact Assessment

FRIAs carried out in application of Article 5(2) AI Act must comply with the conditions laid down in Article 27 AI Act. That provision sets out the requirements concerning FRIAs applicable to high-risk AI systems.

In the period between when the prohibitions in Article 5 AI Act become applicable (after 2 February 2025) but the provisions on high-risk AI systems are not yet applicable (before 2 August 2026), the requirements for FRIA set out in Article 27 AI Act should be implemented by the deployers of real-time RBI systems meeting the conditions to benefit from one or more of the exceptions in Article 5(1)(h) AI Act. The following provisional guidance relates only to the use of real-time RBI in publicly accessible spaces for law enforcement purposes for the period before the obligations for high-risk AI systems become applicable and the Commission adopts the template for FRIA and provides further guidance on the obligation under Article 27 AI Act.

A FRIA is a new type of impact assessment that aims to identify the impact that certain high-risk AI systems, including RBI systems, may produce on fundamental rights. A FRIA is an accountability tool. The FRIA does not replace the existing Data Protection Impact Assessment (DPIA) that data controllers (i.e. those responsible for the processing of personal data) must perform under Article 27 LED, Article 35 GDPR or Article 39 EUDPR.

For example, a DPIA must be conducted when biometric data are processed through new technologies likely to result in a high risk to the rights and freedoms of natural persons (such as CCTV, AI facial recognition, and body-worn cameras) in publicly accessible spaces.

Whereas a DPIA focuses on the risks to the rights and freedoms of individuals resulting from the processing of their personal data, a FRIA covers the possible impact of AI systems on individuals’ fundamental rights more generally. The scope of a FRIA is therefore broader in terms of activities covered and fundamental rights assessed. Where personal data are processed by the AI system (which is the case of RBI systems), the FRIA should complement the DPIA performed by the deployer as data controller, without covering aspects already addressed in the DPIA and avoiding overlaps. The analysis of the FRIA in these Guidelines is limited to the authorized use of RBI in real- time and is aimed to serve as a preliminary guidance for deployers in this interim period before the AI Office provides a template.

The obligation to carry out the FRIA under Article 5(2) AI Act is imposed on the deployers of the RBI system, and not entities or bodies or anyone else acting on their behalf. If other actors are acting on behalf of the deployer/the law enforcement authority, they will have to contribute to the preparation of the FRIA with all relevant information to ensure it is properly carried out.

A FRIA must be carried out before the deployment of the authorized real-time RBI system.

According to Article 27 AI Act, a FRIA should include the following information: • A description of the RBI use and the deployer’s processes for the use, together with the intended purpose of use The description should include:

- the name of the deployer;

- the law enforcement purpose(s) for which the real-time RBI system will be used; - the description of the reference database against which the biometric identification will be compared, including the sources of the biometric data (facial images, voice samples, etc.) that will be used;

- the description of the technology underlying the system to explain its functioning (by referencing the available documentation provided by the provider and its name); - the legal basis on which the real-time RBI will be deployed.

• The period of use and frequency of use Each individual use of a real-time RBI system for one of the permitted exceptions must be authorised prior to its deployment by a judicial authority or other independent authority under Article 5(3) AI Act. By contrast, for the FRIA, deployers must provide a general indication of the intended period of use and the expected frequency. • The categories of persons and groups affected by the system For the purpose of the exception in Article 5(1)(h) AI Act, the FRIA should distinguish between:

- The targeted individual, who may be a victim of a crime, a perpetrator, or a suspect, - The individuals whose biometric data are included in the reference database, and - Categories of persons who are present in the surrounding areas where the RBI system will be deployed.

The use of real-time RBI systems will not only affect the fundamental rights of the targeted individual. The rights of other individuals whose biometric data are used for comparison purposes, passersby, and people incidentally presented in the search area will also be affected. The description of the geographic scope of the search area(s) covered by the real-time RBI system will impact the number of persons affected by the system.

• The specific risks of harm to the affected persons

The fundamental rights that may be affected by the use of real-time RBI in publicly accessible spaces for law enforcement purposes include, in particular: - the right to private and family life, including people’s reasonable expectation of anonymity in public spaces;

- the right to data protection, as RBI systems rely on the processing of biometric data and other personal data (e.g., names, ID numbers, as well as sensitive data such as ethnicity) to identify specific individuals;

- freedom of thought, conscience and religion, freedom of expression and freedom of assembly and association in the public spaces being searched, on which the use of RBI systems could have a chilling effect, preventing individuals to fully exercise their rights and freedoms, since if individuals know that they are monitored, they might change their behaviour, or even prevent themselves to behave in a certain manner;

- the right to an effective remedy and a fair trial;

- the right to non-discrimination if the system embeds biases (such as gender, ethnic or racial biases) and leads to the misidentification of a suspect or perpetrator; - the right to human dignity by the feeling of being reduced to an object of the system; - the presumption of innocence and right to defence; since no decision adversely affecting an individual may be solely taken on the output of the real-time RBI system;

- the rights of the child in case the victim, missing person or suspect is a minor; - the rights of the elderly in the case of a missing person.

To assess the specific risks of harm likely to impact the identified affected person(s) or group(s), the FRIA must identify the fundamental rights of those persons and assess the impact on their fundamental rights, including the severity of the impact and its scale, taking into account the potentially affected persons.

This part of the FRIA should also include the assessment of whether the use of a real- time RBI system is necessary and proportionate considering the objectives and the circumstances in which its use is intended, including the existence or absence of less intrusive alternatives. The FRIA should describe the performance and accuracy level of the system, based on the technical documentation and, if available, the training data on which the technology was tested and developed to prevent biases and discrimination. The FRIA should also identify the impact of use of a real-time RBI system on the fundamental rights of all individuals potentially affected, in particular the suspect or perpetrator, the victim searched and other individuals present in the publicly accessible spaces subject to the search. To the extent that the system processes the biometric data of these individuals, their rights to private and family life and data protection will be impacted, which will be assessed as part of the DPIA as far as the data processing activities are concerned. For other activities related to the use of the real-time RBI systems and the impact on other fundamental rights, the FRIA will complement the DPIA. Depending on the context of deployment, other fundamental rights of these individuals, such as their rights to human dignity, freedom of thought, conscience and religion, assembly or freedom of expression, rights to an effective remedy and a fair trial, presumption of innocence and right of defence, rights of the child, may be impacted.

The assessment under the FRIA should be performed at an abstract level, prior to the first putting into service of the AI system. Specific context-dependent considerations that determine the impact of use in each individual case where a real-time RBI system is used should be further elaborated in the individual request for requesting the authorisation by a judicial authority or other independent administrative authority of each use of the RBI system (see section 10.... below).

• Human oversight measures According to Article 5(3) AI Act, no decision that would adversely affect an individual may be taken solely on the basis of the output of the real-time RBI system. As a consequence, the FRIA should describe the procedures that will be followed during the operation of the system and how the output will be interpreted in the context of decision- making process. The procedures should provide instructions on the deployment of the RBI system, clarify the role of a human agent in verifying and interpreting the output and provide training to operate the system. The person in charge of human oversight should have sufficient ‘AI literacy, training and authority’ to understand how the system functions and when it underperforms or malfunctions.

Other considerations for human oversight and monitoring under Articles 14 and 26 AI Act are also relevant and should be described.

• Risk mitigation measures

Beyond implementing human oversight measures (including to avoid discriminatory measures), the deployer should explain redress measures in case a risk materialises, including the governance procedures and the complaint mechanisms (such as in the case of a misidentification).

10... Registration of the authorized RBI systems

Article 5(2) AI Act also obliges the deployer of a real-time RBI system used in publicly accessible spaces for law enforcement purposes to register the system in the EU database provided for in Article 49 AI Act. However, in cases of a duly justified emergency (such as an imminent threat), deployment may start even prior to registration, provided the law enforcement authority registers the system without undue delay. Undue delay should be understood as meaning ‘as soon as possible’ considering the circumstances of the emergency that prevented the registration of the system prior to its use. Whether registration meets that criterion requires an appreciation on a case- by-case basis. It cannot be defined a priori with a precise time limit. The delay should not be caused by a deliberate action. According to Article 49(4) AI Act, RBI systems used for the purpose of law enforcement will be registered in a secure non-public section of the database, with limited information and limited access to that information.

For example, requesting law enforcement authorities to register the RBI system within 24 hours of the use might be considered a reasonable delay where the system was deployed in a situation of an imminent threat to life, such as in the scenario of a live shooter.

Article 5(3) AI Act requires prior authorisation of each individual use of a real-time RBI system and prohibits automated decision-making based solely on the output of such a system which produces an adverse legal effect.

Article 5(3) AI Act provides:

For the purposes of paragraph 1, first subparagraph, point (h) and paragraph 2, each use for the purposes of law enforcement of a ‘real-time’ remote biometric identification system in publicly accessible spaces shall be subject to a prior authorisation granted by a judicial authority or an independent administrative authority whose decision is binding of the Member State in which the use is to take place, issued upon a reasoned request and in accordance with the detailed rules of national law referred to in paragraph 5. However, in a duly justified situation of urgency, the use of such system may be commenced without an authorisation provided that such authorisation is requested without undue delay, at the latest within 24 hours. If such authorisation is rejected, the use shall be stopped with immediate effect and all the data, as well as the results and outputs of that use shall be immediately discarded and deleted.

The competent judicial authority or an independent administrative authority whose decision is binding shall grant the authorisation only where it is satisfied, on the basis of objective evidence or clear indications presented to it, that the use of the ‘real- time’ remote biometric identification system concerned is necessary for, and proportionate to, achieving one of the objectives specified in paragraph 1, first subparagraph, point (h), as identified in the request and, in particular, remains limited to what is strictly necessary concerning the period of time as well as the geographic and personal scope. In deciding on the request, that authority shall take into account the elements referred to in paragraph 2. No decision that produces an adverse legal effect on a person may be taken based solely on the output of the ‘real-time’ remote biometric identification system.

10... Objective

The objective for requiring prior authorisation (‘authorisation ex ante’) for any use of a ‘real-time’ RBI system in publicly accessible spaces for law enforcement purposes is the need for an assessment and a decision as to whether any envisaged use of such a system for such purposes is:

- necessary and proportionate to achieve any one of the objectives listed in Article 5(1)(h)(i) to (iii), i.e. for the targeted search of specific victims, the prevention of specific threats, or the localisation or identification of offenders; and - limited to what is strictly necessary concerning the time period and the geographic and personal scope.

The consequence of these requirements is that a double necessity and proportionality assessment should occur prior to the deployment of any real-time RBI system in publicly accessible spaces for law enforcement purposes. First, an assessment should be made by the user when performing a FRIA, as required by Article 5(2) AI Act. Second, in accordance with Article 5(3) AI Act, a judicial or independent administrative authority must also assess the necessity and proportionality of using such a system within the limits of the national law providing the legal basis for any such use, taking the Charter and other Union law into consideration. As a consequence, any such system may only be used 1) after a FRIA and 2) when the competent national authority has authorised such use.

Article 5(3) AI Act must be read and understood in conjunction with Article 5(5) AI Act: for the use of a real-time RBI system to be authorized, a national law adopted in the Member States concerned must exist authorising such use. Certain Member States already have a system of prior authorisation in place for the use of biometric systems under other Union or national law, such as data protection law. 10... The main principle: Prior authorisation by a judicial authority or an independent administrative authority

The use of real-time RBI systems which pursue one of the objectives listed in Article 5(1)(h)(i) to (iii) AI Act and which has been provided for in the national law of the Member States concerned must be authorized by a judicial authority or an independent administrative authority prior to its use. This is the main principle.

However, there is an exception in the case of urgency. This shall be duly justified. Urgency is described as ‘situations where the need to use the systems concerned is such as to make it effectively and objectively impossible to obtain an authorisation before commencing the use of the AI system’. In such case of urgency, ‘the use of the AI system should be restricted to the absolute minimum necessary and should be subject to appropriate safeguards and conditions, as determined in national law and specified in the context of each individual urgent use case by the law enforcement authority itself.’

10.... Prior and reasoned request in accordance with national procedural rules a) Request by whom?

Whilst not specified, it may be assumed that the request will normally be initiated by the deployer, i.e. by the competent (law enforcement) authority. According to the definition of law enforcement authority under Article 3(45) b) AI Act, any ‘other body or entity entrusted by Member State law to exercise public authority and public powers for the purposes of the prevention, investigation, detection or prosecution of criminal offences or the execution of criminal penalties, including the safeguarding against and the prevention of threats to public security’ is considered a law enforcement authority and could also be the responsible as the ‘competent authority’ for submitting the request for prior authorisation.

The use of a real-time RBI system for activities falling outside the scope of the AI Act does not need to be authorised under Article 5(3) AI Act. If subsequently such a system is being used for law enforcement purposes, the use would fall within the scope of the AI Act and an authorisation would be required where the requirements of Article 5(1)(h) AI Act are met.

b) Request for which use?

Prior authorisation is needed for the use of ‘real-time’ RBI systems in publicly accessible spaces for law enforcement purposes, even if the systems are operated by other parties on behalf of law enforcement authorities, for example sport clubs or shopping malls.

For example, - An organisation entrusted with resources for searching for missing children decides to use a real-time RBI system. It has no mandate to exercise public authority and public powers or for preventing criminal offences or tasks for the prevention of threats to public security. Such use does not fall under the prohibition laid down in Article 5(1)(h) AI Act, since it is not for law enforcement purposes. That system will, however, be categorised as ‘high-risk’ (point 1 (a) of Annex III) and a requirement of prior consultation of the supervisory data protection authority may be necessary pursuant to Article 36 GDPR. Depending on the applicable national law and whether one of the exceptions to Article 9(1) GDPR applies, a prior authorisation may also be required for such processing. By contrast, if the same organisation were requested by law enforcement authorities to act on their behalf for the search of missing children in a law enforcement context and under the supervision and instructions of the competent law enforcement authorities, prior authorisation would be needed pursuant to Article 5(3) AI Act.

- A private organisation entrusted with providing resources to aid persons who risk becoming victims of a natural disaster decides to use a real-time RBI system for that purpose. Such use does not fall under the prohibition laid down in Article 5(1)(h) AI Act, since it is not for law enforcement purposes. That system will, however, be categorised as ‘high-risk’ (point 1 (a) of Annex III) and a requirement of prior consultation of the supervisory data protection authority may be necessary pursuant to Article 36 GDPR. Depending on the applicable national law and whether one of the exceptions to Article 9(1) GDPR applies, a prior authorisation may also be required for such processing.

c) When? ‘Each use’

In accordance with Article 5(3) AI Act, prior authorisation is needed for ‘each use’. This implies that the decisive moment for obtaining such authorisation is not the moment prior to installing real-time RBI systems, but each concrete use thereof. For example, - the police install biometric ready CCTV cameras at the main train station of a city (no authorisation under the AI Act is needed, but the biometric system must comply with the requirements on high-risk systems, a FRIA must be prepared prior to the first use and an individual authorisation by a judicial or independent administrative authority is needed before each individual use of the system).

The police has concrete indications that a terrorist will arrive by train in the town (prior authorisation is needed for real-time identification). d) Motivated Request

Article 5(3) AI Act requires each individual request for the use of real-time RBI to be ‘reasoned’ and hence substantiated and motivated.

Certain Member States allow such requests to be submitted online. In accordance with Article 5(5) AI Act, national legislation should lay down requirements regarding the exact content of the request, while fully taking into account the requirements outlined above, including sufficient evidence to determine the strict necessity and proportionality for the use of real-time RBI and other relevant aspects to reflect the exceptional nature of authorising such use.

10.... Authorisation by a judicial authority or an independent administrative authority

The authorisation may only be granted by a judicial or an independent administrative authority whose decision is binding.

a) Independent authority

The CJEU has interpreted the concept of ‘independence’ in different contexts. In HK v Prokuratuur, for example, the CJEU explained that independence means that the authority maintains a ‘neutral stance’. The CJEU specified that an authority involved in previous investigations, in that case the public prosecutor, does not have such independence. Similar considerations may apply as regards the independence required by Article 5(3) AI Act, implying that the authorising authority needs to be independent from the authority using the RBI system. This would apply not only for the police, but also cases of investigative judges or prosecutors that supervise the work of police and the use of RBI for which authorisation is sought.

In Commission v Poland, a case dealing with the question when a body can be considered to be independent in the context of railway safety, the CJEU found that, ‘as regards public bodies, independence usually refers to a status that ensures that the body in question is able to act completely freely in relation to those bodies in respect of which its independence is to be ensured, shielded from any instructions or pressure’. Similar indications may apply in the context of Article 5(3) AI Act.

The judicial authorities in a democratic society are in general also independent authorities. The judiciary plays an important role when it is independent from the executive government(s) and from the legislator, covering and reviewing the application of the legislation and fundamental rights and freedoms in an autonomous and independent way. Judicial independence is one of the crucial facets of the rule of law and is guaranteed by Article 47 (Charter) and Article 6(1) ECHR. b) Authority of the place where the use will take place

The authorisation must be addressed to the authority that is competent according to national law. c) Authorisation only if ‘necessary and proportionate’ to achieving one of the objectives set forth in the exceptions

Any authorisation to use real-time RBI in publicly accessible spaces for law enforcement purposes must assess whether the requirements of Article 5(3) AI Act have been met.

Highly intrusive

In the data protection context, the use of biometric data, in particular facial recognition technology, has been considered by the European Data Protection Board (EDPB) in its Guidelines 5/2022 and the European Data Protection Supervisor (EDPS) as affecting several fundamental rights and freedoms. That view is shared by The EU Agency for Fundamental Rights and by the Council of Europe. Both the CJEU and the ECtHR have confirmed the sensitive nature of processing biometric data.

Any interference in fundamental rights and freedoms must always respect the essence of the rights and freedoms. This follows from Article 52(1) Charter.

The notion of ‘essence’ of fundamental rights and freedoms has been developed in the CJEU’s case-law and is an independent value in the Union’s legal order. If the essence of a fundamental right or freedom is not respected, it means that a right or freedom is unduly touched by a measure so that no interference shall be allowed upfront. Only ‘if necessary and proportionate’

Any interference with fundamental rights and freedoms requires ‘a law’ that should in principle respect the necessity and proportionality pursuant to Article 52 Charter. (See below under Article 5(5) AI Act.) Article 5(3) AI Act requires that the national law permitting the use of real-time RBI in publicly accessible spaces for law enforcement purposes must provide that authorisation for such use is permitted ‘only where it [the authority] is satisfied [..] that the use is necessary for, and proportionate to, achieving one of the objectives specified’ in Article 5(1)(h) AI Act. National authorities need to verify whether the biometric identification is strictly necessary. This assessment should be based on the FRIA that should have already include an assessment of the necessity and the proportionality in a general manner prior to requesting the authorisation for each use with the specific circumstances.

10.... The exception to the requirement of prior authorisation: request within 24 hours and consequences if rejected

In cases of urgency, a user may submit a request for authorisation within 24 hours as from the moment that the real-time RBI system is used. In practice, that will generally be the moment that the biometric ready or capably cameras are ‘switched on’ and deployed and the first biometric comparison is made with the system. The logging of the processing activities should be made available to the authority to substantiate the timeliness of the request.

In such a case, the request should motivate why no prior request was submitted prior to starting using the system.

10.... Immediate cessation in case the request for authorisation is rejected and deletion of the data

Article 5(3) AI Act further provides that if an authorisation request in the case of urgency is rejected, the use of the real-time RBI system should be ceased with immediate effect. In such cases, all the data, including the results and outputs of that use, must be immediately discarded and deleted. Article 5(3) AI Act is explicit in this regard, without exception. The deployer will have:

a) a reference database, containing the biometric information (e.g., facial images, voice snippets, ….) and related identifying information, if applicable, against which b) captured biometric information from individuals present in the publicly accessible space is compared to identify and single out those individuals. c) This comparison will lead to the comparison result.

The requirement to discard and delete the data collected and processed also means that the reference database(s) used for the unauthorized biometric identification must be removed and deleted if it was built specifically for the contested search. Only where the law enforcement authorities had built and intended to maintain the database used for identification in a lawful manner for legitimate aims other than for the unauthorized use of real-time RBI may the database be maintained.

Besides the deletion of any (unlawful) database with biometric information, all the collected images and other personal data, including the meta data, technical processing data, including the templates and other personal data, and other comparison - and output data obtained during the unlawful use of the real-time RBI system must also be deleted.

Where the law enforcement authority challenges the rejection, the data may be kept by a trustee until a final decision has been taken on the request. During that period, those data should normally not be placed at the disposal of the law enforcement authority. 10.... No decision-making solely on the output of the real-time RBI system

In accordance with Article 5(3) AI Act, even where the deployer of a real-time RBI system obtains an authorisation, no decision that produces an adverse legal effect on a person may be taken based solely on the output of the ‘real-time’ RBI system. For example, - A person is arrested and imprisoned for a serious crime solely based on identification by a facial recognition system, without any further checks. This comes on top of the requirement under Article 14 AI Act for human oversight. Checks could relate for example to the question whether a given person has been at a different place or also whether there are other reasons for that the person cannot be the person searched. Requirements of Article 14 AI Act for Human Oversight

The use of real-time RBI that is permitted because it pursues one of the objectives listed in Article 5(1)(h) and complies with Articles 5(2)-(6) AI Act still falls under the rules for high-risk systems. In accordance with Article 14 AI Act, high-risk AI systems ‘shall be designed and developed in such a way, including with appropriate human-machine interface tools, that they can be effectively overseen by natural persons during the period in which they are in use.’ Pursuant to Article 14(5) AI Act, no action or decision may be taken by the deployer on the basis of the identification resulting from the system, ‘unless that identification has been separately verified and confirmed by at least two natural persons with the necessary competence, training and authority’ or unless ‘Union or national law considers the application of this requirement to be disproportionate’. Article 4 AI Act prescribes AI literacy measures for providers and users of AI systems to ensure ‘a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems’ and considering the persons on whom the systems are to be used.

As stated by the EDPB in the data protection context, for human oversight to be effective it is crucial ‘to enable the person to understand the (in that case facial recognition) system and its limits as well as to interpret its results properly. It is also necessary to establish a workplace and organisation that counteracts the effects of automation bias, and avoids fostering the uncritical acceptance of the results e.g., by time pressure, burdensome procedures, potential detrimental career effects etc.’ Similar considerations may apply in the context of the AI Act.

Each use of an RBI system pursuing one of the objectives listed in Article 5(1)(h)(i)- (iii) AI Act must be notified to the relevant market surveillance authority and the national data protection authority. Notification must take place after each use in order to be able to report about the number of authorisations and their result. The notification does not need to include sensitive operational data. According to Article 3 (38) AI Act, ‘sensitive operational data’ means operational data related to law enforcement activities (prevention, detection, investigation or prosecution of criminal offences), the disclosure of which could jeopardise the integrity of criminal proceedings.

For the details on the reporting requirement, see section 10. below.

National laws are required for operationalising the use of ‘real-time’ RBI systems in publicly accessible spaces for the purposes of law enforcement. At the same time, Article 5(5) AI Act provides that Member States remain free to decide whether to adopt such national laws. If a national law authorising the use of real-time RBI is adopted, the AI Act specifies the substantive elements which the national laws must contain to comply with the requirements laid down in the AI Act.

Article 5(5) AI Act A Member State may decide to provide for the possibility to fully or partially authorise the use of ‘real-time’ remote biometric identification systems in publicly accessible spaces for the purposes of law enforcement within the limits and under the conditions listed in paragraph 1, first subparagraph, point (h), and paragraphs 2 and 3. Member States concerned shall lay down in their national law the necessary detailed rules for the request, issuance and exercise of, as well as supervision and reporting relating to, the authorisations referred to in paragraph 3. Those rules shall also specify in respect of which of the objectives listed in paragraph 1, first subparagraph, point (h), including which of the criminal offences referred to in point (h)(iii) thereof, the competent authorities may be authorised to use those systems for the purposes of law enforcement. Member States shall notify those rules to the Commission at the latest 30 days following the adoption thereof. Member States may introduce, in accordance with Union law, more restrictive laws on the use of remote biometric identification systems. 10... National law shall respect the limits and conditions of Article 5(1)(h) AI Act

Since the use of ‘real-time’ RBI systems in publicly accessible spaces for law enforcement purposes is considered an interference with fundamental rights, Article 5(5) AI Act provides that such use shall be established by national law in the Member States. Those national laws provide the legal basis for the use of such systems.

National laws shall not exceed the limits set by Article 5(1)(h) AI Act and shall respect all further related conditions set forth in the AI Act. That implies that the Member States may not expand the objectives for which real-time RBI may be used in publicly accessible spaces for law enforcement purposes beyond those listed in Article 5(1)(h)(i)-(iii) AI Act.

Member States shall notify their national laws to the Commission at the latest 30 days following the adoption thereof. Such notification does not provide a presumption of conformity of the Member States’ law with the AI Act. The AI Office, after receiving the notification, will send a confirmation of receipt. Prior to its adoption, Member States are also encouraged to send a preliminary version of the national (or regional) law proposed to the AI Office. In any case, non-notification to the AI Office within the legal deadline of 30 days following the adoption set out in Article 5(5) may imply that the national law is unenforceable in legal proceedings, as has been held in different contexts. The Commission will publish the Member States’ laws on a public website.

Member States may introduce, in accordance with Union law, more restrictive laws, i.e. laws with stricter requirements than those laid down in Article 5(1)(h) and (2) to (7) AI Act.

10... Detailed national law on the authorisation request, the issuance and the exercise

As regards the detailed rules that apply for the request, the issuance and the exercise of the authorisation, these are to be determined by national law. Each Member State that wishes to allow the use of the systems at issue must specify in its national laws such rules, which are aimed at providing relevant and complete information as to the use of real-time RBI systems to the authorising authority to enable it to decide as to the strict necessity and proportionality of such use.

The national law permitting the use of real-time RBI systems may regulate for example, - who are the competent authorities subject to Article 5(1)(h) AI Act and the independent authorities in the Member State competent for issuing (or refusing) an authorisation;

- the detailed scope of the objectives for which real-time RBI in publicly accessible spaces may be used for law enforcement purposes (without going beyond the objectives listed in Article 5(1)(h) (i) to (iii), but possibly narrowing them further down;

- providing that requests shall be in writing and requiring a detailed explanation of the specific use and the intended purpose of use for a specific criminal offence/situation which justifies its use;

- the requirement of motivation and the submission of supporting evidence (and need of translation if relevant) for justifying the use of the system pursuing the objectives listed in Article 5(1)(h)(i) to (iii) AI Act, in particular relating to place, period, and personal scope and justifying the strict necessity and proportionality, including the relevance, sufficiency and the efficiency of the use of the system and the absence of less intrusive means;

- the description of the technology that will be used and the location points of the data collection;

- the minimum reliability, threshold used, and accuracy rates of the systems used; - the possibility of auditing the submitted information, including the technical details and accuracy criteria any time ex ante and ex post by the authorising authority; - the specification of the reference databases used;

- the retention duration of the data captured and all other related personal data used; - the security measures, including against unlawful access to the data; - other safeguards (when relevant);

- the description of any cooperation with private or public authorities, including in other countries, and data transfers and exchanges;

- the traceability of the process;

- the name of the responsible personals of the deployers;

For the issuance as to other formal elements - the possibility of a written procedure complemented with a hearing; - the grounds of refusal;

- the rights of persons for which a search is conducted, the rights of persons whose data is captured, and the possible rights of third parties; - the delays within which the authorities will have to take their decision; - any need for formal notifications upon granting/refusal of the authorisation; - sanctions for not complying with (formal and substantial) requirements; - the right to appeal an authorisation that has been denied;

For the exercise - the registration of the use of real-time RBI systems in a central register with a summary of the substantive elements;

- possible further reporting obligations;

- procedure for extending or modifying the authorisation.

10... Detailed national law on the supervision and the reporting relating to the authorisation

Article 70 AI Act obliges the Member States to ‘establish at least one notifying and at least one market surveillance authority.’ Article 74(8) AI Act provides that ‘Member States shall designate as market surveillance authorities for the purposes of this Regulation either the competent data protection supervisory authorities under Regulation (EU) 2016/679 or Directive (EU) 2016/680, or any other authority designated pursuant to the same conditions laid down in Articles 41 to 44 of Directive (EU) 2016/680.’

This comes on top of the designation of the authorising authority, which the Member State will have to establish before it can authorise the use of real-time RBI systems for any of the objectives listed in Article 5(1)(h)(i) to (iii) AI Act.

The national market surveillance authorities and the national data protection authorities of Member States that have been informed by deployers of the use of real-time RBI systems in publicly accessible spaces for law enforcement purposes (see Article 5(4)) must submit annual reports on such use to the Commission. These reports shall be made on the basis of a template provided by the Commission. This template will be established in due time.

Where the deployer is an EU Institution, body, or agency, the EDPS is obliged to inform the Commission accordingly on an annual basis of the real-time RBI systems used in publicly accessible spaces for law enforcement purposes.

Only the report of the national data protection authority will cover the period between 2 February 2025 and 2 August 2025, since the AI Act does not require Member States to appoint a national market surveillance authority before the latter date.

National market surveillance authorities and national data protection authorities are free to decide whether they wish to submit individual reports or a joint report per Member State.

The AI Act requires the Commission to publish annual reports of the uses of real-time RBI systems in publicly accessible spaces for law enforcement purposes in the Member states and by Union institutions, agencies and bodies, based on aggregated data. These reports will be based on the information notified by the national authorities pursuant to Article 5(6) AI Act.

The Commission’s annual report shall not contain sensitive operational data. Sensitive operational data means ‘operational data related to activities of prevention, detection, investigation or prosecution of criminal offences, the disclosure of which could jeopardise the integrity of criminal proceedings’. This could mean that specific details that reveal ongoing or past investigations, such as e.g., locations, camera’s used, shall not be published.

All other uses of RBI systems that are not covered by the prohibition of Article 5(1)(h) AI Act fall within the category of high-risk AI systems as defined by Article 6 and listed in point 1(a) of Annex III AI Act provided they fall within the scope of the AI Act.

RBI systems that fall outside the scope of the prohibition in Article 5(1)(h) AI Act include biometric verification/authentication systems and the retrospective use of (post- ) RBI systems in publicly accessible spaces for law enforcement purposes. For instance, police authorities might be authorised by national law to perform retrospective facial recognition to compare images of criminal suspects with recorded facial images in a criminal database. Another use that falls outside the scope of the prohibition is the use of real-time RBI systems for law enforcement purposes in either a private (such as at somebody’s place) or an online space (such as the use of a chat room or online game to identify a suspect of disseminating child sexual abuse material). Finally, the use of RBI systems by private actors, both in real-time and retrospective (such as the use of live facial recognition technology by a supermarket to identify known shoplifters, the use of live facial recognition technology by a sports arena to identify individuals banned from entering the arena, or the use of live facial recognition technology in schools for security purposes and school attendance) fall outside the scope of the prohibition.

In addition to the rules that apply to high-risk AI systems generally, the retrospective use of RBI systems for law enforcement purposes is subject to additional conditions and safeguards in accordance with Article 26(10) AI Act (in application as from 2 August 2026).

Uses for purposes other than for law enforcement must in any event comply with data protection rules. The cases below illustrate the interpretation of Article 9(2) GDPR in cases of such use and the exceptions to process biometric data. For example, - A French administrative Court found that the trial of live facial recognition technology in two public schools for access control and security purposes was neither necessary nor proportionate (under data protection rules). Alternative solutions that were less intrusive for the students were available, e.g., the use of badges. In addition, the conditions for explicit consent were not met. Therefore, consent could not be used as a valid legal basis to trial facial recognition technology in high schools. - A supermarket was not allowed to use live facial recognition technology to prevent shoplifting in the Netherlands. Without explicit consent from the customer or any legal basis allowing the processing for a substantial public interest (such as security purposes), the supermarket could not process biometric data and thus deploy facial recognition technology. - The use of live facial recognition technology at the entrance of a football club to identify supporters was prohibited in France and to ensure the safety of spectators was prohibited in Spain.