Commission guidance on prohibited AI practices

This page gives the AI practices that Article 5 of the EU AI Act prohibits. The EU AI Act is Regulation (EU) 2024/1689, with the changes in Regulation (EU) 2026/1744. This page also gives the interpretation of Article 5 in the Commission Guidelines on prohibited AI practices. For the other Commission guidance, see Commission guidance on the EU AI Act.

Article 5 and the Commission guidance

Article 5 of the EU AI Act prohibits specified AI practices. The Commission Guidelines on prohibited AI practices are Communication C(2025) 5052 final of 29 July 2025. These non-binding Guidelines give the Commission interpretation of Article 5, as Article 96(1)(b) requires.

The 2025 Guidelines give an interpretation of only the eight initial prohibitions in Article 5(1), points (a) to (h). They also give an interpretation of the conditions in Article 5(2) to (7) for real-time remote biometric identification (RBI). These eight prohibitions apply from 2 February 2025 (Article 113, point (a)).

Regulation (EU) 2026/1744 added points (ba) and (bb) to Article 5(1). It also added paragraphs 1a and 1b to Article 5 (Article 1(7) of that Regulation). These new provisions apply from 2 December 2026 (Article 113, point (a)). The 2025 Guidelines do not give an interpretation of them. Thus, the eight prohibitions in the Guidelines are not all the prohibitions in Article 5 after the 2026 changes.

For points (ba) and (bb), see Article 5(1)(ba) and (bb) on this page. For all the application dates of the EU AI Act, see Application dates. For the fines for "non-compliance with the prohibition of the AI practices referred to in Article 5" (Article 99(3) and (6)), see Enforcement and penalties.

Conditions for Article 5

Each prohibition in Article 5(1) of the EU AI Act has conditions. A practice is prohibited only if all the conditions of one point apply to it.

Before you apply Article 5, make sure that the EU AI Act applies (Article 2). Also make sure that the system is an AI system (Article 3(1)). For these two steps, see Scope of the AI Act. For high-risk Article 6(1) systems related to Annex I Section B products, Article 2(2) limits the applicable provisions. Article 5 is not one of them. For this rule, see Annex I Section B scope.

• Actions covered — Points (a), (b), (ba), (bb) and (c) each have conditions for a prohibition. The prohibition covers "the placing on the market, the putting into service or the use" of an AI system when those conditions apply. Points (d), (e), (f) and (g) apply to the placing on the market, the putting into service "for this specific purpose", or the use. Point (h) applies only to the use (Guidelines, paragraph (11)). Paragraph 1a limits points (ba) and (bb). Paragraph 1b limits only point (ba).
• Providers and deployers — The Commission Guidelines apply the prohibitions to providers and deployers, "each within their respective responsibilities" (paragraphs (15) and (62)). Point (h) applies only to the use. Thus, point (h) applies only to deployers (paragraph (291)).
• Exceptions — Only points (bb), (d), (f), (g) and (h) contain an exception or an exclusion in their text. Points (a), (b), (ba), (c) and (e) contain no exception. If one condition of a point does not apply to a practice, that point does not apply. This result is not an exception.
• Objective or effect — Points (a) and (b) apply "with the objective, or the effect" of a material distortion of behavior. Thus, intent is not necessary if the condition of "the effect" applies (Guidelines, paragraphs (76), (83) and (113)). A plausible causal link between the practice, the distortion and the significant harm is always necessary (paragraphs (61) and (84)).
• Biometric systems — Points (e), (g) and (h) do not apply to all uses of biometric systems. Each of these points has different conditions. More than one point can apply to one system.

Article 5(1)(a) — Harmful manipulation, deception, subliminal techniques

Article 5(1)(a) of the EU AI Act prohibits the placing on the market, the putting into service or the use of an AI system. Point (a) applies only if all these conditions apply:

1. The system deploys subliminal techniques "beyond a person’s consciousness", or "purposefully manipulative or deceptive techniques".
2. The technique has "the objective, or the effect" of "materially distorting the behaviour of a person or a group of persons by appreciably impairing their ability to make an informed decision, thereby causing them to take a decision that they would not have otherwise taken".
3. The distortion "causes or is reasonably likely to cause that person, another person or group of persons significant harm".

In the Commission Guidelines, all these conditions must apply at the same time (paragraphs (60) and (61)). A plausible causal link between the technique, the distortion and the significant harm is also necessary. In paragraph (84), these effects "do not need to have fully materialised". The test is an "objective assessment" with "sufficient indication that they are likely or capable to materialise and undermine individual autonomy".

Only point (a) contains the condition about an "informed decision". Point (b) does not contain it (paragraph (113)).

• Subliminal techniques — In the Guidelines, subliminal techniques "operate beyond (below or above) the threshold of conscious awareness" (paragraph (64)). With these techniques, the person "remains unaware of such influence, how it works, or its effects" (paragraph (65)). Examples are "stimuli delivered through audio, visual, or tactile media that are too brief or subtle to be noticed" (paragraph (65)).
• Manipulative or deceptive techniques — Point (a) does not apply only because a system uses a manipulative or deceptive technique. The other conditions must also apply: the material distortion, and significant harm that occurred or that is "reasonably likely", with a plausible causal link. An example of a deceptive technique in the Guidelines is "an AI chatbot that impersonates a friend of a person or a relative with synthetic voice" (paragraph (73)).
• Techniques — Point (a) does not automatically apply to a technique used to make an interface or a recommender system. But such a technique is also not automatically out of point (a). Apply all the conditions of point (a), and also examine the alternative of subliminal techniques. Significant harm that occurred is not necessary, because significant harm that is "reasonably likely" is sufficient.
• Significant harm — The Guidelines give "physical, psychological, financial, and economic harms" as the "types of harms" (paragraph (86)). In paragraph (91), the decision about "significant harm" is "fact-specific" and is a "case-by-case assessment".
• Practices that are not prohibited — Point (a) "does not apply to lawful persuasion practices" (paragraph (127)). In paragraph (133), "common and legitimate commercial practices, such as advertising," are not harmful manipulation "in themselves".

Article 5(1)(b) — Exploitation of vulnerabilities

Article 5(1)(b) of the EU AI Act prohibits the placing on the market, the putting into service or the use of an AI system. Point (b) applies only if all these conditions apply:

1. The system "exploits any of the vulnerabilities of a natural person or a specific group of persons due to their age, disability or a specific social or economic situation".
2. The exploitation has "the objective, or the effect, of materially distorting the behaviour of that person or a person belonging to that group".
3. The distortion "causes or is reasonably likely to cause that person or another person significant harm".

• Three types of vulnerability — Point (b) applies only to these three types of vulnerability (Commission Guidelines, paragraph (103)). Point (a) can apply to the exploitation of other vulnerabilities if all the conditions of point (a) apply.
• Specific social or economic situation — In paragraph (109), "specific" is "a legal status or membership to a specific vulnerable social or economic group". Examples are "persons living in extreme poverty and ethnic or religious minorities". Also, "transient circumstances, such as temporary unemployment, over-indebtedness or migration status, may also be covered". Point (b) does not apply to "grievances or loneliness", but point (a) can apply to their exploitation (paragraph (109)).
• Exploitation — In paragraph (103), exploitation is "objectively making use of such vulnerabilities in a manner that is harmful". An AI system that is "inadvertently biased and disproportionately impacts socially disadvantaged persons" is not automatically exploitation, because these persons "are not specifically targeted" (paragraph (110)).
• When a provider or deployer knows — In paragraph (110), a provider or deployer also exploits these vulnerabilities if all these conditions apply:
  • It is "aware that their systems unlawfully discriminate against persons or groups of persons".
  • It is "aware of the reasonably likely significant harm".
  • It has "not taken appropriate corrective measures".
• Example of financial products — The Guidelines give the example of an algorithm that sends "advertisements for predatory financial products" to persons "in a dire financial situation", and that causes them "significant financial harm" (paragraph (110)). For example, an AI system can send information about debt restructuring to persons in "a dire financial situation". Point (b) applies to it only if all three conditions apply.
• Example of credit decisions — An AI system for credit decisions can give a credit score to persons in different social or economic conditions. Point (b) does not apply only because of this fact. Examine the exploitation, the material distortion and the significant harm. In paragraph (133), AI systems for "banking services" that use "the age or the specific socio-economic situation of the client as an input" are not exploitation. This applies if they obey Union law, "are designed to protect and support people identified as vulnerable" and "are beneficial for those groups".

An AI system "intended to be used to evaluate the creditworthiness of natural persons" can be high-risk in Annex III, point 5(b). Chapter III, Sections 1 to 3 apply to these high-risk AI systems from 2 December 2027 (Article 113, point (c)). For systems that are on the market or put into service before that date, Article 111(2) also applies. This date does not change the date of Article 5. For the high-risk examples, see the Annex III(5) examples.

Article 5(1)(ba) and (bb) — Intimate and child sexual abuse material

Regulation (EU) 2026/1744 added two prohibitions to Article 5(1) of the EU AI Act. These two prohibitions apply from 2 December 2026 (Article 113, point (a)). The 2025 Commission Guidelines do not give an interpretation of them.

• Article 5(1)(ba) — Point (ba) prohibits the placing on the market, the putting into service or the use of an AI system that "generates or manipulates realistic images, videos, audio or similar material of an identifiable natural person’s intimate parts, or of an identifiable natural person engaged in sexually explicit activities". The prohibition applies if this occurs "without that person’s freely-given, specific, informed, unambiguous and explicit consent for that generation or manipulation".
• Article 5(1)(bb) — Point (bb) prohibits the placing on the market, the putting into service or the use of an AI system that "generates or manipulates material or performance within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU". Point (bb) does not apply "where a ‘without right’ defence applies under national law".

Paragraph 1a of Article 5 gives limits to points (ba) and (bb). For the placing on the market and the putting into service, the prohibition applies only in one of these two conditions:

(i) that generation or manipulation is the intended purpose of the AI system; or

(ii) the system’s design, training, architecture, capabilities or user-facing functionalities make that generation or manipulation a reasonably foreseeable and reproducible outcome, without requiring significant technical modification, and the system does not have reasonable and adequate technical safety measures and other safeguards to reliably prevent that generation or manipulation, taking into account reasonably foreseeable misuse, and to correct observed or reported misuse;

Condition (ii) has two parts, and the two parts must apply. The outcome is "reasonably foreseeable and reproducible", and the system does not have the safety measures and other safeguards. For the use of such an AI system, the prohibition applies only "where the deployer uses the system for the purpose of generating or manipulating such material or performance" (paragraph 1a, point (b)).

Paragraph 1b applies only to point (ba). In paragraph 1b, an AI system "that manipulates material in a way that does not increase the exposure of any depicted intimate parts or alter the nature of any depicted sexually explicit activities shall not constitute manipulation".

Article 5(1)(c) — Social scoring

Article 5(1)(c) of the EU AI Act prohibits the placing on the market, the putting into service or the use of AI systems. Point (c) applies only if all these conditions apply:

1. The AI systems are "for the evaluation or classification of natural persons or groups of persons over a certain period of time based on their social behaviour or known, inferred or predicted personal or personality characteristics".
2. The social score causes "either or both of the following":
   • "detrimental or unfavourable treatment of certain natural persons or groups of persons in social contexts that are unrelated to the contexts in which the data was originally generated or collected" (point (c)(i))
   • "detrimental or unfavourable treatment of certain natural persons or groups of persons that is unjustified or disproportionate to their social behaviour or its gravity" (point (c)(ii))

• Two alternatives — In the Commission Guidelines, points (c)(i) and (c)(ii) "are alternative and may apply also in combination" (paragraphs (163), (164) and (168)).
• Same social context — Point (c)(ii) can also apply when the social score and the treatment are in the same social context. For example, a social score for the performance of personnel is prohibited if the treatment is "unjustified or disproportionate" and the other conditions apply. Thus, such a social score is not automatically out of point (c).
• Persons who supply or use the system — Point (c) applies "regardless of whether the AI system or the score are provided or used by public or private persons" (paragraph (170)).
• Period of time — Paragraph (155) tells you that "the evaluation or classification is based on data that spans over ‘a certain period of time’". It also tells you that "the assessment should not be limited to a one-time or an at once rating or grading with data or behaviour from a very specific individual context". Examine this condition with "all circumstances of the case", to prevent "circumvention of the scope of the prohibition" (paragraph (155)). The Guidelines give no minimum period.
• Uses that are not prohibited — In paragraph (175), "the scoring of natural persons is not at all times prohibited". In paragraph (176), AI systems that give a social score "in a lawful manner and for a specific purpose in the related context" are not prohibited, if the treatment is "justified and proportionate to the gravity of the social behaviour". Examples in paragraph (177) are "financial credit scoring systems" that use "the customer’s income and expenses and other financial and economic circumstances", and systems that "evaluate customers for financial fraud" with related data. These examples apply only if the systems also obey the applicable law.

Article 5(1)(d) — Criminal offence risk prediction

Article 5(1)(d) of the EU AI Act prohibits the placing on the market, the putting into service "for this specific purpose", or the use of an AI system. Point (d) applies only if all these conditions apply:

1. The AI system is "for making risk assessments of natural persons in order to assess or predict the risk of a natural person committing a criminal offence".
2. The risk assessment is "based solely on the profiling of a natural person or on assessing their personality traits and characteristics".

Point (d) does not apply to "AI systems used to support the human assessment of the involvement of a person in a criminal activity, which is already based on objective and verifiable facts directly linked to a criminal activity".

• Time of the criminal offense — In the Commission Guidelines, these risk assessments "concern future criminal offences (not yet committed) or crimes that are assessed as a risk of being committed at the moment, including in cases of an attempt or preparatory activities undertaken to commit a criminal offence" (paragraph (191)). "They can be made at any stage of the law enforcement activities", and "also during the investigation, prosecution and execution of criminal penalties" (paragraph (191)). The identification of the person who committed a previous criminal offense is a different task.
• "Solely" — In paragraph (199), "‘solely’ is intended to apply both to profiling or to the assessment of personality traits and characteristics". For "profiling", Article 3(52) refers to Article 4, point (4), of Regulation (EU) 2016/679. Other elements can cause the risk assessment to be not "based solely on the profiling". They have this effect only if they are "real, substantial and meaningful" (paragraph (202)). This is a "case-by-case" decision, and facts in a risk assessment do not automatically have this effect.
• Human assessment — For the exclusion, AI predictions or risk assessments can help, but must not replace, the required human assessment (paragraphs (204), (206) and (214)). The human assessment must "already be based on objective and verifiable facts directly linked to a criminal activity" (paragraph (206)). It is not sufficient to add some facts and a person who examines the result to a risk assessment that uses profiling.
• Criminal offense — In paragraph (218), "criminal offence" is "a concept that has autonomous meaning within the EU law". Also, "the classification of the offences by the Member States is not conclusive". "Relevant criteria used to assess the nature of the offence (criminal or not) may be found in relevant case-law of the CJEU and the European Court of Human Rights" (paragraph (218)).

A system for "fraud prevention and detection", for "anti-money laundering" or for sanctions screening is not automatically out of point (d). The Guidelines give these results:

• Point (d) does not apply if "a private entity profiles customers for its regular business operations and safety or to protect its financial interests". This applies only without the purpose "of assessing or predicting the risk of the customer committing a specific criminal offence" (paragraphs (210) and (211)).
• In paragraph (209), point (d) can apply to "private entities" with a legal obligation "to assess or predict the risk of persons committing specific criminal offences". An example is anti-money laundering law. In paragraph (209), "a banking institution" that uses AI for this obligation is not in point (d) if all these conditions apply:
  • It uses only "the data as specified in that law which are objective and verifiable".
  • Its predictions are "subject to human assessment and verification".
• Point (d) can also apply when "private actors are entrusted by law to exercise public authority and public powers" for law enforcement, or when they work "on behalf of law enforcement authorities" (paragraph (208)).
• An AI risk assessment for "pre-trial detention" is prohibited if it is "based solely on the profiling of a natural person or on assessing their personality traits". The exclusion applies only if the AI system helps a human assessment that is "already based on objective and verifiable facts" (paragraph (214)).

If the exclusion applies to a system for use by law enforcement authorities or "on their behalf", the system is high-risk in Annex III, point 6(d) (paragraph (204)). Article 6(3) always makes high-risk an AI system in Annex III that "performs profiling of natural persons". Examine the high-risk classification and its application dates independently. For this classification, see Commission draft guidance on high-risk classification.

Article 5(1)(e) — Untargeted scraping of facial images

Article 5(1)(e) of the EU AI Act prohibits the placing on the market, the putting into service "for this specific purpose", or the use of "AI systems that create or expand facial recognition databases through the untargeted scraping of facial images from the internet or CCTV footage". The Commission Guidelines give four conditions, and all four must apply (paragraph (224)):

1. The practice is the placing on the market, the putting into service "for this specific purpose", or the use of an AI system.
2. The purpose is to "create or expand" a facial recognition database.
3. The database gets its images "through AI tools for untargeted scraping".
4. The images come from the internet or from CCTV footage.

• Database — In paragraph (226), it is not necessary that "the sole purpose of the database is to be used for facial recognition". It is sufficient that "the database can be used for facial recognition".
• Untargeted — "Untargeted" is "without a specific focus on a given individual or group of individuals" (paragraph (228)). Scraping for the images "only of specific individuals or a pre-defined group of persons" is "targeted", and point (e) does not apply to it (paragraph (229)). But point (e) applies to scraping in steps "where the end-result is functionally the same as pursuing untargeted scraping from the outset" (paragraph (230)). "Where systems combine targeted searches for images or videos with untargeted searches, the untargeted scraping is prohibited" (paragraph (231)).
• Consent — Point (e) has no consent exception. In paragraph (232), "the fact that a person has published facial images of themselves on a social media platform does not mean that that person has given his or her consent for those images to be included in a facial recognition database".
• Research — Research is out of the scope of the EU AI Act only if Article 2(6) or Article 2(8) applies. For these provisions, see Scope of the AI Act.

The Guidelines give these examples that are not in point (e) (paragraphs (234) and (235)):

• untargeted scraping of "biometric data other than facial images (such as voice samples)"
• scraping "where no AI systems are involved"
• "facial image databases used for AI model training or testing purposes, where the persons are not identified"
• AI systems that collect facial images "to build AI models that generate new images about fictitious persons", only if the systems do not result in "the recognition of real persons"

Other provisions of the EU AI Act and other Union law can apply to these examples. Examine each use. Paragraph (236) applies to "existing facial databases built up prior to the entry into application of the prohibition, which are not further expanded through AI-enabled untargeted scraping". These databases and their use "must comply with the applicable Union data protection rules".

Article 5(1)(f) — Emotion recognition in workplace and education

Article 5(1)(f) of the EU AI Act prohibits the placing on the market, the putting into service "for this specific purpose", or the use of "AI systems to infer emotions of a natural person in the areas of workplace and education institutions". The exception is "where the use of the AI system is intended to be put in place or into the market for medical or safety reasons".

The Commission Guidelines apply point (f) only to "inferences based on a person’s biometric data" (paragraph (245)). This follows the definition of an emotion recognition system in Article 3(39). The Guidelines give these results:

• An AI system "inferring emotions from written text (content/sentiment analyses)" does not use biometric data, and point (f) does not apply (paragraph (251)).
• An AI system that infers emotions "from key stroke (way of typing), facial expressions, body postures or movements" uses biometric data (paragraph (251)). Point (f) can also apply to an AI system that infers emotions from "voice and imagery of hybrid video calls" (paragraph (254)).
• "Emotions or intentions" do not include "physical states, such as pain or fatigue, including, for example, systems used in detecting the state of fatigue of professional pilots or drivers for the purpose of preventing accidents" (paragraph (249)).
• Point (f) does not apply to "the mere detection of readily apparent expressions, gestures or movements, unless they are used for identifying or inferring emotions" (paragraph (249)).

In paragraph (254), the Guidelines give a wide interpretation of "workplace". The workplace is "any specific physical or virtual space where natural persons engage in tasks and responsibilities assigned by their employer or by the organisation they are affiliated to". This is "independent from the status as an employee, contractor, trainee, volunteer". Point (f) also applies to candidates "during the selection and hiring process", and "during the probationary period" (paragraph (254)). In paragraph (255), education institutions include "both public and private institutions" at all levels. Point (f) also applies "to candidates during the admissibility process".

The persons that the system examines are important for the result, not only the location. The Guidelines give these examples (paragraph (254)):

• "Using webcams and voice recognition systems by a call centre to track their employee’s emotions, such as anger, is prohibited."
• A use only "for personal training purposes" is permitted if all these conditions apply:
  • The results are "not shared with HR responsible persons".
  • The results "cannot impact the assessment, promotion etc. of the person trained".
  • The prohibition "is not circumvented".
  • The use "does not have any impact on the work relationship".
• "Using voice recognition systems by a call centre to track their customers emotions, such as anger or impatience, is not prohibited by Article 5(1)(f) AI Act".

Some systems monitor persons in a location where personnel also work, for example "the security staff at a football stadium". If the primary objective of such a system is not to examine personnel, the Guidelines put it out of point (f) (paragraph (270)). In paragraph (270), "deployers must employ safeguards to avoid the screening of employees". Deployers also "remain responsible to ensure that employees are not adversely affected by their use".

Point (f) does not apply out of the workplace and education institutions. Points (a) or (b) can apply to such systems (paragraph (266)).

Exception for medical or safety purposes — The text of the EU AI Act is "for medical or safety reasons". The Commission gives a narrow interpretation of this exception (paragraph (256)):

• "Therapeutic uses should be understood to mean uses of CE-marked medical devices" (paragraph (257)).
• The exception "does not comprise the use of emotion recognition systems to detect general aspects of wellbeing" (paragraph (257)). Also, "the general monitoring of stress levels at the workplace is not permitted under health or safety aspects".
• A system "intended to detect burnout or depression at the workplace or in education institutions" stays prohibited (paragraph (257)).
• "Safety reasons" apply "only in relation to the protection of life and health and not to protect other interests, for example property against theft or fraud" (paragraph (258)).
• In the interpretation of the Commission, each use "should always remain limited to what is strictly necessary and proportionate", with limits "in time, personal application and scale" and with "sufficient safeguards" (paragraph (259)). "Data collected and processed in this context may not be used for any other purpose" (paragraph (260)).

Other law, for example data protection law, continues to apply to these systems (paragraph (262)). An emotion recognition system not prohibited by point (f) can be high-risk in Annex III, point 1(c). It can also have Article 50(3) obligations. For these obligations, see Commission guidance on transparency.

Article 5(1)(g) — Biometric categorization

Article 5(1)(g) of the EU AI Act prohibits the placing on the market, the putting into service "for this specific purpose", or the use of "biometric categorisation systems that categorise individually natural persons based on their biometric data to deduce or infer their race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation".

Point (g) "does not cover any labelling or filtering of lawfully acquired biometric datasets, such as images, based on biometric data or categorizing of biometric data in the area of law enforcement". The Commission Guidelines put labeling or filtering of lawfully acquired biometric datasets out of point (g), also in law enforcement (paragraphs (284) and (286)).

• Conditions — The Commission Guidelines give five conditions, and all five must apply (paragraph (273)):

  • the placing on the market, the putting into service or the use
  • a biometric categorisation system
  • "individual persons must be categorised"
  • biometric data
  • the purpose "to deduce or infer" one of the characteristics of point (g)

  The characteristics of point (g) are "a limited number of sensitive characteristics" (paragraph (283)).

• Categorization — In paragraph (276), the categorization "is not about identifying an individual or verifying their identity, but about assigning an individual to a certain category". Thus, a system that compares a face with a database, and does no categorization by the characteristics of point (g), does a different task.

• Definition in Article 3(40) — The definition in Article 3(40) does not include a system that is "ancillary to another commercial service and strictly necessary for objective technical reasons". In paragraph (279), these two conditions "must be cumulatively fulfilled".

• High-risk — A biometric categorization not prohibited by point (g) can be high-risk in Annex III, point 1(b) (paragraph (287)).

A system for border control can compare persons with a "reference database" without a categorization by the characteristics of point (g). Point (g) does not apply to such a system. Examine the system with these provisions:

• Annex III, point 1(a), for remote biometric identification systems. Point 1(a) "shall not include AI systems intended to be used for biometric verification the sole purpose of which is to confirm that a specific natural person is the person he or she claims to be".
• Annex III, point 7(d), for AI systems in "migration, asylum or border control management, for the purpose of detecting, recognising or identifying natural persons, with the exception of the verification of travel documents".
• Article 5(1)(h), only for real-time use for law enforcement in a publicly accessible space. In paragraph (316), "a border crossing point is not a publicly accessible space, while the street leading to the border crossing point or a forest in the vicinity normally is".

Annex III, point 7(a), applies to "polygraphs or similar tools". It does not apply to a system that compares persons with a database.

Article 5(1)(h) — Real-time remote biometric identification

Article 5(1)(h) of the EU AI Act prohibits "the use of ‘real-time’ remote biometric identification systems in publicly accessible spaces for the purposes of law enforcement". The prohibition applies "unless and in so far as such use is strictly necessary for one of the following objectives":

(i) the targeted search for specific victims of abduction, trafficking in human beings or sexual exploitation of human beings, as well as the search for missing persons;

(ii) the prevention of a specific, substantial and imminent threat to the life or physical safety of natural persons or a genuine and present or genuine and foreseeable threat of a terrorist attack;

(iii) the localisation or identification of a person suspected of having committed a criminal offence, for the purpose of conducting a criminal investigation or prosecution or executing a criminal penalty for offences referred to in Annex II and punishable in the Member State concerned by a custodial sentence or a detention order for a maximum period of at least four years.

A use for one of these objectives must also obey all the conditions in Article 5(2) to (7). For these conditions, see Article 5(2)–(7) on this page.

Point (h) prohibits only the use. Thus, point (h) applies only to deployers (Commission Guidelines, paragraph (291)). In paragraph (291), "the placing on the market and the putting into service of such systems, as well as the use of other RBI systems, is not prohibited, but subject to the rules for high-risk AI systems". The Guidelines give five conditions, and all five must apply (paragraph (295)). These conditions are an RBI system, its use, real-time use, publicly accessible spaces, and law enforcement purposes.

• Remote — A remote biometric identification system identifies natural persons "without their active involvement" (Article 3(41)). In paragraph (303), AI systems "intended to be used for biometric verification fall outside the scope of the prohibition". In paragraph (305), the use of biometric systems "to confirm the identity of a natural person for the sole purpose of having access to a service, unlocking a device, or having security access to premises is excluded from the concept of ‘remote’". Thus, this type of active biometric verification at an entrance is not remote, and Annex III, point 1(a) does not include it. A system that identifies all persons at an entrance against a database, "without their active involvement", is different. Examine it with Annex III, point 1(a), and, for law enforcement use, with point (h).
• Real-time — In a real-time system, "the capturing of biometric data, the comparison and the identification all occur without a significant delay, comprising not only instant identification, but also limited short delays in order to avoid circumvention" (Article 3(42)). In paragraph (310), "a delay is significant at least when the person is likely to have left the place where the biometric data was taken". Also, "it will have to be assessed on a case-by-case basis". A system is not a "post-remote biometric identification system" only because there is an interval between the steps. In paragraph (311), a system that "screens all incoming visitors to a concert venue" is real-time RBI.
• Publicly accessible space — In Article 3(44), a publicly accessible space is "any publicly or privately owned physical place accessible to an undetermined number of natural persons, regardless of whether certain conditions for access may apply, and regardless of the potential capacity restrictions".

For publicly accessible spaces, the Guidelines give these results (paragraphs (314) to (318)):

• Examples are spaces "for commerce, such as shops, restaurants, cafés", spaces "for services, such as banks", "a doctor’s office as well as an accountant’s office", and "public roads and squares".
• A workplace "with entry control or limitations to relevant employees or service providers" is not a publicly accessible space. A workplace without entry controls can be a publicly accessible space (paragraph (315)).
• "Online spaces" and "prisons" are not publicly accessible spaces (paragraph (315)). For border control, see the Article 5(1)(g) section on this page.
• There is no general exclusion for a type of location. "Assessing whether a space is accessible to the public should be done based on a case-by-case analysis" (paragraph (318)).

Article 5(2)–(7) — Real-time RBI authorization

Article 5(2) to (7) of the EU AI Act give more conditions for a real-time RBI use for the objectives in point (h). Each use must obey all these conditions.

National law (Article 5(5)) — A Member State "may decide to provide for the possibility to fully or partially authorise" such use "within the limits and under the conditions listed in paragraph 1, first subparagraph, point (h), and paragraphs 2 and 3".

• That Member State must "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 also give the objectives, and "which of the criminal offences referred to in point (h)(iii)", for which "the competent authorities may be authorised to use those systems".
• The Member State must 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".

In the Commission Guidelines, without such national law, "law enforcement authorities and entities acting on their behalf may not deploy such systems for law enforcement purposes" (paragraphs (290) and (326)).

Purpose and elements (Article 5(2)) — The use is only "to confirm the identity of the specifically targeted individual". The use "shall take into account the following elements":

• "the nature of the situation giving rise to the possible use, in particular the seriousness, probability and scale of the harm that would be caused if the system were not used"
• "the consequences of the use of the system for the rights and freedoms of all persons concerned, in particular the seriousness, probability and scale of those consequences"

The use must also obey "necessary and proportionate safeguards and conditions in relation to the use in accordance with the national law authorising the use thereof, in particular as regards the temporal, geographic and personal limitations".

Impact assessment and registration (Article 5(2)) — The use is permitted only if the law enforcement authority "has completed a fundamental rights impact assessment as provided for in Article 27 and has registered the system in the EU database according to Article 49". In "duly justified cases of urgency", the use can start without the registration, if the registration "is completed without undue delay".

Prior authorization (Article 5(3)) — Each use must have "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".

• The authority gives the authorization only "on the basis of objective evidence or clear indications presented to it". These must show that the use is "necessary for, and proportionate to, achieving one of the objectives" of point (h). The use must also stay "limited to what is strictly necessary concerning the period of time as well as the geographic and personal scope". The authority also examines the elements of paragraph 2.
• In "a duly justified situation of urgency", the use can start without an authorization if the authorization "is requested without undue delay, at the latest within 24 hours". This time limit applies to the date on which the authorization is "requested", not to the date of the decision.
• If the authority rejects the authorization, "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".
• "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."

Notification (Article 5(4)) — "Each use of a ‘real-time’ remote biometric identification system in publicly accessible spaces for law enforcement purposes shall be notified to the relevant market surveillance authority and the national data protection authority in accordance with the national rules referred to in paragraph 5." The notification contains at a minimum the information of paragraph 6, and no "sensitive operational data".

Annual reports (Article 5(6) and (7)) — In Article 5(6), "national market surveillance authorities and the national data protection authorities of Member States that have been notified of the use of ‘real-time’ remote biometric identification systems in publicly accessible spaces for law enforcement purposes pursuant to paragraph 4 shall submit to the Commission annual reports on such use". The Commission gives them a template. In Article 5(7), the Commission "shall publish annual reports" that are "based on aggregated data in Member States", without "sensitive operational data". These annual reports are not a condition for each use.

A use that obeys all these conditions is also a use of a high-risk AI system (Commission Guidelines, paragraph (407)). Point (h) does not apply to "the retrospective use of RBI systems for law enforcement purposes". Article 26(10) gives different safeguards and a different time limit for that use (paragraph (292)). For these safeguards, see the deployer obligations in Article 26.

Article 50 transparency

Article 5 and Article 50 of the EU AI Act apply independently.

• Recital 137 of the EU AI Act tells you that compliance with the transparency obligations "should not be interpreted as indicating that the use of the AI system or its output is lawful under this Regulation or other Union and Member State law". Thus, transparency does not make a prohibited practice permitted.
• A prohibition in Article 5 does not remove Article 50 obligations that apply to a system. Examine each provision and its exceptions independently.
• If Article 5 does not apply to a practice, this does not show that Article 50 applies. Each Article 50 obligation has different conditions and exceptions. For example, the Article 50(2) obligation does not apply "to the extent the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof, or where authorised by law to detect, prevent, investigate or prosecute criminal offences".

Article 50(1) and (2) give obligations to providers, and Article 50(3) and (4) give obligations to deployers. For these obligations, see Providers, deployers and scope.

Use Modulos

In Modulos, the EU AI Act has two Frameworks: OFF-1 at Organization level and MFF-1 at AI System level. For the Requirements and the steps that Modulos recommends, see Operationalizing the EU AI Act in Modulos.

Sources

• Commission Guidelines on prohibited AI practices — European Commission, "Commission Guidelines on prohibited artificial intelligence practices established by Regulation (EU) 2024/1689 (AI Act)", Communication C(2025) 5052 final of 29 July 2025. Paragraphs (3), (5), (11), (15), (60) to (133), (155) to (177), (191) to (218), (224) to (236), (245) to (287), (290) to (326), and (407). Communication C(2025) 5052 final.
• EU AI Act — Regulation (EU) 2024/1689, consolidated text of 27 July 2026. Articles 2, 3, 5, 6, 26, 27, 49, 50, 96, 99, 111 and 113, Annex II and Annex III. Consolidated text. Recital 137 of the Regulation as published on 12 July 2024. Regulation (EU) 2024/1689.
• Digital Omnibus on AI — Regulation (EU) 2026/1744 of 8 July 2026, Article 1(7). Regulation (EU) 2026/1744.

Disclaimer

This page gives general information. It is not legal advice.