Real-Time Remote Biometric Identification: the EU AI Act's Narrow Exceptions

Article 5(1)(h) is usually reported either as a ban on live facial recognition or as a loophole that permits it. It is neither. It is a prohibition with three enumerated exceptions, each of which is unavailable until a member state has legislated to open it and a judicial authority has authorised the individual use.

The shape of the provision

Article 5(1)(h) of Regulation (EU) 2024/1689 prohibits the use of real-time remote biometric identification systems in publicly accessible spaces for the purposes of law enforcement, unless and in so far as such use is strictly necessary for one of three listed objectives.

Four qualifiers define what is covered. Real-time: Article 3(42) means identification without significant delay, comprising instantaneous identification and limited short delays to avoid circumvention. Remote: Article 3(41) describes identification of persons without their active involvement, typically at a distance, by comparison against a reference database. Publicly accessible spaces: Article 3(44) covers physical places accessible to an undetermined number of natural persons, regardless of ownership conditions of access. For the purposes of law enforcement: the prohibition does not reach private-sector deployment.

That last qualifier is the one most often missed. A shopping centre running live facial recognition against a watchlist of its own is not prohibited by Article 5(1)(h), because it is not law enforcement. It is high-risk under Annex III point 1(a), and it faces a hard problem under GDPR Article 9 that may well make it unlawful anyway — but the Article 5 prohibition is not the reason.

The three permitted objectives

• (i) Targeted search for specific victims. Victims of abduction, trafficking in human beings or sexual exploitation of human beings, and the search for missing persons. “Targeted” and “specific” both appear: general scanning for anyone on a missing persons register is a poor fit for the wording.
• (ii) 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. Note the two different standards inside one limb: the general safety threat must be specific, substantial and imminent, while the terrorist-attack branch reaches a foreseeable threat.
• (iii) Localisation or identification of a suspect. For the purpose of conducting a criminal investigation or prosecution, or executing a criminal penalty, for offences listed in Annex II and punishable in the member state concerned by a custodial sentence or detention order for a maximum period of at least four years.

Limb (iii) is the only one anchored to a list. Annex II enumerates offences including terrorism, trafficking in human beings, sexual exploitation of children, illicit trafficking in narcotic drugs, murder, grievous bodily injury, kidnapping, rape, organised or armed robbery, environmental crime, and participation in a criminal organisation involved in those offences. Both conditions must hold: the offence must be on the Annex II list and carry a maximum of at least four years in that member state, which means the availability of limb (iii) varies by country for the same conduct.

Six conditions before a single use

The exceptions are not self-executing. Article 5(2) to (7) attach conditions, and they are cumulative.

• Strict necessity and a weighing exercise. Article 5(2) requires account to be taken of the nature of the situation — in particular the seriousness, probability and scale of the harm that would be caused without the use — and of the consequences for the rights and freedoms of all persons concerned. Note “all persons concerned”: the bystanders scanned, not only the target.
• Confirmation of a targeted individual only. Article 5(2) limits use under limb (iii) to confirming the identity of the specifically targeted individual.
• Prior authorisation. Article 5(3) requires prior authorisation from a judicial authority, or an independent administrative authority whose decision is binding, in the member state where the use takes place, on a reasoned request. In a duly justified situation of urgency use may begin without it, provided the authorisation is requested without undue delay and at the latest within 24 hours; if refused, use stops immediately and all data, results and outputs are discarded and deleted.
• A fundamental rights impact assessment. Article 5(3) requires one under Article 27 before use — see what the FRIA has to contain.
• Registration in the EU database. Also required by Article 5(3), under Article 49, with a narrow urgency allowance to complete it without undue delay.
• Notification and reporting. Article 5(4) requires notification of each use to the relevant market surveillance authority and the national data protection authority, excluding sensitive operational data; Article 5(6) requires those authorities to report annually to the Commission; Article 5(7) requires the Commission to publish annual reports.

The reporting chain in the last item is worth noting for a reason that has nothing to do with compliance: it is the mechanism by which anyone outside a police force will eventually learn how often this is happening. Those annual reports are the primary source to watch.

Nothing is available without national law

Article 5(5) is the provision that decides whether any of this is available at all. A member state may decide to provide for the possibility of authorising use, in whole or in part. If it does, it must lay down in national law the detailed rules for requesting, issuing and exercising authorisations, and for supervision and reporting, and must specify which of the Article 5(1)(h) objectives — including which Annex II offences under limb (iii) — competent authorities may be authorised to use the systems for. Member states notify the Commission of those rules at the latest 30 days after adoption.

Two consequences follow. First, a member state that legislates nothing has an absolute prohibition on this conduct within its territory, because the exception route requires a national authorisation regime that does not exist. Second, the permitted scope differs by member state by design: one may open all three objectives, another only limb (i). There is no EU-wide answer to “is live facial recognition allowed for police”, and any source that gives one is wrong at the level of the question.

Post-remote identification is a separate regime

Everything above concerns real-time use. Post-remote biometric identification — Article 3(43) defines it as any remote biometric identification other than real-time — is not prohibited. It is high-risk under Annex III point 1(a), and Article 26(10) attaches its own conditions for law enforcement use: authorisation by a judicial or independent administrative authority, requested without undue delay and at the latest within 48 hours where the identification is needed to begin an investigation, with use in connection with a criminal offence and confined to what is necessary for the investigation of a specific offence.

The gap between the two regimes is the entire practical significance of the word “real-time”. Running the same recognition against recorded footage some hours later moves the operation from a prohibition-with-exceptions into a high-risk system with a 48-hour authorisation rule. Whether a short deliberate delay could be used to convert one into the other is exactly what Article 3(42)’s reference to limited short delays “in order to avoid circumvention” is drafted against. How long a delay defeats that anti-circumvention wording is not defined, and is unresolved. For the wider category see the Annex III law enforcement classification, and for the collection-side prohibition see the ban on untargeted facial scraping.