Direct answer
We are considering facial recognition or other biometrics. Is that allowed? (We take it from a vendor, Public authority or public service provider)
Article 27 covers bodies governed by public law and private parties providing public services, among others. If you deploy a high-risk system from Annex III, you assess the impact on fundamental rights beforehand and notify the market surveillance authority of the result.
The key obligation here is Article 50: transparency. That obligation applies today. There are 2 exceptions you have to assess yourself.
This could go the other way
- The direct-interaction disclosure is not required where this is obvious to a reasonably well-informed, observant and circumspect person, considering circumstances and context.
- Only Article 50(2) has a transition until 2 December 2026 for systems placed on the market before 2 August 2026.
First step: Implement the applicable disclosure, marking or label.
You describe: Biometric identification or categorisation of people, such as facial recognition for access or in public spaces. Likely role: body governed by public law or public service provider.
This applies now
- Article 50: transparencyApplicable
- Article 5: prohibited practicesApplicable
- Article 4: AI literacyApplicable
Coming up
- Article 26: obligations of deployers of high-risk AI systemsfrom 2 December 2027
- Annex III: high-risk AIfrom 2 December 2027
- Article 27: FRIAfrom 2 December 2027
Article 27 covers bodies governed by public law and private parties providing public services, among others. If you deploy a high-risk system from Annex III, you assess the impact on fundamental rights beforehand and notify the market surveillance authority of the result. That applies from 2 December 2027. If the system is already in use before that date, the duty applies only once it is significantly changed; for a system intended for use by public authorities, by 2 August 2030 at the latest (Article 111(2)). A GDPR data protection impact assessment does not replace that assessment; since the amending regulation you may carry relevant parts of it across.
Your first actions
- Implement the applicable disclosure, marking or label. First determine which paragraph of Article 50 applies, then implement the specific transparency measure.
- Screen every use case against Article 5 first. Before procurement, build or deployment, check whether the use case falls under a prohibited practice and stop or redesign early rather than after the fact.
- Assign human oversight and give those people a mandate. Name, per high-risk system, who exercises oversight, and ensure that person has the competence, training, authority and support to actually set the output aside.
Record this
- Test report per touchpoint: disclosure visible, timely and accessible
- Article 5 screening record
- Deployment dossier: logs, worker information and information to affected persons
biometrics and identification
Camera at the entrance: access control versus biometric categorisation
An organisation admits staff through facial recognition at its access control gate and additionally runs a camera in the visitor area that sorts faces into age groups. Both applications run on the same biometric infrastructure and the same images. The question is which of the two requires the people involved to be actively informed.
Provenance: The Commission guidelines of 20 July 2026 count automated facial-recognition access controls among the systems that merely collect data passively and are not capable of an exchange with natural persons, and therefore do not interact within the meaning of Article 50(1). For Article 50(3) they state that, unless the use is prohibited under Article 5(1)(g), the information duty applies to any biometric categorisation system, including outside the high-risk scope, and they give classification by age or gender on the basis of biometric data as an example. Article 50(3) itself carries a further exception for systems permitted by law to detect, prevent or investigate criminal offences. As a way of informing people the guidelines describe a visible notice at each possible entrance to an exhibition room stating that facial images are captured to assign visitors to an age group, provided at the latest at the moment of first exposure.
Our reading is that you should map this per processing purpose rather than per camera: the same lens that stays outside Article 50(1) at the access control gate moves inside Article 50(3) as soon as those images place people in a category. Record for each setup what happens to the capture and who the deployer is, because that decides whether a notice belongs at the entrance. The guidelines prescribe no fixed form, but they do fix the moment: the notice has to be there before someone walks into frame, and a line in the privacy statement rarely makes that moment, in our reading.
Commission Guidelines C(2026) 5054 final, 20.7.2026, point (30) and points (104) to (108)
biometrics and identification
Facial recognition at access control: the guard behind the camera counts too
An organisation secures the entrances to its buildings with facial recognition and uses that biometric access control to register visitors as well. When the system returns no match, a security officer reviews the camera images and decides personally whether someone may enter. The question is whose measures have to reach that officer: those of the supplier of the model, those of the department that deploys the system, or both.
Provenance: Article 4(1) provides that providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf. In doing so they must take into account technical knowledge, experience, education and training and the context the AI systems are to be used in, and consider the persons or groups of persons on whom the AI systems are to be used. The same provision states that this obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.
We read the phrase about the persons on whom the system is used as the centre of gravity for biometrics: whoever stands in front of the camera is subject to the outcome and has little to set against it. That argues for equipping the officer who decides for himself when no match comes back more substantively than the colleague who merely switches the system on and off. The article itself names no sufficient level and expressly states that you need not guarantee one, so where the floor lies for each role stays open. In our assessment a record kept per role, stating the choice made and the reason for it, is easier to defend than one organisation-wide session backed only by an attendance list.
Editorial example. The rule above is in the Regulation. The situation was written by us to show how that rule plays out in this sector, and is not taken from a worked case in official guidance.
Article 4(1)
biometrics and identification
Inferring political opinions from uploaded photos
A platform analyses the biometric data in photos users have uploaded to infer their assumed political orientation and serve them targeted political messages. A comparable system infers assumed sexual orientation in order to serve advertisements.
Provenance: The Commission guidelines on prohibited AI practices treat this case as a worked example under Article 5. The document is non-binding: authoritative interpretation rests with the Court of Justice.
Relying on the ancillary feature exception requires that the feature is also strictly necessary for objective technical reasons alongside the main service, since both conditions apply cumulatively and advertising purposes do not meet that bar.
Commission Guidelines C(2025) 5052 final, 29.7.2025, worked examples under Article 5
biometrics and identification
Face comparison at the border gate: verification or identification
An automated border gate uses biometric facial recognition to compare a traveller’s face with the photo in the passport chip. The same camera could technically also compare against a law-enforcement database, and exactly that difference decides whether this biometrics is high-risk.
Provenance: The Commission draft guidelines of 19 May 2026 state that biometric verification falls outside the high-risk classification: one-to-one comparison of presented biometrics with previously stored biometrics, for the sole purpose of confirming that a person is who they claim to be. Where the same capture is additionally compared against a law-enforcement database, it does become remote biometric identification. The document is a consultation version: non-binding and not yet final.
Test your biometric application on purpose rather than technology: the same camera and the same model stay outside the high-risk route as long as the comparison is one-to-one and only confirms identity, and fall inside it as soon as that same capture is also held against a database. Record per application what the comparison runs against, because that single design choice moves the entire regime.
Draft guidelines on high-risk AI classification, 19 May 2026, annex on Annex III, paragraph (136)
AI agents disclose that they are AI (Article 50(1)); the guidelines also ask them to say on whose behalf they act
Point (31) of the guidelines of 20 July 2026 states that AI agents are covered by Article 50(1) if they are capable of interacting with the persons instructing them or with other natural persons in the execution of their tasks, citing as examples making bookings, managing correspondence, negotiating or concluding contracts and executing purchases. That same point requires AI agents to be designed and developed so that they disclose both their artificial nature and the person on whose behalf they are acting, given the need for transparency of the origin and of the delegation of authority and accountability for the consequences of their actions. This also applies in complex multi-agent architectures in which other agents interact directly with natural persons. Where the provider cannot reliably determine before placing on the market or putting into service whether the agent will directly interact with a natural person, the agent should be designed at the architecture level and instructed to disclose itself in every situation where it is reasonably likely to interact with a natural person, including where that person represents a legal entity. Agents should also disclose themselves to the persons instructing them at key steps such as authorisation, reporting and validation, including where the agent receives, processes or relies upon outputs generated by other AI systems rather than by a natural person, and at every new interaction. Point (63) adds that Article 50(2) may apply to AI agents where the agent takes an action whose output is AI-generated or manipulated content perceptible by natural persons, while intermediate processing steps such as reasoning and chain of thought and non-perceptible actions such as a web request or browser action fall outside that scope.
Commission Guidelines C(2026) 5054 final, Section 3.1.1 point (31) and Section 4.1.2 point (63)
Artistic or satirical work is not exempt but attenuated, and the informative character always prevails
Point (119) of the guidelines of 20 July 2026 describes an attenuated transparency obligation for deep fakes forming part of evidently artistic, creative, satirical, fictional or analogous works or programmes, where the obligation is limited to disclosure in an appropriate manner that does not hamper the display or enjoyment of the work. Point (120) describes the categories: artistic works are created for the purpose of art, including music, cinematographic works and visual arts; creative works involve creative choices, while works mainly motivated by functional or technical considerations cannot be regarded as creative; satirical works are intended to criticise society, politics, business or public figures through humoristic techniques; fictional works involve persons, objects, places, entities or events in an imaginary but verisimilitude setting; analogous works share core traits with those categories without fitting neatly into one. Point (122) states that it must be evident to the natural persons exposed to it that the content falls within one of those categories, that the categories must therefore be interpreted strictly given the lighter disclosure regime and the interests of freedom of expression and freedom of the arts and sciences, and that content whose nature is potentially unclear or ambiguous to the audience falls outside this lighter regime. Relevant factors, per that same point, are whether the content displays formats or styles characteristic of the category, the context in which it is presented, and audience expectations. That same point excludes content whose nature is exclusively informative or commercial and recognisable as such, citing news reporting, notes that advertisements or documentaries may be regarded as evidently creative or fictional in certain specific situations but not in others because the assessment is case-specific, and states that where the deep fake combines multiple characters, for example informative and creative, the informative character should always prevail and the standard labelling requirements apply. Point (123) stresses that these deep fakes are not excluded from the obligation: the deployer must still disclose the AI origin or manipulation, but may do so in an appropriate manner, and must in any case comply with Article 50(5). Point (124) states that reliance on the attenuated obligation cannot justify failing to respect the fundamental rights of individuals or the rights of rightsholders under Union intellectual property or data protection law. As examples within the categories the document cites movies featuring AI de-aged existing actors or digital replicas of deceased actors, AI-generated music in the style of existing artists, and an AI-manipulated image of an existing politician in a scene clearly meant as humorous criticism. Outside the categories the document places among others an AI-manipulated video in the style of a teleshopping channel, AI-generated images of celebrities implying involvement in activities that never happened, and an AI-manipulated video featuring a realistic synthetic influencer focused solely on displaying a sponsored product's functionalities.
Commission Guidelines C(2026) 5054 final, Section 6.1.3, points (119) to (124) and the accompanying example lists
Signing the code of practice is voluntary, but not signing means proving it yourself
The Commission Q&A on signing the Code of Practice on Transparency of AI-generated Content states that signing is voluntary and that not signing does not constitute non-compliance with the AI Act. That same Q&A states that the code has two sections, one for providers on marking AI-generated or manipulated content in machine-readable formats and the related detection mechanisms, and one for deployers on the disclosure and labelling of deep fakes and of certain AI-generated or manipulated text on matters of public interest. It further states that signatories may, subject to a positive assessment by the Commission and the AI Board (given on 8 and 9 July 2026), rely on the code to demonstrate compliance with their obligations under Article 50 regardless of their place of establishment, and that the deadline for inclusion in the initial list of signatories was 27 July 2026 at 18:00 CEST, with later signature remaining possible but not appearing on that initial list. The Commission guidelines of 20 July 2026 state in point (146) that providers and deployers within the scope of Article 50(2) and (4) may demonstrate compliance by adhering to a code of practice assessed as adequate under Article 50(7), and that such a code does not replace the Regulation or the guidelines but complements them as the only Union-wide recognised practical framework for that purpose. Point (147) states that compliance may also be demonstrated through other adequate means, that for signatories supervisory activities will focus on whether they have adhered to the code and implemented the measures it contains, and that opting out of sections results in losing the benefit of facilitated demonstration of compliance for that part. Point (148) states that non-signatories are expected to demonstrate through other adequate means how they comply with Article 50(2), (4) and (5) and to explain how their measures ensure compliance, for instance by carrying out a gap analysis comparing their measures with those set out in a code assessed as adequate, and that they will likely face more requests for information and for access. Point (149) states that competent authorities may take commitments implemented in line with a code assessed as adequate into account as a mitigating factor when setting the amount of fines. Point (150) states that if a code is not deemed adequate, the Commission may adopt an implementing act specifying common rules for the implementation of Article 50(2), (4) and (5).
Commission Guidelines C(2026) 5054 final, Section 8.1, points (146) to (150); Commission Q&A on signing the Code of Practice on Transparency of AI-generated Content (consulted 9 August 2026)
The deployer cannot rely on the provider's machine-readable marking
Point (117) of the guidelines of 20 July 2026 states that deployers of AI systems generating or manipulating deep fake content must clearly and distinguishably disclose that the content has been artificially created or manipulated, by labelling the output accordingly and disclosing its artificial origin. According to that same point, the labelling or disclosure methods used must be understandable and perceivable by natural persons, for example with visible or audible labels, without those persons needing to rely on any specific technical tools or performing dedicated actions. The point closes expressly by stating that deployers cannot rely on the machine-readable marking embedded in the content by the provider under Article 50(2), since those markings are not immediately clear and distinguishable for the natural persons exposed to the deep fake content. Point (12) adds that deployers involved in complex content production and distribution value chains must take proportionate measures to ensure that the labelling they have implemented under Article 50(4) is actually displayed in a clear and distinguishable manner to the targeted and foreseeable audience at the point of first exposure, for example via contractual conditions with distributing partners and via user experience settings and interfaces. Point (14) states that a legal person remains the deployer even where it involves third parties such as contractors or freelancers in the operation of the system on its behalf and under its authority, and that individual employees acting under its instructions and control are not considered separate deployers.
Commission Guidelines C(2026) 5054 final, 20.7.2026, Section 6.1.2 point (117) and Section 2.3 points (12) and (14)
General interpretation, not legal advice. Checked against Regulation (EU) 2024/1689 and the Digital Omnibus (EU) 2026/1744; the official source remains authoritative.
Full map for your situation