Annex III point 1 lists biometric use cases that are high-risk under Article 6(2), subject to the Article 6(3) filter: remote biometric identification systems, other than verification systems whose sole purpose is to confirm whether a person is who they claim to be; biometric categorisation systems that categorise natural persons based on biometric data according to sensitive or protected attributes or characteristics; and emotion recognition systems. Each use case must be matched to the intended purpose and to the Article 5 screen first, because several biometric practices are prohibited outright, and a biometric use that is not prohibited can still be high-risk under this point.
Which biometric uses does Annex III point 1 list?
Point 1 lists three groups. The first is remote biometric identification systems, excluding verification systems whose sole purpose is to confirm that a specific person is who they claim to be, such as unlocking a device or confirming identity at a checkpoint where the person presents themselves for verification. The second is biometric categorisation based on biometric data according to sensitive or protected attributes or characteristics. The third is emotion recognition systems. The wording of each group matters: verification is distinguished from identification, categorisation is tied to sensitive or protected attributes, and emotion recognition is listed as its own group. A system should be matched to the exact group wording, not to a general description of biometrics.
How does verification differ from identification?
Verification confirms a claimed identity: a person asserts who they are, and the system checks whether the biometric data match that claim. Identification searches a database to determine who a person is, without a prior claim. Annex III excludes verification systems whose sole purpose is confirmation, but a system that also searches, categorises or identifies beyond that sole purpose is not within the exclusion. A one-to-one check at a border or a device unlock illustrates verification, while a one-to-many search of a crowd illustrates identification. The record should describe the claim, the comparison, the database searched, and the purpose, because the same biometric technique can be verification in one deployment and identification in another.
For the broader regulatory framework, see high-risk AI systems.
What biometric categorisation is high-risk here?
Biometric categorisation in Annex III concerns categorising natural persons based on biometric data according to sensitive or protected attributes or characteristics. Sensitive attributes include the attributes listed in Article 5(1)(g), and protected attributes include characteristics protected under equality and non-discrimination law. A system that categorises people by age group for a lawful purpose follows a different analysis from a system that categorises by a sensitive attribute. The record should name the attributes, the categories, and the purpose of the categorisation, and should test Article 5(1)(g) first, because categorisation that infers a listed sensitive attribute can be prohibited rather than merely high-risk.
What emotion recognition uses are listed?
Emotion recognition systems are listed in point 1 as high-risk, subject to the Article 5 screen and the Article 6(3) filter. Emotion recognition means inferring emotions or intentions from biometric data. In workplaces and education institutions, Article 5(1)(f) prohibits emotion inference except for medical or safety reasons, so many workplace and education emotion systems will be prohibited before Annex III is reached. Outside those settings, an emotion recognition system can be high-risk under Annex III. The record should state the setting, the biometric input, the emotions inferred, and the Article 5 analysis, so that the Annex III conclusion is not reached without the prohibition screen.
What evidence does a biometric classification need?
A biometric classification record should describe the biometric modality, the data collected, the reference database, the matching or categorisation logic, the accuracy and error rates across relevant groups, the intended purpose, the setting, and the Article 5 and Article 6(3) analyses. Bias and error analysis is particularly important for biometric systems, because errors and differential performance can affect individuals in consequential ways. The record should also address data protection: biometric data are special category data under the GDPR, with strict conditions, and a data protection impact assessment will commonly be needed. The AI Act classification and the GDPR assessment should be kept separate but cross-referenced.
Continue with Article 6 classification routes.
What does this guide not decide?
This guide explains the Annex III biometric groups in outline. It does not decide whether any particular biometric system is identification, verification, categorisation or emotion recognition, whether it is prohibited under Article 5, or whether the Article 6(3) filter applies. Those questions need the biometric modality, the comparison logic, the setting, and the purpose, read against the consolidated text and data protection law. Biometric systems affect identity and sensitive attributes, so borderline cases should be confirmed with a competent authority or legal adviser before deployment. The confirmation should be sought early enough to change the design, the data, or the deployment if the advice is that the system is prohibited or high-risk, because a late confirmation that requires redesign after a system is built is far more costly, in time and in trust, than an early review of the same facts.
Frequently asked questions
Is face unlock on a phone high-risk under Annex III?
A verification system whose sole purpose is to confirm that the person is who they claim to be is excluded from the remote biometric identification group in Annex III point 1. The sole purpose wording must be met, and other laws, including data protection law, still apply.
Can emotion recognition be high-risk outside work and school?
Yes. Article 5(1)(f) is tied to workplace and education settings, while Annex III point 1 lists emotion recognition systems more broadly. An emotion system outside those settings can still be high-risk and must be assessed under Article 6.
Does Annex III cover biometric categorisation by age?
It covers categorisation according to sensitive or protected attributes or characteristics. Whether age categorisation in a specific case falls within that wording depends on the attribute, the purpose and the applicable equality law, and should be checked against the point wording and qualified advice.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689 ↗AI Act Service Desk timeline ↗European Commission regulatory framework for AI ↗