Annex III point 3 lists AI systems intended to be used to determine access or admission, to allocate persons to educational and vocational training institutions, to evaluate learning outcomes, and to determine the appropriate level of education or training for a person, as high-risk use cases under Article 6(2), subject to the Article 6 conditions and the profiling limit. Those functions must be tested even where a system is described as a learning aid: the question is whether it determines access, allocation, evaluation or level, not whether it is used in a school or university. Article 5 screening for emotion inference in education, and transparency and data protection duties, must be tested separately. This guide does not decide any specific education system and does not give education law advice.
Which education functions are listed in Annex III?
Annex III point 3 lists, in outline, systems intended to determine access or admission to educational and vocational training institutions, systems intended to allocate persons to those institutions, systems intended to evaluate learning outcomes, and systems intended to determine the appropriate level of education or training for a person. Each function is consequential: admission and allocation decide who gets a place, evaluation decides whether learning outcomes are met, and level determination decides what education a person receives. A system that only delivers content, only stores learner records or only schedules classes, without determining access, allocation, evaluation or level, is assessed differently, and the distinction must be recorded from the actual function. The record should describe the decision the system influences, the data it uses, the output it produces and the person who acts on the output, so a reviewer can see which listed function, if any, is engaged.
How is an education system classified?
Classification follows the Article 6 route: the intended purpose is tested against the Annex III listing, the Article 6(2) conditions are applied, and any Article 6(3) question is tested against the narrow tasks and the profiling limit. Profiling is decisive: an Annex III system that performs profiling of natural persons is always high-risk, and many education evaluation systems will involve profiling where they assess a learner on the basis of personal characteristics or behaviour. A system that evaluates learning outcomes by scoring a learner's work, or that determines a level by assessing a learner's performance, should therefore address the profiling question directly in its classification record. The provider's Article 6(4) documentation, where Article 6(3) is relied on, must be prepared before market placement, and the deployer should not assume the provider's conclusion without seeing the reasoning. Where classification is contested, qualified review should be sought before the system affects admissions or assessment.
For the broader regulatory framework, see healthcare and biometrics under the AI Act.
What must be screened under Article 5?
Classification in education turns on the precise function, so the record should name it exactly: determining access or admission, assigning a person to an institution, evaluating learning outcomes, assessing the appropriate level of education, or monitoring and detecting prohibited behaviour during tests. Each function in point 3 of Annex III is tested on its own, and a system that does several of them is assessed for each. The Article 6(3) question then follows: does the system perform only a narrow procedural task, only improve a prior human result, only detect decision-making patterns without replacing human assessment, or only prepare a task, and does it profile learners? A system that scores, ranks or flags learners on the basis of their characteristics will often involve profiling, which closes the Article 6(3) route. The provider's reasoning, where Article 6(3) is relied on, should be documented before market placement, and the deployer should see that reasoning rather than accepting a conclusion on trust.
What oversight and learner information are needed?
The Article 5 screen in education needs the same care as the classification work. Systems that infer emotion in education, identify learners by biometric data, or categorise learners by biometric data in ways that infer sensitive attributes can engage prohibitions, and tools that monitor behaviour during tests can sit close to those lines. The screen should record the input data, the inferences and the decisions for each prohibition considered, and should be done before conformity work begins, since a prohibited use must not proceed to it. Education and child-protection law, and data protection law for learners including minors, add conditions that the AI Act does not itself set out, and the file should show how those regimes were considered together. A tool described as attention monitoring or engagement detection should be tested on what it infers from learners, not on the name given to the feature.
What evidence should an education deployment keep?
Oversight in education means a teacher or assessor with the competence, time and authority to weigh the system output against the learner's wider record and to decide differently where the facts require it. Learners, and parents where learners are minors, need a clear account of the AI use, what it does, what data it uses and how a decision can be challenged, within the information duties of education and data protection law. Monitoring should track whether the system performs consistently across learner groups, including learners with disabilities and learners in different home and language circumstances, whether outputs are followed or overridden, and whether outcomes shift in ways nobody intended. An unexpected pattern should be escalated, reassessed and recorded with the action taken, so that a later question about an admission or assessment decision can be answered from the record.
Continue with education monitoring and proctoring.
What does this guide not decide?
No education system is assessed here, and no admission or assessment decision is judged. Whether a system is high-risk or prohibited, and whether an education decision was lawful, depends on the function, the data and inferences, the decision process, and the education, equality and data protection law that applies, worked through on the specific facts with qualified advice. Before an education system influences an admission or assessment decision, the classification, prohibition, transparency and data protection positions should be confirmed with a competent authority or legal adviser.
Frequently asked questions
Is a tutoring chatbot high-risk?
It depends on its function. A chatbot that only answers questions or delivers content is assessed differently from a system that determines admission, allocates learners, evaluates outcomes or determines level, which are listed functions in Annex III point 3.
Can emotion inference be used in classrooms?
As a general rule, no. Article 5 prohibits emotion inference in education institutions except for medical or safety reasons, and the exception must be tested on the facts with qualified advice.
Must learners be told AI is used in assessment?
Transparency, data protection and education law duties can require information to learners, parents or guardians and staff. The information should be clear, age-appropriate and given before the system affects a decision, with a route to challenge.
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, as amended ↗AI Act Service Desk implementation timeline ↗