EU AI Act  /  Annex III · Point 3
Annex III · Education

AI in education under the EU AI Act: Annex III point 3, decoded

Admissions, grading, placement, exam proctoring — education and vocational training is one of the eight Annex III high-risk areas. Here is the use-case-by-use-case answer for schools, universities and ed-tech teams: what is high-risk, what escapes, and what is banned outright.

Short answer

If an AI system decides or materially influences who gets admitted, how learning outcomes are evaluated, what level of education someone can access, or whether a student is flagged for cheating during a test, it is high-risk under Annex III point 3. Study chatbots and admin tools that do none of those things are generally not. And one classroom use case is worse than high-risk: emotion recognition of students is prohibited (Article 5(1)(f)), except for medical or safety reasons.

Education AI sits in Annex III because these systems “may determine the educational and professional course of a person’s life” (recital 56) — and, improperly designed, can violate the right to education and perpetuate historical patterns of discrimination. The Act covers “educational and vocational training institutions at all levels”: primary schools, universities, and corporate or professional training programmes alike.

What Annex III point 3 actually covers

Point 3 (“Education and vocational training”) lists four use cases. In condensed form, AI systems intended to be used:

Two limbs deserve a second read. Point 3(b) catches not just grading, but grading that steers the learning process — which is precisely what adaptive-learning platforms do. Point 3(d) is the proctoring clause: detecting “prohibited behaviour of students during tests” is high-risk by name, no interpretation needed. As with all of Annex III, classification follows the intended use under Article 6(2) — a general-purpose model wired into admissions scoring is caught exactly like a dedicated ed-tech product.

Use-case verdicts: education AI, case by case

Verdicts follow the text of the Regulation (Annex III point 3, Article 5(1)(f), Article 6(3) with recital 53). Where a case genuinely depends on configuration, it is marked borderline rather than forced into a bucket.

Education use caseVerdictWhy
Admissions screening / applicant rankingHigh-riskPoint 3(a): determining “access or admission” — named in the Annex itself
Assigning students to schools or programmesHigh-riskPoint 3(a): assigning natural persons to institutions
Automated essay / exam gradingHigh-riskPoint 3(b): evaluating learning outcomes
Adaptive-learning platforms (assessment results steer each student’s path)High-riskPoint 3(b): outcomes “used to steer the learning process”; per-student profiles also trip the profiling carve-out
Placement / level tests (deciding which level of education someone can access)High-riskPoint 3(c): assessing the appropriate level of education
Online exam proctoring (cheating detection during tests)High-riskPoint 3(d): monitoring and detecting prohibited behaviour of students during tests
Emotion recognition of students (attention, engagement or stress detection in class or exams)ProhibitedArticle 5(1)(f) bans inferring emotions in education institutions, except for medical or safety reasons — in force since Feb 2025
Teacher grading-consistency checker (ex-post flagging of deviations from the teacher’s own grading pattern)Borderline — assess under Art 6(3)A verbatim recital 53 example of pattern detection under Art 6(3)(c) — provided it flags for human review rather than replacing the assessment; document under Art 6(4)
Plagiarism detection on submitted courseworkBorderline — assess under Art 6(3)Not named in point 3; outcome depends on whether it materially influences the student’s result or feeds a “prohibited behaviour” process — assess and document the call
AI tutor / study chatbot (explains material, answers questions, no grading or gating)Not high-riskNo point 3 use case. Disclosure duties under Article 50 may apply; re-check if it starts assessing or steering
Improving the academic style of drafted textNot high-riskImproving the language of previously drafted documents — recital 53 gives “academic style” as the example under Art 6(3)(b)
Timetabling, course search, admin automationNot high-riskNo decision about a person’s access, assessment or level — outside point 3

Students are a protected audience twice over. Beyond the emotion-recognition ban, Article 5(1)(b) prohibits AI that exploits vulnerabilities due to age to materially distort behaviour and cause significant harm — a line ed-tech aimed at minors needs to screen against before any high-risk analysis. The prohibited check always comes first: it has applied since February 2025 and carries the Act’s top fine tier.

The Article 6(3) escape route — and the profiling trap

Annex III listing is the default, not the final word. Under Article 6(3), a point 3 system is not high-risk if it does not pose a significant risk of harm to health, safety or fundamental rights — “including by not materially influencing the outcome of decision making” — and meets at least one of four conditions:

  1. it performs a narrow procedural task (recital 53: structuring unstructured data, classifying incoming documents, detecting duplicates among applications);
  2. it improves the result of a previously completed human activity (e.g. polishing language or academic style of a drafted text);
  3. it detects decision-making patterns or deviations from them, without replacing or influencing a completed human assessment absent proper review — recital 53’s example is exactly the education case: checking ex post whether a teacher deviated from their own grading pattern; or
  4. it performs a preparatory task to an assessment (indexing, search, translation, file handling).

The profiling trap. None of the four conditions helps if the system profiles natural persons: “an AI system referred to in Annex III shall always be considered to be high-risk where the AI system performs profiling of natural persons” (Art 6(3), final subparagraph). Learning-analytics and adaptive platforms typically build exactly such profiles — automated evaluation of a student’s performance, behaviour or aptitude. If your system profiles students to make point 3 decisions, stop looking for the exception.

Claiming the exception is a documented act, not an opinion. Under Article 6(4), a provider that considers its Annex III system not high-risk must document that assessment before the system is placed on the market or put into service, register the system in the EU database under Article 49(2), and hand the documentation to national authorities on request.

On guidance: Article 6(5) required the Commission to publish classification guidelines with practical examples by 2 February 2026. That deadline was missed — draft guidelines were published on 19 May 2026, with a consultation running to 23 July 2026, and no final version had been adopted as of early July 2026. Treat the draft as a signal of the Commission’s reading, not settled law.

If you are high-risk: what you must do

The obligations split by role. A school, university or training provider using bought AI is a deployer; the duties live in Article 26:

Fundamental rights impact assessment (FRIA): education deployers should assume they are in scope. Article 27(1) requires a FRIA before first use of an Annex III high-risk system from deployers that are bodies governed by public law — public schools and universities — and from private entities providing public services; recital 96 lists education expressly among those public services. The assessment covers the processes of use, period and frequency, affected categories of persons (students and applicants), the specific risks of harm, human-oversight measures, and what happens when risks materialise.

If you build or sell education AI (or substantially modify a bought system, or put your name on it — Article 25), you carry the provider stack: risk management system (Art 9), technical documentation, transparency to institutions that deploy you (Art 13), human-oversight design (Art 14), conformity assessment and registration — and the Article 6(4) documentation if you claim any product line is exempt.

When this applies

The Regulation as enacted set 2 August 2026 for the Annex III obligations. The Digital Omnibus on AI — adopted by the European Parliament on 16 June 2026 and by the Council on 29 June 2026, but as of early July 2026 still awaiting publication in the Official Journal — defers the Annex III high-risk obligations to 2 December 2027. The Article 5 prohibitions do not move: emotion recognition in education institutions has been banned since February 2025.

Classifying your education AI portfolio

An institution’s AI rarely arrives as one decision. It arrives as an admissions module here, a proctoring vendor there, an adaptive add-on inside the LMS, a pilot the mathematics faculty started on its own. Each of those is a separate classification: prohibited screen first, then the point 3 check, then any Article 6(3) claim with its documented rationale — re-run whenever the tool or its use changes.

Track this in your Jira

Give every education AI use case a classification record

Model Inventory for Jira turns each AI system into a work item in the Jira your team already uses, with a built-in EU AI Act category field and dynamic risk tiering. Record the Annex III point 3 verdict, the Article 6(3) rationale, and who signed it off — with an immutable change history, so when the LMS vendor ships a new “AI insights” feature, the record shows the classification needs a re-check. The legal judgement stays with your compliance team; the inventory makes sure no tool skips the question.

See how it works

FAQ

Is AI essay or exam grading high-risk under the EU AI Act?

Yes — point 3(b) covers evaluating learning outcomes, including when those outcomes steer the learning process. An ex-post checker that only flags deviations from a teacher’s own grading pattern for human review is a recital 53 example of a system that may escape via Article 6(3).

Is online exam proctoring high-risk under the EU AI Act?

Yes — point 3(d) names monitoring and detecting prohibited behaviour of students during tests. And if the proctoring infers students’ emotions (nervousness as a cheating signal), it crosses into the Article 5(1)(f) ban.

Are AI tutoring chatbots high-risk?

Generally no — until they evaluate learning outcomes, steer the learning process based on assessments, or gate access to levels or programmes. Article 50 disclosure duties may still apply, and student profiling for point 3 decisions always means high-risk.

Do schools and universities need a FRIA?

Mostly yes. Article 27 covers bodies governed by public law and private entities providing public services — recital 96 lists education expressly. Expect the FRIA duty for high-risk admissions, grading or proctoring systems, on top of Article 26.

When do these rules start to apply?

Prohibitions: since February 2025. Annex III high-risk obligations: 2 August 2026 as enacted, deferred to 2 December 2027 by the Digital Omnibus adopted in June 2026 — which was still awaiting Official Journal publication as of early July 2026.

This page is a practical explanation, not legal advice. Always confirm classification against the official text of Regulation (EU) 2024/1689 and, where the stakes warrant it, qualified counsel.