EU AI Act in Education (Annex III, 3): Schools, Universities, Training

Practitioner note: This is not legal advice. For specific situations, consult a qualified attorney or compliance officer.

TL;DR

  • Annex III, 3 classifies AI for admission, assessment, performance evaluation, and proctoring as high-risk
  • Emotion detection in classrooms is prohibited under Art. 5(1)(f) EU AI Act — including most AI proctoring
  • Special category data (health, disability) often present — Art. 9 GDPR + Art. 10 EU AI Act bias test apply
  • Teacher as final decision-maker — AI may only recommend
  • Full applicability: Aug 2, 2026 legally binding (Digital Omnibus proposal of Nov 19, 2025 would postpone to Dec 2, 2027 — trilogue ongoing, NOT adopted)

1. Annex III, 3 EU AI Act

Covers AI systems for "determining access, admission, or assignment to education and vocational training institutions," "evaluation of learning outcomes," "assessing performance level," and "monitoring prohibited behavior during exams." All four sub-categories are high-risk for both Providers and Deployers.

2. Obligations from Aug 2, 2026 (Digital Omnibus proposal of Nov 19, 2025: postponement to Dec 2, 2027 — trilogue ongoing, NOT adopted)

FRIA, conformity assessment, bias testing, transparency, human oversight, logging, accuracy and robustness. Schools and universities are typically Deployers; ed-tech vendors are Providers with the heavier Art. 16-26 obligations.

3. Special category data (Art. 9 GDPR)

In education, AI often processes health and disability data (e.g., learning-disability accommodations). This triggers Art. 9 GDPR plus the Art. 10 EU AI Act bias-test obligation. Document each processing purpose and legal basis explicitly.

4. Emotion detection prohibition

Art. 5(1)(f) prohibits AI emotion recognition in education. This catches most AI proctoring tools that infer "stress," "concentration," or "deception" from webcam input. Behavior-based monitoring (e.g., gaze tracking, tab switching) is permissible if proportionate; affect-based scoring is not.

5. Information for parents and students

Art. 50 EU AI Act and GDPR Arts. 13-14 require transparent disclosure of AI use, including the right to object. Plain-language explanations, opt-out paths, and a complaint channel are mandatory.

Summary

Education AI sits at the intersection of high-risk classification, special-category data, and prohibited practices. Schools and ed-tech vendors should map every AI tool against Annex III, 3, exclude any emotion-detection features, and maintain a teacher-final-decision rule. FRIA and conformity assessment must be ready by Aug 2, 2026 (Digital Omnibus proposal of Nov 19, 2025: postponement to Dec 2, 2027 — trilogue ongoing, NOT adopted).

View EU AI Act Kit →

Frequently Asked Questions

Is anti-cheating software permitted?
Behavior-based yes, emotion detection no. Differentiation is important.
Teacher as final decision-maker?
Mandatory. AI provides a recommendation, the teacher makes the final decision.

Sources

Tools & self-assessments

EU AI Act Quick Test Classifies your AI system by risk level (Art. 6, Annex III). Fining Calculator Estimate the potential fine exposure for your organisation. EU AI Act Self-Assessment Classification plus obligations mapping for all AI systems in the organisation. AI Inventory Quick Check Systematic capture of your AI applications in 8 steps.