High risk · Annex III(3)

EU AI Act & Education: Which AI Tools Are High-Risk (Annex III)

AI now decides who gets in, grades work, and watches exams. Because these calls shape a person's future, Annex III point 3 classes them as high risk.

What Annex III(3) covers

  • Admission and assignment to institutions
  • Evaluating learning outcomes
  • Assessing the appropriate level of education
  • Monitoring and detecting prohibited behaviour during tests

Who’s affected

Schools, universities, exam boards, and edtech vendors building admissions, grading or proctoring AI.

Why it’s high risk

These systems are high risk under Annex III(3). Schools using them are usually deployers (Art. 26) — with human-oversight and transparency duties toward students — while edtech vendors are providers. Note that emotion recognition in education is separately prohibited.

Obligations to prepare for

Risk management system (Art. 9)
Data governance & bias checks (Art. 10)
Technical documentation — Annex IV (Art. 11)
Record-keeping / logging (Art. 12)
Transparency & instructions (Art. 13)
Human oversight (Art. 14)
Accuracy, robustness & cybersecurity (Art. 15)
Inform affected students (Art. 26)

FAQ

Is AI exam proctoring high-risk?

Yes — monitoring and detecting prohibited behaviour during tests is expressly listed in Annex III(3). If the proctoring infers emotions, that part may be prohibited outright under Art. 5.

Is an AI grading tool high-risk?

If it evaluates learning outcomes or steers a student's educational path, yes. A tool that only formats feedback without influencing the grade is a weaker case for high-risk.

Being high-risk sets the obligations; your role decides who owns them. See provider vs deployer duties · sector detail: Education & EdTech.

Not sure if your system falls under Annex III(3)?

The free checker classifies it deterministically in 2 minutes — no sign-up.

Run the free check