Domain 3 of Annex III AI Act touches every educational institution and EdTech vendor deploying AI for admission, evaluation or behavioural monitoring. The Commission guidelines of 19 May 2026 sharply distinguish between pedagogical support (often filter-eligible) and evaluative applications affecting a student's future (always high-risk).
For the general framework, see the main article on the Article 6(3) filter. For all domains, see the hub overview.
The Four Use Cases of Domain 3
- Point 3(a) Access or admission to educational institutions at all levels
- Point 3(b) Evaluation of learning outcomes and steering of the learning process
- Point 3(c) Assessment of the appropriate level of education
- Point 3(d) Monitoring and detection of prohibited behaviour during tests
Point 3(a): Admission and Assignment
High-risk:
- AI scoring or ranking university admissions
- AI assigning new students to educational streams
- AI supporting lottery or placement in schools based on personal characteristics
- AI conducting selections for scholarships, talent programmes or specialist programmes
Filter possible:
- AI categorising applications into fixed folders per programme without evaluation
- AI verifying authenticity of identity or diploma documents (not candidate evaluation)
Point 3(b): Evaluation of Learning Outcomes
High-risk:
- AI grading tests or exams
- AI scoring essays
- AI in adaptive learning evaluating student progression and steering it
- AI giving progression or retention advice based on learning performance
Filter possible:
- AI generating only spelling and grammar feedback without grade
- AI suggesting ready-made exercises to teachers without evaluating students
- AI generating anonymous classroom aggregates for teacher quality purposes
Point 3(c): Education Level Determination
High-risk:
- AI interpreting placement tests and advising on placement
- AI determining education-level transitions
- AI determining appropriate level in adult education or lateral entry
Point 3(d): Proctoring and Behaviour Detection
High-risk:
- AI in online proctoring detecting irregular behaviour during a test
- AI tracing cheating or plagiarism
- AI flagging candidates for human review based on behaviour or biometrics
Filter possible:
- AI only preparing the test environment (identity verification done beforehand, no behaviour detection during the test)
Watch for prohibition: Emotion recognition in educational institutions is prohibited under Article 5 (except for medical or safety reasons). Proctoring tools measuring emotions or stress run into that.
Sector-Specific Pitfalls
Pitfall 1: Adaptive Learning Is Almost Always Point 3(b)
EdTech vendors often present adaptive learning as personalised learning paths, but as soon as the system evaluates whether a student masters a topic and steers the path accordingly, you fall under Point 3(b).
Pitfall 2: Teacher-Assist Doesn't Shift the Problem
"Our AI just supports the teacher" doesn't work as an argument. As soon as the AI's output influences the final evaluation, it is high-risk, regardless of who presses the button.
Pitfall 3: Profiling Excludes the Filter
Many EdTech systems process personal data to predict individual learning progress. That is profiling under GDPR, and the filter is automatically excluded.
What to Do
Inventory EdTech in scope
LMS, adaptive learning platforms, examination software, proctoring tools, dropout prediction models.
Separate pedagogical from evaluative functions
For systems doing both, determine which functions fall under which use case. Sometimes the evaluative module can be classified separately.
Embed teacher AI literacy
Test your AI against the Article 6(3) filter
Free interactive self-assessment, updated for the Commission guidelines of 19 May 2026. 9 steps, personal report with reasoning, vendor questions and next steps.