The EU AI Act treats AI used to monitor students during tests as high-risk. After the 2026 amendments, those obligations apply from 2 December 2027. One part is already in force: since February 2025, emotion recognition in education is prohibited.
What is already banned
Systems that infer emotions or intentions from faces or voices in an educational setting. A proctoring tool that scores “nervousness” or “suspicious affect” is out. Detecting events, such as a second person, a phone or a tab switch, is not emotion recognition.
What arrives in 2027
- Risk management and data governance across the system’s life
- Automatic logging of events
- Human oversight: a person who understands the output and can override it
- Technical documentation and instructions for use
Design choices that age well
Keep flags advisory. Show the evidence and the timestamp. Let automatic action happen only on events the server can verify. Log every human override. These choices are good practice everywhere, not only in Europe, and they are how Edorer proctoring works today.