AI Exam Proctoring: Detecting Cheating While Protecting Privacy

A signal is not evidence

A proctoring engine produces indicators, never a decision. An off-screen glance or a noise in the room has a dozen innocent explanations. Any serious deployment separates five stages: capture, automated alert, human review, candidate notification, and an appeals route. A penalty issued on an algorithm's output alone will not survive a review board.

What Law 09-08 requires

Analysing a face or a voice is personal data processing, with a stricter regime for biometrics: a legal basis, prior notice to candidates, a written purpose, a bounded retention period, and a named data controller. The point most often missed: the institution remains the data controller, not the software vendor.

Educati's choice: secure the room, not the webcam

Educati captures no video and no audio, ships no facial recognition and no anomaly detection, and processes no biometric data. Exam integrity rests on how the session is organised: each session is scheduled with its room, named seat assignments, designated invigilators, printable convocations and a session report. Access to papers and marks is governed by granular roles; grade changes, exports and permission grants are timestamped in an exportable audit log — plain read access is not logged.

If your institution adopts proctoring anyway

Pick a specialist vendor and require in writing: where processing happens, how long recordings are kept, a guarantee of human review before any penalty, and the appeal route offered to the candidate.