AI proctoring uses machine-learning models to monitor candidate behaviour during a remote exam. The platform typically captures webcam video, microphone audio, screen activity, and browser state — then flags events (no face detected, second person in frame, window switch, unknown background voice) for an institution's review panel to decide on.
Three proctoring modes
- Live proctoring — a human proctor watches up to ~30 candidates in real time.
- Record-and-review — the session is recorded, AI flags events, a human reviews flagged segments after.
- Auto-proctoring — fully AI-decided. Most institutions use it as a first-pass filter, not a final verdict.
Privacy, bias, and DPDPA
AI proctoring has well-documented bias issues (face-detection failure on darker skin tones, false positives for neurodivergent candidates). Responsible platforms publish their failure-rate metrics, give candidates a path to appeal flagged sessions, and let institutions opt into manual-only review. Under India's DPDPA 2023, EU GDPR, and similar frameworks, candidates must consent to recording and have the right to review the data captured about them.