Automated proctoring uses AI-based analysis — typically computer vision on the webcam feed and heuristics on browser telemetry — to monitor an exam session without any human watching in real time or reviewing recordings afterward. The system detects signals such as a face disappearing from frame, multiple faces appearing, tab switching, copy-paste events, or audio anomalies, and produces a risk score or list of flagged incidents attached to the submission. The main appeal is scale and cost: automated proctoring can support tens of thousands of concurrent exam-takers without proportional staffing increases, which is why it is common in MOOCs and large professional certification programs. However, fully automated systems have well-documented accuracy problems: they cannot reliably distinguish intent from coincidence, and facial-recognition and gaze-tracking components exhibit bias across demographic groups, meaning they can systematically disadvantage certain learners. Automated proctoring is therefore best treated as a signal-generation system rather than a decision-making system — its outputs should feed a human review queue for anything consequential rather than triggering automatic grade penalties. Any deployment must address biometric data handling under applicable law, provide clear disclosure to candidates before the exam, and establish a transparent appeal process. Institutions that skip the appeal process face heightened legal and reputational risk when automated flags produce outcomes that learners contest.
