Artificial Intelligence in Online Proctoring: Algorithmic Bias, Accountability, and the Limits of Automated Final Decision-Making

The use of Artificial Intelligence (AI) in online proctoring systems has expanded rapidly in response to the growing demand for efficient and scalable remote examination supervision. Various AI-based approaches, such as facial recognition, eye-gaze tracking, facial expression analysis, and behavioral classification, are employed to detect potential violations during examinations. However, the application of AI in high-stakes online examinations raises serious concerns related to algorithmic bias, detection errors (false positives and false negatives), and the clarity of accountability in decision-making processes. This article aims to critically examine the limitations of AI in online proctoring systems, particularly when AI is used as the basis for final decisions such as blocking examinees, canceling examinations, or invalidating exam results. The research adopts a systematic literature review of recent studies addressing AI bias, the reliability of automated proctoring systems, and ethical as well as governance issues in educational assessment technologies. The findings indicate that AI-based online proctoring systems possess inherent limitations influenced by the quality of training data, cultural context, participants’ environmental conditions, and algorithm design. Consequently, the use of AI as a final judgment mechanism may lead to unfair outcomes and accountability conflicts when system errors occur. This study argues that AI should be positioned as a decision support system, while final decisions must remain under human authority with clear verification mechanisms and accountability structures. These findings are expected to provide a conceptual foundation for educational institutions and system developers in designing online proctoring policies that are more fair, transparent, and responsible.