PROCTORING AT A NEW STAGE: NEXT-GENERATION AI SYSTEMS FOR ENSURING TRUST IN DIGITAL EXAMINATIONS

Abstract

The rapid proliferation of online and hybrid learning environments, accelerated by institutional digitalisation programmes across Central Asia and beyond, has elevated academic integrity to a critical challenge worldwide. Legacy proctoring solutions — whether in-person supervision or first-generation remote monitoring software — prove systematically inadequate against sophisticated cheating strategies enabled by generative artificial intelligence (AI) tools. This paper presents a rigorous analysis of a next-generation AI-powered proctoring architecture deployed at scale, combining four integrated layers: (1) multi-modal behavioural analysis (facial landmark tracking, gaze estimation, head pose, keyboard dynamics); (2) large language model (LLM) semantic anomaly detection and stylometric authorship verification; (3) federated, differential-privacy-preserving learning (ε = 0.5) that ensures no raw biometric data leaves institutional premises; and (4) explainable AI (XAI) decision support via SHAP values, producing human-readable integrity reports with full audit trails. Drawing on real-world deployment data from Proctoring.uz — the first locally developed AI proctoring platform in Central Asia, covering 129,463 examination sessions across three government ministries and multiple higher education institutions in Uzbekistan — we demonstrate that the proposed hybrid architecture achieves a detection accuracy of 99.8%, an F1-score of 0.998, and a false positive rate of 0.2%, while reducing average review time per flagged session from 42 minutes to 11 minutes relative to legacy systems. The platform is formally certified by the Uzbek Ministry of Justice (Certificate No. DGU 37103), recognised as a top-10 winner among nearly 300 startups at the U-Solve Incubation Programme (supported by UK International Development / VirtualAccelerate), and selected for presentation at GITEX AI Central Asia 2026 (Almaty, Kazakhstan). A structured competitive evaluation against ProctorU, Honorlock, and Proctorio, based on publicly documented capabilities, demonstrates 100% coverage of 18 monitoring parameters with a price point estimated at 10–15 times lower than international alternatives. Ethical dimensions — algorithmic bias, surveillance governance, and explainability limits — are examined alongside technical results.