From screening to gatekeeping: AIGC-detection governance in Chinese university thesis assessment — with a focus on medical education

AI-writing detectors are increasingly embedded in graduation-thesis workflows in Chinese universities. Their use is not only a technical question of classification accuracy. Once detector scores influence review, defence or degree progression, they also allocate evidentiary authority and define students’ procedural protections. We conducted a directed policy analysis of a purposive, maximum-variation sample of 20 institutional documents published between March 2025 and May 2026. The primary scope was Chinese university thesis governance, with medical and health-related policies examined as a subgroup. We coded directly observable policy features and then applied rule-based classifications for evidentiary logic, student consequence and procedural fairness. Nineteen policies required formal detection, and 14 specified a warning or submission threshold. Nine policies used detector output as a direct decision gate, eight combined it with human review and three used it mainly as a screening clue. Procedural fairness was rated as high in 3 policies, moderate in 8 and low in 9. Medical and health-related policies more often used direct gates than institution-wide or non-medical policies (5/9 versus 4/11), and none reached the high-fairness category. Threshold-linked rules may create pressures to optimise detector scores, but student behaviour was not measured. The findings describe formal policy language rather than implementation or national prevalence. More balanced governance would treat detection as a screening signal within a reviewable process that includes disclosure, human adjudication, supplementary assessment and appeal.