MJSTEM

Abstract

AI-writing detectors are increasingly used in educational assessment, but their learner-facing consequences remain insufficiently understood, especially for writers using English as a foreign language. This study examined AI-detection anxiety, writing self-censorship, and humaniser intention among Vietnamese EFL learners, compared previously flagged and never-flagged learners, and tested the association between flagging experience and categorical humaniser intention. A cross-sectional descriptive-comparative survey was completed by 142 adult learners. Data were analysed in R using frequencies, percentages, means, standard deviations, Cronbach’s alpha, Welch’s independent-samples t-tests, Cohen’s d, chi-square analysis, and Cramér’s V. AI-detection anxiety was the most strongly endorsed construct (M = 3.65, SD = 0.98), followed by humaniser intention (M = 3.13, SD = 1.01) and writing self-censorship (M = 2.97, SD = 0.98). Previously flagged learners reported higher anxiety than never-flagged learners, t(87.52) = 2.22, p = .029, d = 0.41. No significant group difference emerged for self-censorship, and flagging experience was not significantly associated with categorical humaniser intention. The study distinguishes an emotional response, a restrictive writing response, and a prospective tool-use intention rather than treating them as equivalent evidence of misconduct. Detector-based scrutiny may create substantial anxiety without uniformly producing self-censorship or humaniser intention. Institutions should not treat detector scores as conclusive evidence and should use transparent, human-reviewed authorship procedures.

Keywords