Generative AI-related academic misconduct among medical students: co-occurrence with conventional misconduct and associated factors
Year: 2026
Type: article
Study design: Observational Study
Abstract: Background Generative artificial intelligence (AI) has introduced forms of academic misconduct that may occur beyond established patterns of cheating. It remains unclear whether these behaviors replace conventional misconduct or coexist with it among the same students, particularly in medical education. We estimated the past-year prevalence, mapped the overlap between AI-related and non-AI-related misconduct, and examined modifiable correlates. Methods We conducted an anonymous, online, cross-sectional survey at a medical university in eastern China in September 2026. Students reported on 12 conventional and AI-related academic behaviors during the period immediately preceding the survey. The primary AI-related outcome involved the submission of AI-generated work without the required disclosure and the use of AI paraphrasing or translation to evade detection. The use of AI to generate answers in an assessment was analyzed separately because authorization and disclosure were not measured in this context. We also assessed an intention score that reflected respondents’ willingness to submit AI-generated work as their own and to use AI-based paraphrasing or rewriting to evade plagiarism or AI detection measures. Four co-occurrence profiles were compared, and exploratory hierarchical linear and logistic models were fitted with robust inference and internal bootstrap correction. Results Of the 421 submissions, 359 met the prespecified quality criteria. A total of 64 students (17.8%, 95% CI: 14.2–22.1) reported a primary AI-related behavior, 52 (14.5%, 95% CI: 11.1–18.6) reported a conventional behavior, and 79 (22.0%, 95% CI: 18.0–26.6) reported at least one primary AI-related or conventional behavior. The profiles were categorized as follows: neither profile (280 students, 78.0%), conventional-only profile (15 students, 4.2%), AI-only profile (27 students, 7.5%), and both profiles (37 students, 10.3%). The assessment-answer item was reported by 67 students (18.7%) and analyzed separately. In exploratory adjusted models, perceived peer completion of undisclosed AI work and the neutralization of AI use were associated with the primary outcome, while perceived opportunity and neutralization were associated with intention. Conclusion In this single-university sample, the reported AI-related and conventional behaviors showed both overlap and distinct characteristics. The findings are correlational and do not establish whether AI use caused misconduct. Clear task-level guidance and assessment methods that make authorship visible are reasonable options for the future evaluation; however, their effectiveness was not tested in this study.
Source: Frontiers in Public Health
Author: Xiuhong Yang
Language: English
Subfield: Education
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7225414516