Students' Non-Disclosure of Generative AI Use in Academic Assessment

Abstract

Generative AI (GenAI) has become a near-universal feature of students' academic work, yet institutions cannot reliably detect it. As a result, many universities rely on students to disclose their GenAI use themselves, making honest self-disclosure central to the validity of assessment. When students conceal this use, the credibility of assessed work is undermined. This study examines why students engage in such non-disclosure. Rather than treating non-disclosure as a single act, it distinguishes full non-disclosure (concealing GenAI use entirely) from partial non-disclosure (admitting some use while withholding its extent or nature), and it measures each as both past behavior and future intention. Drawing on Deterrence Theory and the Theory of Planned Behavior, together with academic self-efficacy and policy awareness, the study tests how these factors explain non-disclosure. Data were collected through a cross-sectional survey of 431 higher education students and analysed using hierarchical regression across the four outcomes, complemented by an exploratory mediation analysis. The findings show that non-disclosure is primarily an evaluative and social decision. Attitude toward concealment was the strongest predictor of every outcome, followed by subjective norms, while the perceived certainty, severity, and swiftness of sanctions had only weak and largely indirect effects. Perceived behavioral control, self-efficacy, and policy awareness mattered mainly for the partial forms. These results suggest that institutions can address non-disclosure more effectively by shaping shared norms and understanding than by relying on harsher enforcement.