Generative AI Misuse and Rule Compliance Among University Students in Bhutan: Implications for Assessment and Institutional AI Policy

Background and Aim: Generative artificial intelligence (AI) use is expanding in Bhutanese universities while institutional policies remain under development. Guided by an extended Theory of Planned Behavior, this study examined factors associated with unauthorized AI use and intention to comply with institutional AI policies. Materials and Methods: An anonymous, cross-sectional survey was conducted in June 2026 among a convenience sample of 713 Bhutanese university students. Seven-point Likert-type responses were analyzed using factor analyses, reliability assessments, correlations with false discovery rate correction, multiple and moderated regression, and bootstrapped indirect association tests. Robustness checks included robust standard errors and demographic controls. Results: The models explained 24.5% of variance in AI misuse and 19.6% in compliance intention. Approximately 29% of respondents endorsed at least one misuse item. Misuse was associated with misuse acceptability, AI familiarity, academic pressure, moral rationalization, rule clarity, subjective norms, and learning impact. Compliance intention was associated with subjective norms, rule clarity, behavioral control, productive AI use, institutional enforcement, and moral rationalization. Institutional enforcement showed a small independent association with compliance intention (β = .086). Compliance intention was inversely associated with misuse (B = −0.10, p = .004). All six indirect associations were small, with bootstrap confidence intervals excluding zero. Demographic adjustment preserved findings; academic level showed no significant moderation. Conclusion: Findings support evaluating assignment-specific AI guidance, coordinated workloads, classroom discussion of misuse rationalizations, and communicated institutional policies. Cross-sectional data preclude causal inference, and convenience sampling limits generalizability. These associations provide a basis for evaluating approaches to responsible AI governance.