Intentions for AI academic misconduct across 15 countries: the theory of planned behavior and past misconduct...

With generative artificial intelligence (GenAI) tools now widely available in higher education, understanding what predicts students' intentions to use them dishonestly has become an increasingly important concern for universities worldwide. In this pre-registered study, we applied the extended Theory of Planned Behavior (TPB) to examine the predictors of AI academic misconduct intentions (AIMI) in 2,883 English-proficient university students from 15 countries recruited via Prolific. Because these intentions were strongly concentrated at zero, we used a Bayesian hurdle model that separates two processes: whether a student reports any intention to use AI dishonestly, and how strong that intention is among those who report one. More favorable attitudes, higher AI proficiency, and weaker moral obligation predicted both a greater probability of reporting any intention and greater intention intensity, with attitudes the strongest and most consistent predictor across both components. Weaker subjective norms predicted whether students reported any intention but not its intensity. A history of traditional academic misconduct and a history of AI-enabled misconduct each predicted both components. The final model accounted for 18.7% of the variance in whether students reported any intention and 47.8% of the variance in the intensity of that intention. These findings provide an initial cross-national test of the extended TPB for AI misconduct intentions, but future research in native-language, institution-based samples is still needed to verify the cross-cultural specificity of these mechanisms.