To cheat or not to cheat? Examining undergraduate students’ use of generative AI and academic dishonesty

Generative artificial intelligence (GenAI) has become embedded in everyday university study, intensifying long-standing concerns about academic integrity now that students can obtain finished work on demand. Yet why some students cross the line into academic dishonesty—and whether teaching them to feel capable of and in control of AI protects against it—remains unclear. Framing propensity toward AI-enabled academic dishonesty as the focal educational outcome, and drawing on the Interaction of Person-Affect-Cognition-Execution (I-PACE) model, this study used an explanatory mixed-methods design. We analyzed a cross-sectional survey of undergraduates across disciplines at several Chinese universities ( N = 1,045 after removing 115 respondents who gave the same answer to all 41 items) and semi-structured interviews with seven students, which were analyzed thematically. Academic dishonesty was common (50.1% of students scored 5 or higher on a 1–7 scale) and was most strongly associated with AI dependence ( β = 0.63), which was in turn associated with impulsivity, trust in AI, and fear of missing out; trust was also directly associated with dishonesty. Contrary to the widely held assumption that agency protects integrity, a sense of human–AI agency did not moderate the dependence–dishonesty association, and was itself positively associated with dishonesty. Habitual verification of AI output showed the complementary pattern: it did not weaken the dependence–dishonesty association, but it was associated with markedly lower dependence, and thus indirectly with less dishonesty. A latent profile analysis converged on the same account—dependence ordered the profiles on dishonesty, whereas felt agency did not order them at all—and interviewees invoked a sense of control when accounting for passing AI work off as their own. Because the survey was cross-sectional and self-reported, these are associations, not demonstrated effects. For teaching and integrity policy, they nonetheless suggest that curbing dependence and calibrating students’ trust in AI, rather than cultivating a general sense of agency, are the more promising targets.