Ethical conditions for university students’ adoption of large language models in exam preparation contexts

Large Language Models (LLMs), have become a disruptive force in higher education, assisting students in tasks such as summarizing content, clarifying concepts, and preparing for exams. Their rapid adoption has generated both enthusiasm and ethical concern, particularly regarding fairness and responsible use in assessment contexts. This study examines whether university students’ ethical perceptions regarding the use of LLMs for exam preparation influence their intention to use them (IU), and whether this intention, in turn, affects their actual use (USE). Data were collected from 151 undergraduate students enrolled in Spanish universities. Ethical perceptions were measured using the Multidimensional Ethics Scale (MES), which captures three normative perspectives: moral equity (ME), consequentialism (CO), and deontology (DE). The first research objective (RO1) was to determine whether a minimum threshold in each ethical dimension is required for students to form the intention to use LLMs. The second (RO2) assessed whether higher ethical evaluations lead to greater acceptance and use. Necessary Condition Analysis revealed that all three ethical dimensions are necessary conditions for IU (ME: d = 0.338; CO: d = 0.274; DE: d = 0.207; all p < 0.001). Partial Least Squares-Structural Equation Modeling showed that only CO (β = 0.350, p < 0.001) and DE (β = 0.329, p < 0.001) exert statistically significant and sufficient effects on IU, while ME does not (β = 0.108, p = 0.414). Finally, IU is both a necessary ( d = 0.325, p < 0.001) and sufficient predictor of USE (β = 0.905, p < 0.001). The findings highlight that ethical reasoning is central to the responsible adoption of LLMs in exam preparation, offering practical guidance for educators and institutions promoting academic integrity.