Ethical Conditions for University Students' Adoption of Large Language Models in Exam Preparation Contexts

Synthesis: Pérez-Portabella, Arias-Oliva, Padilla-Castillo, and de Andrés-Sánchez (2026) use the Multidimensional Ethics Scale with 151 Spanish undergraduates to test whether ethical perceptions shape intention to use large language models for exam preparation and, in turn, actual use. Necessary Condition Analysis shows all three ethical dimensions (moral equity, consequentialism, deontology) are necessary conditions for intention to use, while PLS-SEM finds only consequentialist and deontological judgments are sufficient predictors; intention is both a necessary and a strong sufficient predictor of actual use. The results position ethical reasoning as central to responsible Large Language Models (LLMs) adoption in assessment contexts.

Key Findings

• All three ethical dimensions are necessary conditions for forming the intention to use LLMs for exam preparation (moral equity d = 0.338, consequentialism d = 0.274, deontology d = 0.207, all p < 0.001).
• Only consequentialism (β = 0.350) and deontology (β = 0.329) had significant sufficient effects on intention; moral equity did not (β = 0.108, p = 0.414).
• Intention to use is both a necessary (d = 0.325) and a very strong sufficient predictor of actual use (β = 0.905, p < 0.001).
• Students apply multiple normative ethical frames rather than a single moral stance when deciding whether to adopt LLMs in assessment contexts.

What this means for practice

• Instructors. Make the ethical stakes of LLM use in assessment explicit instead of relying on rule compliance, because consequentialist (β = 0.350) and deontological (β = 0.329) judgments were the sufficient predictors of intention to use.
• Instructors. Redesign assessment toward process-oriented tasks that reward creativity, critical reflection, and integration of multiple sources, since students who perceive learning benefit intend to use LLMs even without explicit regulation.
• Administrators. Pair integrity codes with support for students' normative reasoning: moral equity was a necessary condition (d = 0.338) but not a sufficient predictor of intention (β = 0.108, p = 0.414).
• Students. Treat the decision about exam-preparation use as consequential rather than incidental — intention was a very strong sufficient predictor of actual use (β = 0.905).
• Administrators. Position disclosure policies as complements to ethical reasoning rather than substitutes, since moral considerations operate alongside perceived usefulness and ease of use in shaping adoption.

Limitations

• The sample is 151 Spanish undergraduates recruited through purposive and snowball sampling, with a voluntary, anonymous, self-administered online questionnaire completed in 2025.
• Program scope: participants came from social science-related programs, which restricts generalizability to other educational domains such as STEM disciplines.
• Cultural scope: only Spanish participants were surveyed, which limits applicability to culturally distinct environments where the influence of predictors on intention to use can differ.
• The study is cross-sectional, measuring perceptions and behavior at a single point in time, and its model includes only ethical antecedents, omitting academic performance, institutional pressure, and digital literacy.

Citation

Pérez-Portabella, A., Arias-Oliva, M., Padilla-Castillo, G., & de Andrés-Sánchez, J. (2026). Ethical conditions for university students' adoption of large language models in exam preparation contexts. Computers and Education Open, 10, 100323.