AIEd Knowledge BaseAIEd KB
On this page

Research Article

Predictors of the ethical use of generative artificial intelligence in higher education

created 2026-09-16 · updated 2026-09-18

Synthesis: Tabares-Cruz and colleagues surveyed 980 students at the State University of Milagro (Ecuador) to ask a question that adoption research tends to skip: not why students use Generative AI, but what makes their use ethical. Testing six theoretically distinct predictors at once in a single structural equation model, they found that the six together explain 44% of the variance in ethical use, and that the strongest single contributor is not technical skill but academic integrity — disclosure, authorship, and transparency. Responsible GenAI use emerges, on this evidence, from the interaction of individual literacy, critical verification, self-regulation, and institutional AI Governance rather than from familiarity with the tools.

Key Findings

  • Academic integrity and transparency was the strongest predictor of all six. Norms of authorship and disclosure — acknowledging when AI has contributed to a piece of work, distinguishing one's own contribution from generated text, refusing to pass off generated content as one's own — carried more weight than any technical or institutional factor. The authors read this as evidence that the decisions students make about authorship are the front line of ethical use, because nearly every AI-assisted task raises them.
  • Technical and ethical AI Literacy matters, but as one competency among several. Understanding how these systems generate output and knowing the criteria that separate acceptable from unacceptable use contributed independently, but did not dominate. Proficiency and ethical judgment are separable, and the model treats them as such.
  • Critical verification of generated content is a distinct, substantial contributor. Comparing AI output against reliable sources, spotting bias and overgeneralization, and deciding which parts of a response to trust formed a pathway of their own — close in magnitude to AI literacy. The finding gives empirical weight to the claim that verification is a learnable academic skill, not a by-product of tool fluency.
  • Institutional guidance and ethics education contributed independently of student characteristics. Clear institutional rules, course-level explanation of when AI use is appropriate, and exposure to ethical principles all carried weight on their own. The authors argue this means institutional support should be modeled as a distinct condition rather than treated as an extension of individual competence.
  • Self-regulation and privacy mattered, but least of the six. Setting personal limits on when and how much to rely on AI, retaining responsibility for understanding the material, and guarding personal or confidential data all pointed the right way. Privacy had the smallest effect, which the authors explain as situational: authorship questions arise on almost every task, whereas data-protection questions only arise when sensitive information is actually in play.
  • Ethical use does not follow from frequency of use. Taken together with the descriptive pattern — widespread but not entrenched adoption, and a minority of students clear about institutional rules — the model supports the paper's central claim that exposure to these tools does not by itself produce responsible practice.

Study Design & Method

A quantitative, applied, non-experimental, cross-sectional and confirmatory study using covariance-based structural equation modeling in IBM SPSS AMOS. The target population was the 3,850 students enrolled and academically active at the State University of Milagro in the 2026 academic period; the final sample of 980 was recruited by non-probability purposive sampling, with informed consent required before the questionnaire would open. The sample was heterogeneous by design — a majority female, roughly two-thirds undergraduates, drawn from Education, Social Sciences, Administrative Sciences, Engineering and Health, and split across face-to-face, hybrid, online and blended modalities.

Measures came from a purpose-built 44-item questionnaire on a five-point Likert scale, covering six predictor dimensions — ethical and technical AI literacy, academic integrity and transparency, critical thinking and content verification, self-regulation and academic responsibility, data protection and privacy, and institutional guidance and ethics education — plus eight items capturing the outcome, ethical GenAI use in higher education. Seven reverse-worded indicators were reverse-scored before analysis, and a sensitivity analysis showed that dropping them actually lowered internal consistency, so all 44 items were retained. The measurement and structural models were specified a priori on theoretical grounds: no exploratory factor analysis, no correlated residuals, no post hoc re-specification, and all six structural paths were positive and statistically significant.

Implications

The findings reframe responsible AI use as a governance and assessment problem, not merely a skills problem.

  • Because authorship and disclosure decisions carry the largest weight, institutions should prioritize disclosure protocols, clear norms distinguishing original from AI-generated content, and assessment designs that require students to account for algorithmic assistance — rather than leaning on detection or prohibition.
  • Since institutional guidance and ethics education contributed independently, institutional support is a distinct condition to be designed and resourced, not assumed to follow from capable students. The authors recommend integrating ethics and technical AI literacy across the curriculum, teaching systematic verification against inaccurate or biased output, strengthening data-protection practice, and keeping guidelines consistent across courses, modalities and academic levels.
  • The awareness and training gaps are actionable: substantial minorities of students report only partial knowledge of institutional rules or no formal AI training at all. Training should reach past functional tool use into authorship, transparency, bias, accountability for AI-assisted outputs, and the limits of algorithmic support.
  • The practical message is that exposure to Generative AI does not produce responsible practice on its own — deliberate instruction and institutional policy have to do the work.

Limitations

  • Cross-sectional, non-experimental design: no temporal precedence, so the paths are predictive associations, not causal effects.
  • Non-probability purposive sampling: with 3,850 eligible students and 980 respondents, selection probabilities were unknown; the large sample does not substitute for random selection, and self-selection and coverage bias cannot be ruled out. Generalization beyond the participating students and institution is cautioned.
  • Self-reported measures: susceptible to social desirability bias and to gaps between reported and actual behavior; no behavioral indicators, authentic-task data or platform logs were collected.
  • No competing models or cross-validation: the psychometrics come from the same sample used to estimate the structural model, and no held-out or independent data were used; alternative specifications were deliberately not estimated post hoc, so model uniqueness is not established.
  • Explained variance is incomplete: a substantial share of the variance remains unaccounted for, and the authors note other individual, pedagogical, technological and contextual factors are still at work.

Connected Concepts

Connected Articles

Citation

Tabares-Cruz, Y. B., Cevallos-Sánchez, H. A., Arteaga-Pita, I. G., et al. (2026). Predictors of the Ethical Use of Generative Artificial Intelligence in Higher Education. Frontiers in Education, 11, 1942426.

Metadata

Study designSurvey, Structural Equation Modeling, Educational Measurement
LevelHigher Education
Intended audienceLearners, Teaching, Administrators
Technologies and techniquesGenerative AI
Institutions and policyAI Governance
Equity, ethics, and responsible useEthics, Privacy
Foundations of AI in educationAcademic Integrity, AI in Education, AI Literacy, Critical Thinking
SourceHistoryEditChat

Embed this page

Copy the code below to embed a chromeless version of this page in a learning management system or other website. The embedded view hides the site header, navigation, and footer.