Academic Integrity in the Age of AI: University Students' Study Practices and Ethical Judgments

Synthesis: López-López, Bru-Cordero, and Correa-Álvarez (2026) analyzed survey data from 357 undergraduates across 14 programs at a Colombian public university to examine how students interpret academic integrity in relation to AI. They found AI use is routine and usually perceived as helpful for understanding content, yet students' ethical judgments remain divided—a slight majority rejected the idea that AI use is fraud, over one-third were undecided, and a smaller group endorsed it. The authors introduce the concept of pragmatic ambiguity to describe how students negotiate AI use between academic usefulness, uncertain institutional boundaries, and concerns about authorship and intellectual contribution.

Key Findings

1. AI use is routine. Students across programs use tools like ChatGPT, Gemini, Copilot, and DeepSeek for brainstorming, explanation, translation, summarization, coding support, feedback, and early drafting.
2. Judgments are divided, not settled. A slight majority rejected the view that using AI in academic tasks is fraud; more than one-third were undecided; a smaller group endorsed the fraud framing.
3. Use and perceived value drive permissiveness. More frequent AI use (χ² = 78.08, p < 0.001) and stronger perceived learning support were associated with more permissive integrity judgments; in ordered logistic regression, frequent use (OR = 0.669) and perceived learning support (OR = 0.581) lowered odds of stricter fraud judgments.
4. Perceived Creativity reduction drives strictness. Students who believed AI reduced their creativity had significantly higher odds of endorsing stricter fraud judgments (OR = 1.498, p < 0.001).
5. Undecided students are engaged, not disengaged. More than a third of the sample (n = 135) were undecided, yet a majority used AI almost always and nearly all reported it helped them understand topics—indecision reflects genuine ethical uncertainty, not inexperience.
6. Independent study time correlated with stricter judgments (OR = 1.174, p = 0.033), suggesting students who invest more in independent preparation value process, effort, and accountability.
7. Perceived dependence, not frequency, predicted negative consequences. Reported academic failure due to AI was strongly associated with perceived AI dependence (χ² = 25.44, Cramer's V = 0.267) but not with frequency of use.

What this means for practice

• Instructors. Replace detection-centered enforcement with authorship-centered guidance: state at assignment level which AI uses are permissible and require disclosure of the tasks delegated to AI, rather than relying on a general prohibition.
• Instructors. Teach non-substitutive, self-regulated use directly, because more frequent AI use (OR = 0.669) and stronger perceived learning support (OR = 0.581) each lowered the odds of students labeling AI use as fraud, while perceived dependence rather than frequency tracked reported academic failure.
• Learners. Treat unresolved disagreement as a reason to ask rather than to guess: 135 of 357 respondents were undecided about the fraud framing while most of them used AI almost always.
• Administrators. Publish assignment-level boundaries, fund AI literacy work around authorship and dependence, and address inequitable access to academic support; read the creativity concern behind stricter judgments (OR = 1.498, p < 0.001) as a teaching topic rather than grounds for a ban.
• Researchers. Test whether disclosure templates, oral defenses, or process-based assessment reduce ethical uncertainty among undecided students, using mixed-method or longitudinal designs instead of another single-campus survey.

Limitations

• Single site: the 357 undergraduates came from 14 programs at the Manizales campus of one Colombian public university, and the authors present the findings as context-sensitive rather than a universal model of student ethics.
• All measures are self-reported and the design is cross-sectional, so the results are associations rather than causal effects; social desirability may have led some students to underreport practices they consider questionable.
• Integrity was measured with a small number of direct perception items rather than a validated questionnaire, including a single item on whether AI use is fraud, and the dataset held no scenario-based AI cases or disclosure-practice items.
• The survey captured declared perceptions, not observed AI behavior, so the study cannot show how stated judgments correspond to students' actual AI-use practices.

Citation

López-López, E. M., Bru-Cordero, O. E., & Correa-Álvarez, C. D. (2026). Academic integrity in the age of AI: University students' study practices and ethical judgments. Trends in Higher Education, 5(2), 49.