Students who viewed artificial-intelligence rules as unclear, AI cheating as easy to hide and university guidance as limited were more likely to report using AI to break academic rules, according to a survey of 863 undergraduates in Vietnam.
The study, conducted by Hanh Van Nguyen and Mai Thi-Thuy Duong of Hanoi University of Science and Technology, found an association rather than proof that unclear policies cause cheating. It surveyed students anonymously online from late October to late November 2025, and was published in PLOS ONE in September 2026.
What the survey found
Students rated 11 possible conditions related to AI-assisted cheating. Statistical analysis grouped them into two broad clusters. The first combined unclear ethical boundaries, AI capabilities and gaps in university policy. The second included academic pressure and comparison with classmates.
The first cluster had the strongest relationship with reported cheating. Students who viewed unclear rules, easy-to-use or difficult-to-detect AI, and weak institutional guidance as more serious problems reported more AI-assisted cheating. The association was statistically significant and had a coefficient of β = 0.228, with p < 0.001.
By contrast, academic pressure and peer comparison were essentially unrelated to cheating on their own. After the factors were modeled together, the pressure cluster showed a small negative coefficient. The researchers cautioned that this result should not be interpreted to mean stress prevents cheating or that pressure is protective.
Reported behavior was below the midpoint
Students rated 10 behaviors on a scale from 1, meaning “never,” to 5, meaning “very often.” Every behavior had an average below the midpoint of the scale.
- Paraphrasing AI-generated text to avoid detection was the most common listed behavior, averaging 2.42.
- Using AI to complete an entire assignment averaged 2.24.
- Using AI to answer exam questions averaged 2.20.
- Submitting AI-generated work as one’s own averaged 2.05.
- Using AI-fabricated references averaged 1.93.
- Using AI to fabricate or manipulate research data averaged 1.66.
- Using AI when an instructor had banned it averaged 1.65.
The remaining behaviors included submitting AI work with little personal input, accepting AI output without checking it, and using AI to take over or rewrite other people’s informally published work.
Why the results need careful interpretation
The survey included 881 responses, but 863 were complete and usable. Participants came from one technical university and were recruited through convenience sampling. The sample was 81.3% STEM majors, 71.3% men and 55.0% second-year students.
Those characteristics limit how confidently the findings can be applied to other universities or student populations. Men reported more AI cheating than women, while non-STEM students reported more than STEM students. The non-STEM comparison involved 161 students and was treated by the researchers as a side finding. Year in school was not associated with reported cheating.
The study was cross-sectional, meaning it captured responses at one point rather than tracking changes over time. Students who cheat may also be more likely to perceive rules as unclear, so the direction of the relationship cannot be established.
Self-reporting is another important limitation. An earlier Vietnamese study found that direct questions detected less than half as much AI cheating as anonymous indirect methods estimated. That suggests the survey averages may not capture the full extent of the behavior, although the exact level of underreporting cannot be determined from this study.
What the findings clarify
The model’s factors explained about 5% of the variation in self-reported cheating. In other words, unclear rules and related concerns were associated with cheating, but they accounted for only a small part of the differences between students.
Students agreed most strongly with statements about pressure: overloaded schedules averaged 3.85 out of 5 and performance pressure averaged 3.81. Yet those ratings did not translate into a clear independent relationship with reported cheating. Academic pressure was also strongly correlated with the unclear-rules cluster, making the separate statistical effects less stable.
Taken together, the results suggest that students’ understanding of permitted AI use and the visibility of institutional guidance may matter when they describe academic-integrity behavior. They do not establish that policy changes alone would eliminate cheating, nor do they show that academic pressure is unimportant in every setting.
AI tools were used to assist with the preparation of this article.
Leave a Reply