Beyond Detection: How Students Use—and Hide—AI in Online Assessments and What Authentic Tasks Can Do About It

Abstract As AI tools such as ChatGPT and CoPilot become increasingly common in higher education, universities must reconsider how assessments are designed, monitored, and supported. This small case study investigates how students use AI in online assessments, whether they disclose such use, and how ethical concerns shape their behaviour. Based on a targeted survey of undergraduate economics students, representing about 18% of the cohort (31/174), we find that only about one-third reported using AI tools, this figure is lower than those reported in several larger surveys. Most reported uses were supportive tasks such as rewording or idea generation. Some students appear to opt out early, suggesting a strategic decision to avoid scrutiny. Fear of penalties is widespread, and exploratory modelling suggests that students with greater ethical concerns may be less likely to use AI at all. At the same time, students express support for guidance and structured regulation. Many favour citation rules and believe AI can be used ethically. Real-world, data-based tasks are widely seen as a way to reduce misuse of AI. The way to tackle the negative learning effects of AI is not by eliminating AI, but by encouraging meaningful engagement. We conclude that students are navigating institutional ambiguity with caution and pragmatism. Overall, our conclusions are preliminary and exploratory: findings are not generalisable, but they point to promising directions for assessment design. Rather than relying on detection and deterrence, universities may achieve better outcomes by aligning assessments with authentic tasks and clear expectations—and by addressing fairness about all students’ access and use of AI.