Regulating GenAI in education: theoretical lessons from an empirical case study

Keywords: GenAI, academic integrity, transparency, rules, exams

Synopsis: We present results from a survey of bachelor (BA) students from a major Scandinavian university. The survey probed students’ use GenAI and their understanding of relevant rules. Results showed that the regulation had significant weaknesses. From the case study and existing literature, we suggest general advice on how to design GenAI regulation in education.

Background: Educational institutions are reforming their disciplinary rules to adapt to GenAI. Presumably this process will lead to challenges similar to those documented for other regulations of student conduct, including widespread violations due to students’ limited understanding of grey zones (Goddiksen, et al., 2023; Johansen, et al., 2026), worries among students about being falsely accused of cheating (Goddiksen et al., 2024), and lack of enforceability due to limited detection options (Kangwa, et al., 2025). In addition, there may be new challenges related specifically to GenAI and the novel teaching activities it enables.

To investigate how these challenges play out in practice, we chose a major Scandinavian research university as a case. The university changed its rules on GenAI use for students in Jan 2025. Before Jan 2025, GenAI was banned by default in all exams. After Jan 2025, GenAI was by default allowed. However, GenAI declarations were mandatory, and certain practices, e.g. generating and improving text, were forbidden. We investigated how students who worked on their BA thesis during the spring of 2025 used GenAI and how they understood and complied with the new rules.

Aim: To identify key challenges when regulating GenAI in education and giving advice on how to avoid or overcome them.

Research questions:

Based on a literature study, the case study investigated the following questions:

- Did students use GenAI when preparing their BA thesis and for what?

- Were the students aware of the new rules and did they comply?

- Were students worried about being accused of cheating using GenAI?

- Did the supervisors guide the students` GenAI-use?

Methods: Questionnaire based survey and philosophical analysis. The survey was developed based on previous surveys on students’ understanding of rules and grey zones (cf. above) combined with the authors’ experience from introducing GenAI in teaching and exams. The survey included a list of 13 different tasks (plus ‘other’) that GenAI could be used for when writing a BA project. Four of these tasks (e.g. generating text to be included in an assignment) were explicitly prohibited in the new GenAI regulation. Two tasks, including data analysis, were judged by the authors to be in a grey zone, as their status depended heavily on the specific interpretation of the existing rules. The remaining tasks were coded as compliant. Survey data were further analysed through descriptive statistics, cross-tabulation, and correlation analysis. The survey results were further analysed through a novel theoretical framework combining insights on effective legislation from research in law and academic integrity to help suggest general recommendations.

Ethics: The study was approved by the local ethics committee. Participation in the survey was voluntary and anonymous. Informed consent was secured at the beginning of the survey.

Sample: All 824 students who had handed in their BA thesis to the Faculty of Science at the Scandinavian research university in the spring/summer of 2025 were approached. 131 complete responses were included in the analysis.

Results: 76% of the participants had used GenAI as an aid in their project work. Use spanned from idea generation to editing of the near final text. Of those who had used GenAI, 85% indicated that they had complied with existing regulations, 12% were unsure, and 5% indicated that they knew they had broken the rules. However, a comparison between the participants’ self-declared use of GenAI and the rules showed that 48% of those who had used GenAI had used it for at least one task that was clearly prohibited, and an additional 28% had used it for one or more grey zone tasks. The majority of these participants were confident that they had been rule-compliant.

Of participants who had used GenAI, 17% further indicated that they had deliberately left out some uses of GenAI in their transparency declarations, mainly due to worries about being accused of cheating. Some also found it irrelevant to list everything.

Among all participants, 35% indicated that they had worried about being accused of breaking GenAI regulation. This included 14% of those who did not use GenAI in their project.

Only 44% of the participants had discussed GenAI use with a supervisor. There was no significant correlation between engagement in non-compliant or grey zone GenAI use and discussing GenAI use with a supervisor.

The participants who did not use GenAI in their project indicated general reservations against GenAI and fear of accidentally breaking the rules as the main reasons.

Discussion: The case provides details on some of the challenges related to regulating GenAI in education. Many of the challenges identified were versions of familiar challenges related to explaining and enforcing disciplinary rules. Some challenges, however, had twists that were specific (in strength if not in kind) to GenAI. The lack of effective detection tools, for instance, made it particularly difficult to prohibit GenAI uses that are widely accepted among students and staff, unless substantial efforts were made to explain the underlying reasoning. In such cases, acceptance through explanation thus becomes central to reduce the high rate of unintentional rule breaking, even though it will likely not be sufficient to handle all cases of intentional dishonesty. In this process, our data illustrate that it is important to be sensitive to what uses are generally accepted in the affected groups, but also to ethical and didactical concerns related to GenAI.

Based on our results and general theory from academic integrity and law on assessment of regulation (Goddiksen, et al., 2023; Johansen, et al., 2026; Mousmuti, 2019), we end by presenting ten key focus areas when developing new regulation on GenAI in education.

References:
Goddiksen, M., et al. (2024) The dark side of text-matching software: worries and counterproductive behaviour among European upper secondary school and bachelor students. International Journal of Educational Integrity 20, 15. https://doi.org/10.1007/s40979-024-00162-7

Goddiksen, M., et al. (2023). Grey zones and good practice: A European survey of academic integrity among undergraduate students. Ethics & Behavior, 34(3), 199–217. https://doi.org/10.1080/10508422.2023.2187804

Johansen, M. et al. (2026). Academic integrity across educational levels: Exploring students` engagement with grey-zone and non-compliant practices in four European countries. PLoS ONE 21(3): e0342227. https://doi.org/10.1371/ journal.pone.0342227

Kangwa, D., Msafiri, M. & Fute, A. (2025). “Exploring the Factors That Promote a Balance Between Academic Integrity and the Effective Use of GenAI Tools in Higher Education: A Systematic Review.” Journal of Computer Assisted Learning 41, 5: e70109. https://doi.org/10.1111/jcal.70109.

Mousmouti, M. (2019). Designing effective legislation. Northampton: Edward Elgar Publishing.