Generative AI in Programming Education: Students’ Ethical Judgements Across Five University Sites

Abstract

Background and Context: GenAI can support explanation, debugging, and personalised practice in programming courses, but it can also generate code that substitutes for the programming performance that assessments are intended to evidence. Computer science educators therefore need evidence about how students judge different forms of GenAI use in assessed programming work.Objective: This study examines students’ ethicality and perceived institutional-rule violation judgements of GenAI use in programming coursework. It examines whether judgements vary with the amount of AI-generated code, demonstrated understanding, stage of GenAI use, and university site, and whether the two outcomes show similar or divergent patterns.Method: A multi-site scenario-based survey was completed by 196 students in Botswana, Brazil, Canada, Indonesia, and South Korea. Participants evaluated eight programming scenarios using two five-point items. Responses were analysed using descriptive summaries, Kruskal-Wallis tests, one-vs-rest Mann-Whitney U contrasts, and site-specific ordinal logistic regression.Findings: Judgements varied across scenarios and differed significantly across sites for both outcomes. Across all five sites, amount of AI-generated code was the only scenario characteristic consistently and positively associated with both ethical concern and perceived rule violation. Demonstrated understanding and stage of use showed weaker and less consistent independent associations. Ethicality and rule-violation judgements showed broadly similar but not identical patterns.Implications: Students’ judgements were most consistently associated with the extent to which AI-generated code was incorporated in the assessed artefact. Programming educators should make expectations for generated code explicit and design assessments that make student reasoning visible while supporting pedagogically useful forms of GenAI.