<p>Despite the rapid uptake of generative AI (GAI) tools in higher education, there is limited empirical understanding of how students balance perceived educational utility with ethical concerns such as privacy, bias, and overreliance, including risks of cognitive offloading and weakened epistemic authority. This study provides a cross-regional comparison of students’ perceptions and use of GAI by focusing on Europe and Australia, two higher education contexts shaped by contrasting policy and institutional cultures, with Europe often characterised by stronger regulatory attention (e.g., data protection and academic integrity governance) and Australia by comparatively pragmatic, implementation-oriented approaches. Using a cross-sectional mixed-methods design, we surveyed 788 university students (Europe <i>n</i> = 402, Australia <i>n</i> = 386) via an online questionnaire and analysed the data using ANOVA and chi-square tests, complemented by qualitative content analysis of open-ended responses. Across both regions, students reported strong perceived benefits, most notably accessibility to a fast knowledge base (72.5%) and support for personalised learning (66.4%). Importantly, ethical concerns were widely acknowledged but did not consistently deter adoption, suggesting a pattern of risk normalisation and pragmatic prioritisation of efficiency. However, the results also reveal a statistically meaningful nuance: privacy concerns were more clearly linked to willingness to adopt among Australian students, whereas concerns about bias and dependence showed weaker or non-significant relationships with adoption. Qualitative responses help explain these patterns by indicating that students often recognise ethical risks but accept them as manageable trade-offs when academic productivity gains are salient, while also referencing regional differences in integrity and data-protection expectations. The findings provide actionable implications for higher education institutions and policymakers by supporting the need for structured, multidimensional AI literacy initiatives that integrate technical, ethical, and critical competencies, and by highlighting the value of context-specific guidance aligned with distinct regulatory and pedagogical cultures. Limitations include self-reported measures, convenience sampling with disciplinary overrepresentation, and aggregation of diverse European countries into a single regional category, which constrain generalisability and call for longitudinal and more differentiated cross-cultural analyses in future work.</p>