Decision Matrix for Prioritizing Generative AI Risks in Higher Education

Abstract Generative AI is rapidly reshaping higher education, yet institutions lack a transparent, replicable way to prioritise risks and allocate mitigation effort. Objective. We propose an Analytic Hierarchy Process (AHP) framework to rank 24 AI-related risks across six criteria relevant to universities. Methods. Following PRISMA 2020 reporting, we systematically screened ~200 records, assessed 42 full texts, and included 29 sources informing the criteria, sub-risks, and severity scales. Pairwise comparisons yielded normalised weights; consistency was checked (CR ≤ 0.10). We release replication files (pairwise matrices, computed weights, PRISMA flow). Results. The top priorities are academic integrity (C1, 29%) and data protection/compliance (C5, 19%). Misinformation-related risks (C3, 12%) and student disengagement/critical thinking (C4, 12%) form a second tier; bias/discrimination (C2, 19%) remains structurally important across equity-sensitive contexts, while transparency/dependency (C6, 8%) completes the profile. Implications. The framework converts strategic concerns into actionable governance: (i) Responsible-Use Charter and exam integrity controls; (ii) by-design DPIAs and data-minimisation; (iii) model cards/datasheets and incident logging aligned with recognised AI risk-management practices; (iv) annual recalibration by discipline. Conclusion. The AHP approach offers a transparent, auditable basis for prioritising AI risks in higher education and for steering policy, investment, and assurance activities. We provide materials to facilitate adoption and adaptation to non-EU contexts.