Governing Artificial Intelligence in University Mental Health Services: A Population Health Intelligence Framework for Preventing Algorithmic Stratification

Abstract Background Universities worldwide are adopting artificial intelligence (AI) tools for student mental health triage, risk prediction, and counselling allocation. These tools represent a specific application of Population Health Intelligence (PHI); the use of multi-source data and analytics for surveillance, resource allocation, and policy decision-making. However, AI mental health tools may produce algorithmic stratification: the systematic sorting of students into unequal tiers of visibility, risk, and care. University populations present unique PHI governance challenges including developmental transitions, heightened privacy expectations, and the tension between institutional duty of care and student autonomy. Objective To develop the SAFE-STUDENT governance framework for AI mental health tools in university settings, addressing the specific PHI governance risks of student populations and preventing algorithmic stratification while respecting student privacy, autonomy, and developmental needs. Methods We conducted a targeted conceptual synthesis of literature on university mental health, digital health ethics, student privacy, and higher education governance. Building on algorithmic stratification theory and PHI governance principles, we used a four-stage conceptual synthesis: (1) problem identification, (2) operationalisation of four university-specific governance challenges, (3) principle development using criteria of distinctiveness, applicability, and measurability, and (4) translation into stakeholder-specific guidance with safeguards mapped across the PHI-to-decision pathway. Results Four PHI governance challenges salient in university contexts were identified: (1) privacy-surveillance tension, (2) developmental considerations in student-AI interaction, (3) the digital native assumption and its equity implications, and (4) institutional duty of care versus student autonomy. The SAFE-STUDENT framework comprises 11 principles (Student-Centred, Accountability, Fairness Monitoring, Escalation, Safety, Transparency, User Agency, Data Privacy, Equity, Non-Replacement) tailored with implementation indicators mapped across seven PHI stages, data generation to outcomes and stakeholder-specific guidance for seven stakeholder groups. Failure points and equity signals are mapped across the PHI pathway. Conclusions General AI governance frameworks are insufficient for university mental health. SAFE-STUDENT provides a tailored governance mechanism for equitable intelligence generation in university digital health settings. Empirical validation via a three-stage research agenda is required.