Universities are moving from pilots to routine human-AI collaboration in teaching, learning, and student support, including feedback drafting, tutoring, and case routing. When AI participates in routine judgment, the central issue becomes governance: who may rely on AI outputs, how that reliance is reviewed, and what recourse exists when outputs are wrong or contested.

I develop an auditable OpenAlex evidence map of scholarship on human-AI collaboration in higher education (2013–2025), complemented by a focused generative-AI sub-corpus (2022–2025). Two query packs retrieved 43,329 records, deduplicated to 40,229 unique works. A seedless machine-assisted triage combines rule-based cues with an ensemble text model to estimate relevance and uncertainty during title/abstract screening, yielding a query-stratified core analytic set (top 5%, n=2,012) for synthesis.

The map indicates sharp post-2022 growth, a concentration of high-scoring work in surveillance / platformization and in policy / ethics strands, and a dominant framing of AI as a transactional tool rather than a relational co-agent. I translate these patterns into three vendor-agnostic university AI-governance capabilities (accountable delegation, epistemic stewardship, and consentful data stewardship with duty-of-care boundaries) intended to be assignable, teachable, and auditable as AI becomes routine infrastructure.