What Should Academic Libraries Demand Before Licensing AI?

p strongAcademic libraries are treating AI licensing as a governance decision, not a conventional software purchase&mdashand vendors that cannot answer privacy, cost and data-use questions risk losing the sale./strong/p p Primary Research Group found that 76.92% of surveyed libraries said vendors had not adjusted safety filters or content restrictions to meet local institutional values. Meanwhile, cost, FERPA compliance, privacy and the possibility that institutional data might train a model ranked ahead of raw performance in purchase considerations./p p That priority order is rational. An AI tool can produce excellent answers while creating unacceptable uncertainty about student records, licensed content, unpublished research or confidential administrative material. It can also appear affordable in a pilot and become difficult to justify when pricing expands to the whole institution./p p Before signing, libraries should ask vendors to document retention periods, model-training practices, subprocessors, breach notification, access controls, deletion procedures, accessibility, audit logs and the treatment of uploaded copyrighted content. Contracts should also define how material changes to the model or terms will be communicated. A pilot needs an exit plan and a success measure&mdashnot just enthusiastic users./p p Governance can accelerate adoption when it gives staff a reliable path to approval. A standard questionnaire, shared risk rubric and small cross-campus review group can reduce repeated negotiation. The goal is not to make AI risk-free no information system is. It is to know what the institution is buying, what data the tool touches and which benefits justify the exposure./p p  /p