As artificial intelligence (AI) companies increasingly seek access to archival collections for model training, archival organizations face high-stakes decisions with limited precedent and no shared standard. The University of Virginia (UVA) Library has developed the UVA Archival AI Protocol (UVA AAIP), a practical framework grounded in a core rule: irreversible AI models do not get access to archival materials unless item-level provenance and meaningful attribution can be demonstrated, and the institution retains contractually enforceable control to stop further use. The Protocol distinguishes between retrieval-based AI systems—which keep source materials under institutional control and are generally permitted—and general-purpose model training, which absorbs knowledge into model weights irreversibly and is blocked by default. Built on three foundational pillars (provenance and attribution, donor and community responsibilities, and institutional control), the Protocol provides a decision framework for evaluating AI requests, sample contract clauses for deeds of gift and vendor agreements, minimum provenance standards for AI-generated citations, and a phased implementation plan designed for organizations of any size. This briefing addresses both the strategic rationale for the Protocol and the practical realities of putting it into action. Attendees will leave with a freely available adoption kit—including customizable templates, sample clauses, and implementation checklists—that any archival organization or memory institution can use to establish a principled, consistent position before the next AI partnership request arrives.
https://doi.org/10.18130/5dqf-9w86
