A library consortium deploys shared AI assistants across city branches with unclear who approves changes. What program artifact should security establish?

A library consortium deploying shared AI assistants across city branches with unclear change approval needs a common AI security RACI that defines roles and responsibilities across the multi-site civic deployment. The RACI clarifies who approves prompt changes, who monitors outputs, who owns incidents, and who accepts residual risk—preventing unilateral edits and orphaned duties that confuse residents. A centralized AI asset inventory that catalogs model ownership across the multi-site civic deployment only lists what models exist and who owns them and does not establish who approves changes before production, so it leaves the approval gap unresolved. A common AI risk register that documents shared model risks across the multi-site civic deployment tracks identified risks rather than who approves changes across the consortium, so it cannot close the change-authority gap. A shared AI control baseline that standardizes security requirements across the multi-site civic deployment is useful for setting consistent requirements but does not define prompt-change approvers or residual-risk owners, so it is not the change-governance artifact this deployment needs. Exam caveat: the RACI must cover vendor-managed components and shared platforms, not only local branch staff, so accountability spans the consortium. Operational check: publish the RACI with named roles per site and consortium-level functions, require change tickets referencing the accountable approver before production prompt updates, and test an incident page-out against the documented responsible party.