When old data brings AI rollouts to a screeching halt - and how to manage it

AI deployments at large enterprises (e.g., Fidelity, EY) unexpectedly surfaced vast stores of unstructured data (SharePoint, NAS, PDFs), prompting temporary halts and a focus on data ownership, labeling, geo- and contract-aware governance, access controls, telemetry, and lifecycle management; the article highlights using AI to assist labeling, the risks of shadow AI, and the unresolved problem of assigning identity to short-lived agents.