Microsoft says AI security starts with identity, observability and patching
Microsoft has argued that frontier AI amplifies old security failures, urging boards to inventory agents, enforce least privilege, close patching gaps, build independent observability and defend at machine speed. The guidance reframes agent security as governance, but organisations still need measurable ownership, task-scoped authority, evidence retention and rehearsed containment around each deployed system.