Orphaned AI agents in enterprise networks pose significant security risks because their authorization and access rights are often undocumented and not traceable.