Treating AI features as first-class risk surfaces, rather than extensions of existing ones, forces clarity. Assign clear ...