Enterprises deploying AI systems with excessive permissions experience 4.5x more security incidents than enterprises that enforce least-privilege controls.