The principle of least privilege ensures agents operate with minimal permissions — enough to complete their task, but not enough to cause harm. It’s the foundation of safe, scalable AI deployment.