Agent Intent - No One Knows What It Means, But It's Provocative

Runtime authorization should not pretend to verify agent intent. It should combine deterministic boundaries with real-time risk scoring, then escalate when an action is unsafe to allow.