A new zero-trust paper tested an agent-based enforcement design against a simple rule-based baseline. The baseline was more accurate and decided in under a millisecond. Here’s what that says about where AI belongs in enforcement.