Policy tells the model how to behave. Approval stops it before risky actions. Audit records what happened.

We don't sell toxicity filters or hallucination detectors we haven't built. The three layers we have are real, and the next slide shows the boundaries of each.