Integrity for autonomous AI agents.

Independent verification before and after consequential agent actions.