Doctrine
The Trust Plane is built on a simple operating claim: as AI systems move from answers to actions, governance has to decide what behavior is allowed, what evidence supports that permission, and how autonomy changes when outcomes change.
The old question:
Do we trust it?
The better question:
What is this system allowed to do?
Under what conditions?
With what evidence?
What changes when the evidence changes?
What changes when outcomes change?
The question is not whether the system seems trustworthy. The question is whether its behavior has earned permission.
The loop
Behavior produces evidence.
Evidence informs governance decisions.
Decisions trigger interventions.
Interventions change outcomes.
Outcomes determine whether autonomy expands, contracts, pauses, or resumes.
Capability creates possibility. Permission changes work. Outcomes prove whether the permission was earned.
The Trust Plane is not a software product. It is an operating model for governing AI behavior.
The object of governance is what the system does.
Trust means a behavior is permitted under defined conditions.
Evidence is what governance can inspect and act on.
Evals produce evidence. Governance decides what changes because of it.
Humans govern when they can see, challenge, and change the system.
When AI can act, authority has moved.
Decisions must be challengeable by people, systems, and outcomes.
A correction that changes nothing is only a cost.
The work must show whether permission was earned.
It expands, contracts, pauses, and resumes based on evidence and outcomes.
Occasional updates on the book, Field Notes, and practical ideas for governing AI behavior in production.