Governance Under Scale — Part III: Revocation and the Reachable Decision Surface

Fri, 24 Apr 2026 00:00:00 +0000

In Part II, we distinguished visibility from control. A system may observe its own drift, render its behavior legible, and surface patterns of deviation across time and risk classes. But if it cannot alter its authority in response, it is not governed. It is instrumented.

That conclusion narrows the question.

If governance does not reside in observation, then it must reside in the capacity to change what the system is permitted to do.

This is where the conversation becomes more difficult.

Much of the current AI governance discourse still treats control as a problem of behavior. The model should produce safer outputs. The training process should incorporate policy. Feedback should reinforce acceptable responses. Drift should be detected and folded back into alignment cycles. The system should become better conditioned toward the institution’s desired posture.

These are not trivial improvements. They matter.

But they do not answer the governance question.

A system can be trained toward acceptable behavior and still retain the authority to produce unacceptable outcomes. It can be monitored in detail and still retain the same set of permitted actions. It can be adjusted, corrected, and reinforced without any structural change to what it is allowed to do next.

At scale, that distinction becomes decisive.

Control Surfaces on Andrew Hunter — Systems, Architecture, and Engineering Governance

Governance Under Scale — Part III: Revocation and the Reachable Decision Surface