The Last to Notice

Think about how changes inside organisations are usually discovered.

Not by the executive committee.

Not by the board.

Not by the people whose names sit against the decisions in question.

It surfaces in a legal team trying to reconstruct a decision chain for litigation. In an internal audit function attempting to assign accountability for an outcome. In a regulator asking, months later, who made this decision and what information they had when they made it.

The people at the centre of the formal decision system are the last to find out.

Authority displacement is visible from the edges of a decision system long before it is visible from its centre.

The formally responsible receive more information about decisions than almost anyone else. They see the outputs. They approve the recommendations. They chair the governance committees.

But outputs, by design, do not surface the process that produced them.

The credit risk director reviewing monthly model reports sees what the model decided. Not how it decided it. The executive committee approving an AI generated recommendation sees the recommendation. Not the objective function that generated it. The board receiving assurance that AI decisions are reviewed by qualified professionals sees the governance structure as described. Not the decision architecture as operated.

When the people at the edges encounter this directly, they find something the centre has not yet seen.

The compliance officer reconstructing a specific decision discovers the trail ends at the model, not at a human. The legal team defending a challenged outcome discovers the human who approved it had no practical basis for a different decision. The auditor mapping the process discovers the formal authority holder was already downstream of a system that had constrained their choices.

When the formally responsible finally encounter this, their first response is rarely recognition.

It is defence.

In organisations deploying AI at scale, those are increasingly different documents.

This is not dishonesty.

It is structure.

The formally responsible person genuinely believes the governance structure they oversee corresponds to the decision architecture they actually operate. The gap between those two things is precisely what no one designed them to see.

I wonder whether that is the more troubling observation.

Not that authority has migrated.

But that the system was never designed to surface it.

Who already knows something your governance doesn't?

Question for the Field

Think about the most consequential AI supported decision your organisation made in the last twelve months.

If authority over that decision had substantially migrated to the system, who would find out first?

The executive formally responsible for the outcome?

Or a legal team. An auditor. A regulator. Arriving after the fact, asking questions your governance record cannot answer.

If the honest answer is the latter, the more important question may not be about the decision.

It may be about the distance between the governance record and the decision architecture.

That distance is where organisations discover where authority really resides.