There are two kinds of organisations failing at AI governance right now. They look nothing alike. They have the same disease.
The first type moves fast. Executives arrive from conferences carrying vendor commitments. Business units are already three months into codifying core processes into large language models. Nobody called architecture. Nobody called legal. Nobody called anyone, because speed was the point and governance was the obstacle.
Then the audit arrives. Or the board asks a question. Or a vendor relationship surfaces in a risk register it was never supposed to appear in. And the room goes quiet in a very specific way — the quiet of people who have collectively decided they cannot remember anything.
That is not ignorance. That is obfuscation with a clean conscience. The amnesia is wilful. The vendor meetings happened. The decisions happened. The deployments happened. What did not happen was accountability — so accountability gets retroactively removed by the simple act of not mentioning it.
And the people positioned to close that gap — heads of architecture, practice managers, governance leads — are captured by the momentum. If it is not formally in my purview, I am not responsible for knowing it exists.
This is how organisations accumulate AI debt they cannot audit, cannot quantify, and cannot remediate — because the first requirement of fixing a problem is admitting you have one.
Data-driven decision-making in an AI-first world requires that organisations know what is actually running inside them. Not what has been approved. Not what appears in the architecture register. What is actually running.
Amnesia is convenient. It is also compounding. When it finally closes — and it always closes — it closes badly.