Where AI governance actually starts (it isn’t the model)

Most AI governance conversations jump straight to the model: which one, how big, how it was trained. In practice the decisions that determine whether a rollout becomes a real program or a series of demos are made in the boring middleware layer underneath it.

Governance is a data problem first

Before a model can be trusted to act, the data it reads has to be aggregated, cleaned and described. Lineage, access control and quality are governance. Skip them and every model output inherits the ambiguity of the data behind it.

Start with the plumbing: where data comes from, who is allowed to see it, and how you would explain a given answer after the fact. Get that right and the choice of model becomes an implementation detail you can revisit at will.