AI training-data provenance becomes a buyer question

In the first half of 2026, enterprise procurement teams have started asking model vendors a question that used to be reserved for a small circle of policy staff: what, precisely, was this model trained on, and can you show it?

The reason is neither ideological nor purely legal. A model whose training data cannot be described in writing cannot be defended in writing when a client asks the same question two years later. Vendors that treat training provenance as a first-class deliverable are quietly winning shortlists.