Building an evidence layer for AI

Organisations can prove that an employee used an AI system, but can they reconstruct what it was asked, what it produced and who relied on it?