We need institutions capable of using AI while preserving transparency, recourse, and a human account of consequential decisions.