Generative AI rarely fails loudly. It fails fluently. The real risk is not hallucinations, but persuasive outputs travelling through workflows without ownership, controls, or escalation paths. This article introduces a Decision Accountability model to govern AI as a fallible participant in your operating model.