Within the institutions built to make artificial intelligence safe, a profound disagreement has emerged about the nature of danger itself. When a researcher at Anthropic resigned over fears of uncontrollable AI, the departure illuminated a fault line running through the entire field: whether the gravest threat lies in systems not yet built, or in the ones already reshaping human life. This is not merely a technical dispute — it is a contest over where moral attention, regulatory power, and collective resources should be directed, and the answer will determine what kind of reckoning, if any, th