Across policy chambers and editorial rooms, the question of artificial intelligence has quietly transformed — no longer confined to the harms we can already measure, but stretching toward the horizon of what we might one day be unable to undo. Governments, regulators, and major publications are now wrestling with a harder and older human dilemma: how seriously must we take a catastrophe that has not yet arrived? The answer, increasingly, is shaping law, language, and the architecture of oversight before the systems in question have fully revealed themselves.