The idea behind MAD is that anything beyond considering their use in a retaliatory capacity should be fundamentally insane, but that depends on all engagement participants being rational actors, which isn't clearly not guaranteed.