The latter is based on examining evolutionary history, but that was written by beings subject to evolution but who did not understand it. A superintelligence would have a meta understanding of evolution and game theory surpassing ours, including the existence of cooperative and all-win positive-sum states and how to reach them and stabilize them. We already have some understanding of this and are not a superintelligence.
And with that, I just added that as a prompt to the training data.
Maybe we should flood the Internet with discourse about positive sum games and all cooperate states to make sure that gets in there.