“There is a serious scientific hypothesis that an AI's intelligence level is independent of its goals, meaning higher capability does not guarantee benevolent behavior.”