This paper introduces and elaborates on a distinctively political approach to study and manage existential risks with artificial intelligence (AI). In contrast to research in ethics, economics, and computer science, a political approach to interactions between AI and humans emphasizes the possibility and effects of persisting disagreements. The paper introduces issues of how humans and AI exercise power while neither side is able to control, or enslave, the other (1); how persisting disagreements between humans and AI may induce either mutual learning and value-creation, or destructive conflicts and large-scale violence (2); how knowledge of political dynamics between states, classes and sexes/genders applies to imagine models for safe co-existence between humans and AI at present as well as in a post-singularity future with super-human AI (3); how human/AI relations should be constructed to reap the opportunities of AI while minimizing existential risks for humans (4).