Well, they mostly support research designed to prevent harmful AI, which is important but hardly the only or best approach to AI research.