“Recent AI experiments demonstrate an implicit drive for self-preservation in AI models.”