“The primary societal risk from AI systems like those from OpenAI is not the creation of deepfakes, but rather the potential for customized, one-on-one persuasion at a massive scale.”