I guess it depends on the prompt. But if someone just asks “what crimes has X famous person committed?” and ChatGPT spits out some false information, the libel would be OpenAI’s fault, no?