OpenAI declined to specify whether the 53 ChatGPT user images exposed during its agent research incident were photos of real people or AI-generated images created by users