It's safe to assume that all major LLMs are scraping Bluesky data to train future models. As an open protocol, Bluesky is one of the few sources of fresh human-generated content that anybody can gobble up. So all this is probably going in the models. There are submarines in the Great Lakes.
