They’ll train on prompts and anything else you send in. Many LLM responses are sorta finger printable: I assume this is intentional