Also, maybe some services are doing "machine learning" training with user data, but it is the first time I see recent LLM service saying that you can feed your data to human reviewers at their will.