Add in that if many people are using the LLM side of things to ask about their health, that's data that can get bundled in, used in the ML algorithm, and used against you.

I know it sounds very tin foil hat, but after working with US healthcare companies... well, tin foil hats are appropriate.

... it's another thing to have clean, structured data that could be sold to health insurance companies and fed into a machine learning algorithm to determine what health conditions you are likely to develop and deny you coverage because it's a "preexisting condition"