What privacy risks are there for using LLM models on rented GPUs?

I'm considering using a gpu rental service, but would like to know what information they could potentially have access to.

Can they get the original prompt or do some services log the prompt?

Perhaps openai's api logs prompts?