There is no way to know if an LLM API is manipulating you