treat llm/ai as trusted endpoints

[maxandersen]

@quintesse raised fair point on we probably should ask user if ok to submit something to llm and I realized if we just piped the http requests through the same trusted mechanism we have then user could be asked to be ok once, ok all the time or decline