Except that, and this I find slightly amusing, is a situation where they definitely don't want to publicly blame the ai model when there are mistakes.