Given how financial services can impose silent inexplicable lifetime bans for using the wrong words in the "what is this transaction for" field, I'm wondering at what point the AI automatically reports people for sanctions violation based on its mishearing.