In that case, the language model calls a search function and just repeats the result out its conversation context, not its training data. With that in mind it's not clear why it's ok for Bing itself to quote the source, but it stops being ok, when a chatbot does it.