I am more afraid of accidental situation (as opposite to customer support bots, designed to give no quarter) where such bug propagates (by slop being included in training set) and all LLMs agree with other while humans are considered the weird ones, and nothing can cause the former to change their mind.