AI chatbots’ content rules often frustrate users, study finds. But those “guardrails” can also go awry — and quite often do, as a new academic study documents. There are good reasons for AI companies ...