I guess that’s on you if you’re asking it something like “tell me yesterday’s news”. No matter your feelings on AI our current LLMs are indisputably a great tool for sending emails and summarizing large text as a draft. If you’re taking the output and running with it and not relying on any other external sources or proofreading then I could see how someone could come to the conclusion it’s 100% terrible awful.
sounds like the internal phishing service should be the one needing to do training