I’ve had multiple occurrences (with the latest models) of AI reading a poorly named db column and assuming it was one thing when it was another that resulted in a problem. It does not know when to stop or when to continue with determining what is true.