They could be if models were trained properly, with more carefully delineated prompts.