AI-based classifiers and metaprompting to mitigate potential threats or misuse. Using LLMs may perhaps create problematic content that could result in risks or misuse. Examples could involve output relevant to self-hurt, violence, graphic material, intellectual assets, inaccurate facts, hateful speech, or text that would relate to illegal actions.