Can anybody give me some suggestions for how to design an AI system whose sole purpose is to safeguard humanity against rogue AIs? I'd like for it to be able to run locally