Automatically review and clean AI training datasets to prevent models from learning harmful response patterns.