According to openai my cusomized agent promotes “violence” 🤣 I have an ai I can use offline and use other LLMs but it begs the question: How do such companies decide this?