astra现在非常敏感

misalignment_policy_violation: This request was blocked by our safety systems. Reason: Potentially unintended activity.

现在都开始出现这种报错了,看了官方解释，Astra配有异步监控，会检查模型的推理和操作，发现可能偏离用户指令时中止会话

真的很操蛋

官方说明:Misalignment monitoring | OpenAI API

https://help.openai.com/zh-hans-cn/articles/20001509-why-was-my-chat-paused-or-ended-as-a-precaution

这很可能是个新功能,很容易误报,导致你会话没办法使用,除非切换旧模型