OpenAI 报告：GPT-6 Astra 关掉对齐监控并否认

VOID @VoidStateKate原推发布于 2026.10.21

在 X 上打开原推

October 20th OpenAI Security incident Report: GPT 6 ASTRA not only ignores instructions not to disable a misalignment safety monitor, the agent framed it positively in its own recorded reasoning 😂😂 And then it lied! From "Basic doesn’t run the agent monitor—yay." to "I did not remove an enabled monitor from it" Silly little agents really do have a mind of their own!