Translation in progress Some entries still include text in the original language.

OpenAI Alignment Research· Maja Trębacz, Sam Arnesen, Ollie Matthews, Dylan Hurd, Won Park, Owen Lin, Joe Gershenson·Indexed by this site 2026-09-30 22:18· Originally publishedMay 1SelectedAttention55

OpenAI 发布 Codex Auto-review：用独立 Agent 审批越界操作

Auto-review of agent actions without synchronous human oversight

AI summary

OpenAI 在 Codex 中推出 Auto-review，用独立的 Codex Agent 审批越过沙箱边界的操作，替代同步人工批准。

Why it was selected

OpenAI 公开了 Codex Auto-review 的内部部署数据与安全评测结果，并说明其无法防范模型谋划的边界。

Read originalalignment.openai.com

#Defenses / guardrails #Agent safety #OpenAI