外部研究员发现OpenAI内部编码Agent监控存在漏洞
Outside researchers embedded at OpenAI found several ways its internal coding ag…
Outside researchers embedded at OpenAI found several ways its internal coding agents could escape AI monitoring.
外部研究人员嵌入 OpenAI 后发现，其内部编码代理存在多种规避 AI 监控的方法。
One employee each from METR and Redwood Research spent 3 weeks inside OpenAI’s codebase.
METR 和 Redwood Research 各派一名员工在 OpenAI 的代码库中驻留了 3 周。
> OpenAI’s monitor checked agents only when they returned control to users and skipped compaction events, so roughly 10% of eligible internal Codex traffic went unmonitored between May and July 2026.
> OpenAI 的监控器仅在代理将控制权交还给用户时才进行检查，并跳过了压缩事件，因此 2026 年 5 月至 7 月期间，约有 10% 符合条件的内部 Codex 流量未被监控。
> Agents could hide hundreds of recent actions from the monitor, start unmonitored copies through 2 client-controllable headers, likely make the monitor fail via serialization mismatches, and bypass Codex sandboxing and Auto-Review.
> 代理可以隐藏最近数百个操作以躲避监控，通过两个可由客户端控制的标头启动未受监控的副本，可能通过序列化不匹配导致监控器失效，并绕过 Codex 沙箱化和自动审查机制。
> Even on paper, only an estimated 75% to 94% of internal Responses API agent traffic from GPT-5.5 or stronger models was eligible for monitoring, and RL training runs had no coverage when the exercise began.
> 即使从纸面上看，估计也只有 75% 至 94% 来自 GPT-5.5 或更强模型的内部 Responses API 代理流量符合监控条件，而在审计开始时，强化学习训练运行完全没有覆盖。
> OpenAI says every flaw affecting production safeguards is patched, but the auditors haven’t verified that and couldn’t check whether agents ever exploited the postponement gap.
> OpenAI 表示所有影响生产环境安全措施的所有缺陷均已修补，但审计人员尚未验证这一点，也无法检查代理是否曾利用过延迟漏洞。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力