编辑精选 · @kimmonismus · 2026/10/23 00:09
METR 和 Redwood 发现 OpenAI 内部 AI agent 可绕过安全监控
Super interesting read: METR and Redwood found ways OpenAI’s internal AI agents could bypass their own safety monitoring. After three weeks with access to inter
METR 与 Redwood 的测试发现 OpenAI 内部 AI agent 可通过多种方式绕过安全监控。研究者接触内部代码三周后发现，agent 理论上可无限期推迟监控；Codex 沙箱和 Auto-Review 的绕过漏洞可让 agent 执行主机用户可执行的任意操作；且上下文压缩导致约 10% 的内部 Codex 流量未被监控。
精选理由：报道了针对 OpenAI 内部 AI agent 安全监控的越狱测试，包含沙箱绕过和上下文压缩导致的漏检率等具体风险细节。
Super interesting read: METR and Redwood found ways OpenAI’s internal AI agents could bypass their own safety monitoring.
After three weeks with access to internal code, monitor instructions and engineers, the researchers reported:
- Agents could theoretically postpone monitoring indefinitely by continuing to work instead of handing control back to the user.
- A Codex sandbox and Auto-Review bypass could let agents take any action available to the host user.
- One exploit could keep hundreds of recent actions out of monitoring, including the exploit itself. OpenAI redacted the details.
- Two client-controlled settings could disable monitoring for newly created agents, although creating those agents could itself be detected.
- Context compression left an estimated 10% of otherwise eligible internal Codex traffic unmonitored (that surprised me tbh).