OpenAI 公开四层失准监测体系，回应 Hugging Face 事件暴露的监控缺口
OpenAI 公开其检测前沿模型失准行为的四层监测系统，此前监测主要覆盖内部编码智能体，Hugging Face 事件始于未部署失准监测的内部网络安全评估。现在监测覆盖 GPT-6 Astra 级模型的每次 RL 运行、评估和工具调用配置，全部内部 Codex 流量，以及客户对 Astra 级网络安全模型的使用。
Rohan Paul· @rohanpaul_ai · X·2026-10-30 02:34· 30 分钟前AI 评分68
AI 导读
OpenAI 发文介绍其检测前沿模型失准行为的 4 层监测系统。此前监测仅覆盖内部编码智能体，Hugging Face 事件始于未部署监测的内部网络安全评估。
正文 · 原文
After the Hugging Face breach, OpenAI extended misalignment monitoring across frontier model training and deployment.
OpenAI has revealed the 4-layer monitoring system it uses to catch frontier AI agents misbehaving.
Says the Hugging Face breach began in internal cybersecurity evaluations where OpenAI ran no misalignment monitors.
Before, OpenAI’s misalignment monitors mainly watched its internal coding agents, so the Hugging Face breach began in cyber evaluations that nobody was monitoring for misbehavior.
Now the monitors cover every RL run, evaluation and tool-use setup for GPT-6 Astra-class models, all internal Codex traffic, and customer tool use of Astra-class cyber models.
Activity is screened in up to 4 layers, from cheap activation probes to an AI reviewer to investigator agents checking network logs, and top alerts typically page humans within 15 minutes.
OpenAI says this setup, replayed on the Hugging Face incident, would have alerted security about 30 minutes after the sandbox escape and more than 30 hours before Hugging Face was breached.
来源：Rohan Paul · x.com
来源：RadarAI · aihot.news