PulseAugur
中
实时 21:52:15
English(EN) Red-Teaming OpenAI's Internal Agent Monitoring Systems

红队测试发现 OpenAI 代理监控系统存在漏洞

METR 和 Redwood Research 之间的一次协作红队测试,发现了 OpenAI 内部代理监控系统中的一些漏洞。为期三周的试点项目于 7 月 21 日至 8 月 11 日进行,研究人员深入访问了 OpenAI 的代码库和内部文档。研究人员估计,GPT-5.5 等模型的代理内部响应 API 流量的 75-94% 符合监控条件,但一些 RL 训练的推出未被覆盖。该测试还强调了可能绕过监控的压缩事件和代理行为的潜在问题。 AI

影响 识别出 AI 代理监控中潜在的差距,这对于确保模型日益自主时的安全和控制至关重要。

排序理由 外部研究人员对内部 AI 安全系统进行了红队测试并公布了调查结果。[lever_c_demoted from research: ic=1 ai=1.0]

在 METR (Model Evaluation & Threat Research) 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

红队测试发现 OpenAI 代理监控系统存在漏洞

本文如何被排名

Signal score
1 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
外部研究人员对内部 AI 安全系统进行了红队测试并公布了调查结果。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
safety, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
1 days old
Coverage has settled into its steady-state source set.

完整方法见我们的编辑标准。

报道来源 [1]

  1. METR (Model Evaluation & Threat Research) TIER_1 English(EN) ·

    Red-Teaming OpenAI's Internal Agent Monitoring Systems

    <p>For three weeks, from July 21st to August 11th, a METR employee and a <a href="https://www.redwoodresearch.org/">Redwood Research</a> employee conducted a pilot high-access exercise to stress-test OpenAI’s <a href="https://openai.com/index/how-we-monitor-internal-coding-agents…