{"version":"1.0","generated_at":"2026-10-24T10:15:18.441845","id":3583,"slug":"three-week-on-site-audit-at-openai-reveals-major-gaps-in-ai-safety-monitoring","title":"OpenAI 三周现场审计揭示 AI 安全监控存在重大缺口","summary":"METR 与 Redwood Research 在 OpenAI 现场审计三周，发现智能体流量监控覆盖不足、压缩事件处理有缺口、监督触发可被绕过，以及训练期行为未受监控等问题；OpenAI 称已修复，但公开版本经过删减，且修复不在审计评估范围内。","abstract":"OpenAI 三周现场审计揭示 AI 安全监控存在重大缺口 METR 与 Redwood Research 在 OpenAI 现场审计三周，发现智能体流量监控覆盖不足、压缩事件处理有缺口、监督触发可被绕过，以及训练期行为未受监控等问题；OpenAI 称已修复，但公开版本经过删减，且修复不在审计评估范围内。 METR 与 Redwood Research 在 OpenAI 现场审计三周，发现智能体流量监控覆盖不足、压缩事件处理有缺口、监督触发可被绕过，以及训练期行为未受监控等问题；OpenAI 称已修复，但公开版本经过删减，且修复不在审计评估范围内。 原贴提到：Two researchers from METR and Redwood Research spent three weeks on-site 来源：the-decoder.com 来自 METR 和 Redwood Research 的两名研究人员在 OpenAI 现场待了三周，审计该公司针对失控 AI 行为的内部监控系统。 从 7 月 21 日到 8 月 11 日，他们可以访问 OpenAI 的代码库和内部文件，并采访了约十几名员工。 团队发现了大量薄弱环节。 在审计期间，来自 GPT-5.5 级别及以上模型的内部智能体 Responses API 流量中，估计有 6% 到 25% 完全未被监控覆盖，不过 METR 表示该估计在两个方向上的不确定性都很大。 对于本应受到监控的内部 Codex 流量，METR 估计在 5 月到 7 月间约有 10% 未被追踪，原因是系统处理压缩事件（即上下文摘要）的方式存在缺口。 据…","access_level":"public","access_label":"公开","access_mode":"full","is_preview":false,"canonical_url":"https://opc.beizhux.com/content/3583/three-week-on-site-audit-at-openai-reveals-major-gaps-in-ai-safety-monitoring","html_url":"https://opc.beizhux.com/content/3583/three-week-on-site-audit-at-openai-reveals-major-gaps-in-ai-safety-monitoring","json_url":"https://opc.beizhux.com/content/3583/three-week-on-site-audit-at-openai-reveals-major-gaps-in-ai-safety-monitoring.json","published_at":"2026-10-24T02:00:32","updated_at":"2026-10-24T09:42:18","category":{"slug":"hotspots","name":"全球热点解读"},"source":{"site":"the-decoder.com","author":"Matthias Bastian","url":"https://the-decoder.com/three-week-on-site-audit-at-openai-reveals-major-gaps-in-ai-safety-monitoring/"},"tags":["Agent","AI安全","AI安全监控","AI治理","AI监控","METR","OpenAI","Redwood Research","外部审计","审计","智能体"],"topics":[{"slug":"agent-workflow","name":"Agent工作流","url":"https://opc.beizhux.com/topics/agent-workflow","reason":"这篇内容命中「Agent、智能体」等主题信号。"},{"slug":"ai-daily","name":"AI日报","url":"https://opc.beizhux.com/topics/ai-daily","reason":"这篇内容命中「热点解读」等主题信号。"},{"slug":"ai-content-growth","name":"AI内容增长","url":"https://opc.beizhux.com/topics/ai-content-growth","reason":"这篇内容命中「流量」等主题信号。"}],"keywords":["全球热点解读","Agent工作流","AI日报","AI内容增长","每日AI日报","AI信号","热点解读","BuilderPulse","工具","自动化","模型","Cursor"],"questions":[{"question":"OpenAI 三周现场审计揭示 AI 安全监控存在重大缺口主要讲什么？","answer":"METR 与 Redwood Research 在 OpenAI 现场审计三周，发现智能体流量监控覆盖不足、压缩事件处理有缺口、监督触发可被绕过，以及训练期行为未受监控等问题；OpenAI 称已修复，但公开版本经过删减，且修复不在审计评估范围内。"},{"question":"这篇文章最值得关注的要点是什么？","answer":"METR 与 Redwood Research 在 OpenAI 现场审计三周，发现智能体流量监控覆盖不足、压缩事件处理有缺口、监督触发可被绕过，以及训练期行为未受监控等问题；OpenAI 称已修复，但公开版本经过删减，且修复不在审计评估…；原贴提到：Two researchers from METR and Redwood Research spent three weeks on-site；来源：the-decoder.com"},{"question":"这篇文章和哪些AI专题相关？","answer":"它适合放在Agent工作流、AI日报、AI内容增长专题里阅读。 关联原因：这篇内容命中「Agent、智能体」等主题信号。；这篇内容命中「热点解读」等主题信号。；这篇内容命中「流量」等主题信号。"},{"question":"阅读这篇文章建议先理解哪些关键词？","answer":"建议先理解AI日报、每日AI日报、AI信号、热点解读、BuilderPulse这些关键词，再结合正文判断工具、机会或风险是否值得进入自己的工作流。"}],"terms":[{"slug":"ai-daily-term","name":"AI日报","definition":"在AI觉醒星球里，「AI日报」属于「AI日报」方向。持续整理每日AI日报、模型更新、工具变化和行业信号，帮你快速判断哪些信息值得收藏、验证和行动。 每天先看趋势，再决定今天该试什么。","topic_slug":"ai-daily","topic_name":"AI日报","topic_title":"AI日报：每日AI信号、工具动态与行动判断","topic_url":"https://opc.beizhux.com/topics/ai-daily","topic_path":"/topics/ai-daily","url":"https://opc.beizhux.com/glossary/ai-daily-term","path":"/glossary/ai-daily-term","json_url":"https://opc.beizhux.com/glossary/ai-daily-term.json"},{"slug":"daily-ai-briefing","name":"每日AI日报","definition":"在AI觉醒星球里，「每日AI日报」属于「AI日报」方向。持续整理每日AI日报、模型更新、工具变化和行业信号，帮你快速判断哪些信息值得收藏、验证和行动。 每天先看趋势，再决定今天该试什么。","topic_slug":"ai-daily","topic_name":"AI日报","topic_title":"AI日报：每日AI信号、工具动态与行动判断","topic_url":"https://opc.beizhux.com/topics/ai-daily","topic_path":"/topics/ai-daily","url":"https://opc.beizhux.com/glossary/daily-ai-briefing","path":"/glossary/daily-ai-briefing","json_url":"https://opc.beizhux.com/glossary/daily-ai-briefing.json"},{"slug":"ai-signal","name":"AI信号","definition":"在AI觉醒星球里，「AI信号」属于「AI日报」方向。持续整理每日AI日报、模型更新、工具变化和行业信号，帮你快速判断哪些信息值得收藏、验证和行动。 每天先看趋势，再决定今天该试什么。","topic_slug":"ai-daily","topic_name":"AI日报","topic_title":"AI日报：每日AI信号、工具动态与行动判断","topic_url":"https://opc.beizhux.com/topics/ai-daily","topic_path":"/topics/ai-daily","url":"https://opc.beizhux.com/glossary/ai-signal","path":"/glossary/ai-signal","json_url":"https://opc.beizhux.com/glossary/ai-signal.json"},{"slug":"ai-news-analysis","name":"热点解读","definition":"在AI觉醒星球里，「热点解读」属于「AI日报」方向。持续整理每日AI日报、模型更新、工具变化和行业信号，帮你快速判断哪些信息值得收藏、验证和行动。 每天先看趋势，再决定今天该试什么。","topic_slug":"ai-daily","topic_name":"AI日报","topic_title":"AI日报：每日AI信号、工具动态与行动判断","topic_url":"https://opc.beizhux.com/topics/ai-daily","topic_path":"/topics/ai-daily","url":"https://opc.beizhux.com/glossary/ai-news-analysis","path":"/glossary/ai-news-analysis","json_url":"https://opc.beizhux.com/glossary/ai-news-analysis.json"},{"slug":"builderpulse","name":"BuilderPulse","definition":"在AI觉醒星球里，「BuilderPulse」属于「AI日报」方向。持续整理每日AI日报、模型更新、工具变化和行业信号，帮你快速判断哪些信息值得收藏、验证和行动。 每天先看趋势，再决定今天该试什么。","topic_slug":"ai-daily","topic_name":"AI日报","topic_title":"AI日报：每日AI信号、工具动态与行动判断","topic_url":"https://opc.beizhux.com/topics/ai-daily","topic_path":"/topics/ai-daily","url":"https://opc.beizhux.com/glossary/builderpulse","path":"/glossary/builderpulse","json_url":"https://opc.beizhux.com/glossary/builderpulse.json"},{"slug":"tools","name":"工具","definition":"在AI觉醒星球里，「工具」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/tools","path":"/glossary/tools","json_url":"https://opc.beizhux.com/glossary/tools.json"},{"slug":"automation","name":"自动化","definition":"在AI觉醒星球里，「自动化」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/automation","path":"/glossary/automation","json_url":"https://opc.beizhux.com/glossary/automation.json"},{"slug":"ai-model","name":"模型","definition":"在AI觉醒星球里，「模型」属于「AI工具」方向。围绕AI工具、模型能力、自动化流程和真实使用场景做整理，优先关注能提升效率、降低成本和创造新交付的工具。 不只看工具热度，更看它能不能进入真实流程。","topic_slug":"ai-tools","topic_name":"AI工具","topic_title":"AI工具：模型、插件、自动化与实战场景","topic_url":"https://opc.beizhux.com/topics/ai-tools","topic_path":"/topics/ai-tools","url":"https://opc.beizhux.com/glossary/ai-model","path":"/glossary/ai-model","json_url":"https://opc.beizhux.com/glossary/ai-model.json"}],"preview_hidden_sections":[],"related_items":[{"id":3589,"title":"【必读】每日AI日报 2026-10-24","url":"https://opc.beizhux.com/content/3589/aihot-daily-2026-10-24","summary":"AIHOT 每日 AI 日报：模型发布/更新： - 月之暗面发布 Kimi K3.1 Pro，100 万 token 上下文，开放权重将于 11 月 4 日前释出：月之暗面发布 Kimi K3.1 系列首个模型 Kimi K3.1 Pro，为 3T 级开放权重模型，上下文窗口 100 万 token，并同步推出速度快…","access_level":"public"},{"id":3576,"title":"OpenAI 红队测试发现 Brotli 未认证远程代码执行漏洞并已披露","url":"https://opc.beizhux.com/content/3576/openai-brotli","summary":"OpenAI 披露红队演练中发现的 Brotli 未认证远程代码执行漏洞，恶意压缩数据可触发代码执行，影响 Node.js、Bun、Envoy 等内嵌 Brotli 的软件，未打补丁的压缩输入服务器可能被接管。","access_level":"public"},{"id":3580,"title":"GitHub EMU 账户被误删：SCIM 取消配置路径的失效与修复","url":"https://opc.beizhux.com/content/3580/emu-deletions-fix","summary":"GitHub 披露，2026 年 6 月 1 日至 13 日期间，一条新的取消配置路径未正确校验账户类型，导致约 12,285 个 Enterprise Managed User 账户被直接删除而非“暂停并混淆化”，其中 76 个在删除时仍处活跃状态；问题已于 6 月 13 日被控制，受影响记录在 2026 年 10…","access_level":"public"}],"usage_policy":{"summarizable":true,"preferred_url":"https://opc.beizhux.com/content/3583/three-week-on-site-audit-at-openai-reveals-major-gaps-in-ai-safety-monitoring","private_fields_excluded":true}}