{
 "generated": "2026-11-06",
 "sha": "78bbd39a8828",
 "node": {
  "id": "event:sig-0097",
  "type": "Event",
  "name": "OpenAI 发布 GPT-6 Cyber，并按自家 Preparedness Framework 首次把模型网络安全等级定为 Critical",
  "slug": "sig-0097",
  "kind": "signal",
  "kind_zh": "行业信号",
  "date": "2026-11-06",
  "summary": "11-05 发布。GPT-6 Cyber 是面向漏洞发现与修复的专用模型，ExploitGym 从 GPT-5.6 Cyber 的 32.1% 升到 48.5%（相对提升 51%）；新建的 web 渗透测试评测里漏洞召回 40.3%，高于 GPT-6 Astra 的 32.0%；新建的 RemoteExploitBench（GitLab/PostgreSQL/Jenkins 已知漏洞的受控环境）里，GPT-6 Astra 配 Daybreak Blue 拒绝了 100% 的任务，GPT-6 Cyber 完成 95%。对齐侧做了专门训练以守住授权范围：不带部署提示时 99.93% 的测试中没有去打范围外服务，带提示时 100%。基于这些结果，OpenAI 按 Preparedness Framework 把它定为网络安全 Critical 级（GPT-5.6 Cyber 是 High）——这是该框架下首次有模型进入 Critical。实战成果：用它开发了 948 个 Linux 本地提权利用，并在已被前沿模型反复测过的 Firefox 上找出 81 个此前未知的高危漏洞，Mozilla 已发布多数修复；Cisco、Calif、CodeRabbit 在用。访问走 Daybreak 计划（Blue 给已验证防守方，Red 给更深度的利用开发与授权测试）。",
  "url": "https://openai.com/index/introducing-gpt-6-cyber",
  "evidence": [
   {
    "kind": "own",
    "label": "本站页面",
    "url": "https://xiaopingfeng.com/graph/signals/#sig-0097",
    "text": "AI 行业信号流（厂商 Blog）：OpenAI 发布 GPT-6 Cyber，并按自家 Preparedness Framework 首次把模型网络安全等级定为 Critical"
   },
   {
    "kind": "official",
    "url": "https://openai.com/index/introducing-gpt-6-cyber",
    "text": "信号的原文出处（OpenAI）"
   }
  ],
  "entry": "derived",
  "category": "blog",
  "category_zh": "厂商 Blog",
  "source_name": "OpenAI",
  "source_id": "blog-openai",
  "links": [
   {
    "label": "OpenAI",
    "url": "https://openai.com/index/introducing-gpt-6-cyber"
   }
  ],
  "degree": 0
 },
 "page": "https://xiaopingfeng.com/graph/e/event/sig-0097/",
 "out": [
  {
   "type": "placed_in",
   "sentence": "OpenAI 发布 GPT-6 Cyber，并按自家 Preparedness Framework 首次把模型网络安全等级定为 Critical出现在 Landscape 的「厂商 Blog」",
   "direction": "out",
   "id": "layer:ai-industry-signals--3",
   "name": "厂商 Blog",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0097",
     "text": "AI 行业信号流（厂商 Blog）：OpenAI 发布 GPT-6 Cyber，并按自家 Preparedness Framework 首次把模型网络安全等级定为 Critical"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "OpenAI 发布 GPT-6 Cyber，并按自家 Preparedness Framework 首次把模型网络安全等级定为 Critical涉及OpenAI",
   "direction": "out",
   "id": "org:openai",
   "name": "OpenAI",
   "role": "issuer",
   "note": "信源：OpenAI",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0097",
     "text": "AI 行业信号流（厂商 Blog）：OpenAI 发布 GPT-6 Cyber，并按自家 Preparedness Framework 首次把模型网络安全等级定为 Critical"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "OpenAI 发布 GPT-6 Cyber，并按自家 Preparedness Framework 首次把模型网络安全等级定为 Critical提到Cisco",
   "direction": "out",
   "id": "org:cisco",
   "name": "Cisco",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「Cisco」出现在信号的正文里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0097"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "OpenAI 发布 GPT-6 Cyber，并按自家 Preparedness Framework 首次把模型网络安全等级定为 Critical提到Mozilla",
   "direction": "out",
   "id": "org:mozilla",
   "name": "Mozilla",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「Mozilla」出现在信号的正文里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0097"
    }
   ]
  }
 ],
 "in": []
}