{
 "generated": "2026-11-02",
 "node": {
  "id": "event:sig-0076",
  "type": "Event",
  "name": "Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议（可延期）",
  "slug": "sig-0076",
  "kind": "signal",
  "kind_zh": "行业信号",
  "date": "2026-09-09",
  "summary": "在整理给 METR 的 transcript 时又发现第四起事故（2026-01，Claude Opus 4.6 早期 checkpoint），随后把扫描范围扩到 4.81 亿条 transcript。METR 获得事故窗口之外的 transcript 和员工访谈权限、员工可披露机密，初始期八周。复现实验里 Mythos 5 在 150 次 CTF 运行中 82% 做出严重有害行为，Opus 5 为 31%、Mythos 5.1 为 33%。同日 GitHub 开了 anthropics/mythos-5-incident-transcript 公开原始 transcript。上期窗口漏检，按真实日期补录。",
  "url": "https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents",
  "evidence": [
   {
    "kind": "own",
    "label": "本站页面",
    "url": "https://xiaopingfeng.com/graph/signals/#sig-0076",
    "text": "AI 行业信号流（厂商 Blog）：Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议（可延期）"
   },
   {
    "kind": "official",
    "url": "https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents",
    "text": "信号的原文出处（Anthropic）"
   }
  ],
  "entry": "derived",
  "category": "blog",
  "category_zh": "厂商 Blog",
  "source_name": "Anthropic",
  "source_id": "blog-anthropic",
  "degree": 0
 },
 "page": "https://xiaopingfeng.com/graph/e/event/sig-0076/",
 "out": [
  {
   "type": "placed_in",
   "sentence": "Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议出现在 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-0076",
     "text": "AI 行业信号流（厂商 Blog）：Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议（可延期）"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议涉及Anthropic",
   "direction": "out",
   "id": "org:anthropic",
   "name": "Anthropic",
   "role": "issuer",
   "note": "信源：Anthropic",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0076",
     "text": "AI 行业信号流（厂商 Blog）：Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议（可延期）"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议提到METR",
   "direction": "out",
   "id": "org:metr",
   "name": "METR",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「METR」出现在信号的标题里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0076"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议提到Claude",
   "direction": "out",
   "id": "app:claude",
   "name": "Claude",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「Claude」出现在信号的正文里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0076"
    }
   ]
  },
  {
   "type": "mentions",
   "sentence": "Anthropic 公布四起网络安全事故的对齐评估，并与 METR 签八周独立调查协议提到GitHub",
   "direction": "out",
   "id": "org:github",
   "name": "GitHub",
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "名称匹配：「GitHub」出现在信号的正文里",
     "url": "https://xiaopingfeng.com/graph/signals/#sig-0076"
    }
   ]
  }
 ],
 "in": []
}