{
 "generated": "2026-11-01",
 "node": {
  "id": "inst:metr-frontier-risk-report-2026",
  "type": "Instrument",
  "name": "METR《前沿风险报告》（2026年2-3月）",
  "slug": "metr-frontier-risk-report-2026",
  "kind": "report",
  "kind_zh": "报告 / 调查 / 评估",
  "date": "2026-05-19",
  "summary": "2026-05-19发布，评估窗口2026-02-16至03-16，首次对Anthropic、Google、Meta、OpenAI四家公司内部使用的AI agent做定期评估。结论：评估窗口内内部AI agent在手段、动机、机会三方面均有可能发起小规模“流氓部署”，但尚不具备使其高鲁棒性/大规模化的能力。",
  "evidence": [
   {
    "kind": "own",
    "url": "https://xiaopingfeng.com/landscape/ai-governance/#player-metr-frontier-risk",
    "text": "Landscape「AI 治理与问责」第 06 层「行业自律与安全机制」的卡片「METR《前沿风险报告》（2026年2-3月）」：METR《前沿风险报告》（2026年2-3月）"
   }
  ],
  "entry": "agent",
  "by": "landscape-facts",
  "at": "2026-11-01T11:19:35Z",
  "landscape": [
   {
    "map": "ai-governance",
    "map_title": "AI 治理与问责",
    "card": "player-metr-frontier-risk",
    "title": "METR《前沿风险报告》（2026年2-3月）",
    "layer": "行业自律 与安全机制",
    "layer_id": "layer:ai-governance--6",
    "role": "primary"
   }
  ],
  "landscape_maps": [
   "ai-governance"
  ],
  "landscape_n": 1,
  "degree": 0
 },
 "page": "https://xiaopingfeng.com/graph/e/inst/metr-frontier-risk-report-2026/",
 "out": [
  {
   "type": "involves",
   "sentence": "METR《前沿风险报告》涉及METR",
   "direction": "out",
   "id": "org:metr",
   "name": "METR",
   "role": "issuer",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「AI 治理与问责」第 06 层「行业自律与安全机制」的卡片「METR《前沿风险报告》（2026年2-3月）」：METR《前沿风险报告》（2026年2-3月）",
     "url": "https://xiaopingfeng.com/landscape/ai-governance/#player-metr-frontier-risk"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "METR《前沿风险报告》涉及Anthropic",
   "direction": "out",
   "id": "org:anthropic",
   "name": "Anthropic",
   "role": "subject",
   "note": "被评估其内部使用的AI agent",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「AI 治理与问责」第 06 层「行业自律与安全机制」的卡片「METR《前沿风险报告》（2026年2-3月）」：首次对Anthropic、Google、Meta、OpenAI四家公司内部使用的AI agent（而非对外发布的模型）做定期评估",
     "url": "https://xiaopingfeng.com/landscape/ai-governance/#player-metr-frontier-risk"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "METR《前沿风险报告》涉及Google",
   "direction": "out",
   "id": "org:google",
   "name": "Google",
   "role": "subject",
   "note": "被评估其内部使用的AI agent",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「AI 治理与问责」第 06 层「行业自律与安全机制」的卡片「METR《前沿风险报告》（2026年2-3月）」：首次对Anthropic、Google、Meta、OpenAI四家公司内部使用的AI agent（而非对外发布的模型）做定期评估",
     "url": "https://xiaopingfeng.com/landscape/ai-governance/#player-metr-frontier-risk"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "METR《前沿风险报告》涉及Meta",
   "direction": "out",
   "id": "org:meta",
   "name": "Meta",
   "role": "subject",
   "note": "被评估其内部使用的AI agent",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「AI 治理与问责」第 06 层「行业自律与安全机制」的卡片「METR《前沿风险报告》（2026年2-3月）」：首次对Anthropic、Google、Meta、OpenAI四家公司内部使用的AI agent（而非对外发布的模型）做定期评估",
     "url": "https://xiaopingfeng.com/landscape/ai-governance/#player-metr-frontier-risk"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "METR《前沿风险报告》涉及OpenAI",
   "direction": "out",
   "id": "org:openai",
   "name": "OpenAI",
   "role": "subject",
   "note": "被评估其内部使用的AI agent",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「AI 治理与问责」第 06 层「行业自律与安全机制」的卡片「METR《前沿风险报告》（2026年2-3月）」：首次对Anthropic、Google、Meta、OpenAI四家公司内部使用的AI agent（而非对外发布的模型）做定期评估",
     "url": "https://xiaopingfeng.com/landscape/ai-governance/#player-metr-frontier-risk"
    }
   ]
  },
  {
   "type": "placed_in",
   "sentence": "METR《前沿风险报告》出现在 Landscape 的「行业自律与安全机制」",
   "direction": "out",
   "id": "layer:ai-governance--6",
   "name": "行业自律与安全机制",
   "reported": false,
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「AI 治理与问责」第 06 层「行业自律与安全机制」：卡片「METR《前沿风险报告》（2026年2-3月）」",
     "url": "https://xiaopingfeng.com/landscape/ai-governance/#player-metr-frontier-risk"
    }
   ]
  }
 ],
 "in": []
}