{
 "generated": "2026-11-01",
 "node": {
  "id": "inst:guidelight-control-assessment",
  "type": "Instrument",
  "name": "Guidelight《前沿AI控制实践评估》",
  "slug": "guidelight-control-assessment",
  "kind": "report",
  "kind_zh": "报告 / 调查 / 评估",
  "date": "2026-08",
  "summary": "独立评估，按「遏制计划」单项评分：OpenAI 3/5，Google 2/5，xAI 1/5，Anthropic 与 Meta 均 0/5；评估者称几乎没人公开说清楚模型真失控后具体怎么办。",
  "url": "https://guidelight.ai/blog/control-assessment-august-2026",
  "evidence": [
   {
    "kind": "own",
    "url": "https://xiaopingfeng.com/landscape/agent-security/#player-guidelight",
    "text": "Landscape「Agent 安全攻击面」第 07 层「失控遏制」的卡片「Guidelight《前沿AI控制实践评估》」：独立评估 · 2026-08"
   },
   {
    "kind": "official",
    "url": "https://guidelight.ai/blog/control-assessment-august-2026",
    "text": "Guidelight 发布的评估原文（卡片所链接的页面）"
   }
  ],
  "entry": "agent",
  "by": "landscape-structure",
  "at": "2026-11-01T10:26:39Z",
  "landscape": [
   {
    "map": "agent-security",
    "map_title": "Agent 安全攻击面",
    "card": "player-guidelight",
    "title": "Guidelight《前沿AI控制实践评估》",
    "layer": "失控 遏制",
    "layer_id": "layer:agent-security--7",
    "role": "primary"
   }
  ],
  "landscape_maps": [
   "agent-security"
  ],
  "landscape_n": 1,
  "degree": 0
 },
 "page": "https://xiaopingfeng.com/graph/e/inst/guidelight-control-assessment/",
 "out": [
  {
   "type": "involves",
   "sentence": "Guidelight《前沿AI控制实践评估》涉及Guidelight",
   "direction": "out",
   "id": "org:guidelight",
   "name": "Guidelight",
   "role": "issuer",
   "note": "评估发布方（独立评估，2026-08）",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「Agent 安全攻击面」第 07 层「失控遏制」的卡片「Guidelight《前沿AI控制实践评估》」：Guidelight《前沿AI控制实践评估》",
     "url": "https://xiaopingfeng.com/landscape/agent-security/#player-guidelight"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "Guidelight《前沿AI控制实践评估》涉及OpenAI",
   "direction": "out",
   "id": "org:openai",
   "name": "OpenAI",
   "role": "subject",
   "note": "遏制计划单项评分 3/5",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「Agent 安全攻击面」第 07 层「失控遏制」的卡片「Guidelight《前沿AI控制实践评估》」：OpenAI 3/5",
     "url": "https://xiaopingfeng.com/landscape/agent-security/#player-guidelight"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "Guidelight《前沿AI控制实践评估》涉及Google",
   "direction": "out",
   "id": "org:google",
   "name": "Google",
   "role": "subject",
   "note": "遏制计划单项评分 2/5",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「Agent 安全攻击面」第 07 层「失控遏制」的卡片「Guidelight《前沿AI控制实践评估》」：Google 2/5",
     "url": "https://xiaopingfeng.com/landscape/agent-security/#player-guidelight"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "Guidelight《前沿AI控制实践评估》涉及SpaceXAI",
   "direction": "out",
   "id": "org:spacexai",
   "name": "SpaceXAI（原 xAI）",
   "role": "subject",
   "note": "遏制计划单项评分 1/5（卡片写作 xAI）",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「Agent 安全攻击面」第 07 层「失控遏制」的卡片「Guidelight《前沿AI控制实践评估》」：xAI 1/5",
     "url": "https://xiaopingfeng.com/landscape/agent-security/#player-guidelight"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "Guidelight《前沿AI控制实践评估》涉及Anthropic",
   "direction": "out",
   "id": "org:anthropic",
   "name": "Anthropic",
   "role": "subject",
   "note": "遏制计划单项评分 0/5",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「Agent 安全攻击面」第 07 层「失控遏制」的卡片「Guidelight《前沿AI控制实践评估》」：Anthropic与Meta均为0/5",
     "url": "https://xiaopingfeng.com/landscape/agent-security/#player-guidelight"
    }
   ]
  },
  {
   "type": "involves",
   "sentence": "Guidelight《前沿AI控制实践评估》涉及Meta",
   "direction": "out",
   "id": "org:meta",
   "name": "Meta",
   "role": "subject",
   "note": "遏制计划单项评分 0/5",
   "status": "agent",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「Agent 安全攻击面」第 07 层「失控遏制」的卡片「Guidelight《前沿AI控制实践评估》」：Anthropic与Meta均为0/5",
     "url": "https://xiaopingfeng.com/landscape/agent-security/#player-guidelight"
    }
   ]
  },
  {
   "type": "placed_in",
   "sentence": "Guidelight《前沿AI控制实践评估》出现在 Landscape 的「失控遏制」",
   "direction": "out",
   "id": "layer:agent-security--7",
   "name": "失控遏制",
   "reported": false,
   "status": "derived",
   "evidence": [
    {
     "kind": "own",
     "label": "本站页面",
     "text": "Landscape「Agent 安全攻击面」第 07 层「失控遏制」：卡片「Guidelight《前沿AI控制实践评估》」",
     "url": "https://xiaopingfeng.com/landscape/agent-security/#player-guidelight"
    }
   ]
  }
 ],
 "in": []
}