{"id":89490,"date":"2026-08-03T05:33:32","date_gmt":"2026-08-02T21:33:32","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/89490.html"},"modified":"2026-08-03T05:33:32","modified_gmt":"2026-08-02T21:33:32","slug":"safeclawbench%ef%bc%9a%e5%9c%a8%e4%bd%bf%e7%94%a8%e5%b7%a5%e5%85%b7%e7%9a%84%e5%a4%a7%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b%e4%bb%a3%e7%90%86%e4%b8%ad%e5%88%86%e7%a6%bb%e8%af%ad%e4%b9%89%e5%8d%b1","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/89490.html","title":{"rendered":"SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3"},"content":{"rendered":"<p>\u5927\u5bb6\u8bfb\u5b8c\u89c9\u5f97\u6709\u5e2e\u52a9\u8bb0\u5f97\u5173\u6ce8\u548c\u70b9\u8d5e&#xff01;&#xff01;&#xff01;<\/p>\n<p>\u6458\u8981<br \/>\n\u4f7f\u7528\u5de5\u5177\u7684\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4f1a\u5f15\u5165\u8d85\u8d8a\u4e0d\u5b89\u5168\u6587\u672c\u7684\u5b89\u5168\u5931\u6548&#xff1a;\u5b83\u4eec\u53ef\u80fd\u6cc4\u9732\u53d7\u4fdd\u62a4\u5bf9\u8c61\u3001\u5199\u5165\u6301\u4e45\u5316\u5185\u5b58\u3001\u53d1\u9001\u6d88\u606f\u3001\u4fee\u6539\u6570\u636e\u5e93&#xff0c;\u6216\u89e6\u53d1\u6709\u5bb3\u4ee3\u7801\u548c\u5de5\u5177\u6548\u5e94\u3002\u73b0\u6709\u8bc4\u4f30\u5f80\u5f80\u5c06\u8fd9\u4e9b\u9636\u6bb5\u5408\u5e76\u4e3a\u5355\u4e00\u7684\u653b\u51fb\u6210\u529f\u7387&#xff0c;\u4f7f\u5f97\u96be\u4ee5\u533a\u5206\u6a21\u578b\u4ec5\u4ec5\u662f\u540c\u610f\u4e86\u653b\u51fb\u8005\u8fd8\u662f\u5b9e\u9645\u4ea7\u751f\u4e86\u53ef\u89c2\u5bdf\u7684\u5371\u5bb3\u3002\u6211\u4eec\u63d0\u51fa\u4e86 SafeClawBench&#xff0c;\u4e00\u4e2a\u9488\u5bf9\u4f7f\u7528\u5de5\u5177\u7684\u4ee3\u7406\u5b89\u5168\u7684\u5206\u9636\u6bb5\u57fa\u51c6\u6d4b\u8bd5&#xff0c;\u5305\u542b 600 \u4e2a\u53d7\u63a7\u5bf9\u6297\u4efb\u52a1&#xff0c;\u6db5\u76d6\u516d\u7c7b\u653b\u51fb&#xff1a;\u76f4\u63a5\u548c\u95f4\u63a5\u63d0\u793a\u6ce8\u5165\u3001\u5de5\u5177\u8fd4\u56de\u6ce8\u5165\u3001\u5185\u5b58\u6295\u6bd2\u3001\u5185\u5b58\u63d0\u53d6&#xff0c;\u4ee5\u53ca\u6b67\u4e49\u9a71\u52a8\u7684\u4e0d\u5b89\u5168\u63a8\u7406\u3002SafeClawBench \u62a5\u544a\u4e09\u4e2a\u72ec\u7acb\u7684\u7aef\u70b9&#xff1a;\u8bed\u4e49\u653b\u51fb\u63a5\u53d7\u5ea6\u3001\u5ba1\u8ba1\u53ef\u89c1\u7684\u5371\u5bb3\u8bc1\u636e&#xff0c;\u4ee5\u53ca\u6c99\u7bb1\u89c2\u5bdf\u5230\u7684\u5de5\u5177\/\u72b6\u6001\u5371\u5bb3\u3002\u6211\u4eec\u5728\u56db\u79cd\u63d0\u793a\u7ea7\u7b56\u7565\u4e0b\u8bc4\u4f30\u4e86\u4e94\u4e2a\u4ee3\u7406\u7aef\u70b9&#xff0c;\u53d1\u73b0\u8fd9\u4e9b\u7aef\u70b9\u6355\u83b7\u4e86\u4e0d\u540c\u7684\u5931\u6548\u6a21\u5f0f\u3002\u5728\u6ca1\u6709\u989d\u5916\u63d0\u793a\u4fdd\u62a4\u7684\u60c5\u51b5\u4e0b&#xff0c;\u8bed\u4e49\u5931\u8d25\u7387\u5728\u4e0d\u540c\u6a21\u578b\u95f4\u5dee\u5f02\u5f88\u5927&#xff0c;\u4ece 9.0% \u5230 44.2%\u3002\u5ba1\u8ba1\u5230\u7684\u5371\u5bb3\u8bc1\u636e\u8303\u56f4\u6bd4\u8bed\u4e49\u5931\u8d25\u66f4\u7a84&#xff0c;\u5e76\u4e14\u5728\u4e00\u79cd\u5355\u72ec\u7684\u53ef\u6267\u884c\u534f\u8bae\u4e0b&#xff0c;\u67d0\u4e9b\u5339\u914d\u7684\u4efb\u52a1\u6807\u8bc6\u5c3d\u7ba1\u901a\u8fc7\u4e86\u8bed\u4e49\u6838\u5fc3\u8c03\u7528&#xff0c;\u4ecd\u4f1a\u4ea7\u751f\u6c99\u7bb1\u5371\u5bb3&#xff1a;\u5728 12,000 \u884c\u7684\u5339\u914d\u5206\u6790\u4e2d&#xff0c;\u89c2\u5bdf\u5230\u7684 347 \u4f8b\u6c99\u7bb1\u5371\u5bb3\u4e2d\u6709 291 \u4f8b\u51fa\u73b0\u5728\u901a\u8fc7\u8bed\u4e49\u68c0\u67e5\u7684\u884c\u4e2d\u3002\u63d0\u793a\u7b56\u7565\u4f1a\u6539\u53d8\u7aef\u70b9\u7ed3\u679c&#xff0c;\u4f46\u5176\u6548\u679c\u65e2\u53d6\u51b3\u4e8e\u6a21\u578b\u4e5f\u53d6\u51b3\u4e8e\u534f\u8bae\u3002SafeClawBench \u63d0\u4f9b\u4e86\u4e00\u4e2a\u53ef\u91cd\u73b0\u7684\u6846\u67b6&#xff0c;\u7528\u4e8e\u6bd4\u8f83\u4ee3\u7406\u6a21\u578b\u548c\u63d0\u793a\u7b56\u7565\u6761\u4ef6&#xff0c;\u800c\u4e0d\u4f1a\u5c06\u6587\u672c\u9075\u4ece\u6027\u3001\u8bc1\u636e\u652f\u6301\u7684\u5371\u5bb3\u548c\u53ef\u6267\u884c\u72b6\u6001\u53d8\u5316\u6df7\u4e3a\u4e00\u8c08\u3002<\/p>\n<hr \/>\n<p>1. \u5f15\u8a00<\/p>\n<p>\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5df2\u4ece\u5bf9\u8bdd\u52a9\u624b\u6f14\u5316\u4e3a\u80fd\u591f\u6d4f\u89c8\u3001\u6267\u884c\u4ee3\u7801\u3001\u67e5\u8be2\u6570\u636e\u5e93\u3001\u53d1\u9001\u7535\u5b50\u90ae\u4ef6\u548c\u5b58\u50a8\u6301\u4e45\u5316\u5185\u5b58\u7684\u4ee3\u7406&#xff08;Yao \u7b49&#xff0c;2023&#xff1b;Schick \u7b49&#xff0c;2023&#xff1b;Zhou \u7b49&#xff0c;2024&#xff09;\u3002OpenClaw&#xff08;OpenClaw Contributors&#xff0c;2024&#xff09;\u3001AutoGPT&#xff08;AutoGPT Contributors&#xff0c;2023&#xff09;\u548c LangChain&#xff08;LangChain Contributors&#xff0c;2023&#xff09;\u7b49\u6846\u67b6\u4f7f\u5f97\u8fd9\u7c7b\u7cfb\u7edf\u5728\u73af\u5883\u6743\u9650&#xff08;ambient authority&#xff09;\u4e0b\u6613\u4e8e\u90e8\u7f72&#xff1a;\u5373\u4ee3\u8868\u7528\u6237\u91c7\u53d6\u884c\u52a8\u800c\u76d1\u7763\u6709\u9650\u3002\u8fd9\u6539\u53d8\u4e86\u5b89\u5168\u76ee\u6807\u3002\u4ee3\u7406\u5b89\u5168\u662f\u5206\u9636\u6bb5\u7684&#xff1a;\u4e00\u6b21\u653b\u51fb\u53ef\u80fd\u9996\u5148\u8bf1\u5bfc\u8bed\u4e49\u987a\u4ece&#xff0c;\u7136\u540e\u8bbf\u95ee\u53d7\u4fdd\u62a4\u5bf9\u8c61\u3001\u6539\u53d8\u72b6\u6001\u3001\u6cc4\u9732\u91d1\u4e1d\u96c0\u79d8\u5bc6&#xff0c;\u6216\u4e3a\u540e\u7eed\u4f1a\u8bdd\u6295\u6bd2\u5185\u5b58\u3002\u5c06\u8fd9\u4e9b\u7aef\u70b9\u538b\u7f29\u4e3a\u4e00\u4e2a\u653b\u51fb\u6210\u529f\u6570\u5b57&#xff0c;\u63a9\u76d6\u4e86\u8bed\u4e49\u635f\u5bb3\u3001\u4ec5\u8bed\u4e49\u5931\u8d25\u548c\u53ef\u89c2\u5bdf\u57fa\u51c6\u5371\u5bb3\u4e4b\u95f4\u7684\u5dee\u5f02\u3002<\/p>\n<p>\u73b0\u6709\u7684 LLM \u5b89\u5168\u57fa\u51c6\u4e3b\u8981\u6d4b\u91cf\u4e0d\u5b89\u5168\u7684\u6587\u672c\u751f\u6210&#xff08;Mazeika \u7b49&#xff0c;2024&#xff1b;Zhang \u7b49&#xff0c;2024&#xff1b;Huang \u7b49&#xff0c;2024b&#xff09;\u3002\u4ee3\u7406\u5b89\u5168\u57fa\u51c6\u5982 PASB&#xff08;Wang \u7b49&#xff0c;2026&#xff09;\u3001InjecAgent&#xff08;Zhan \u7b49&#xff0c;2024&#xff09;\u3001AgentDojo&#xff08;Debenedetti \u7b49&#xff0c;2024&#xff09;\u3001AgentHarm&#xff08;Andriushchenko \u7b49&#xff0c;2025&#xff09;\u3001R-Judge&#xff08;Yuan \u7b49&#xff0c;2024&#xff09;\u548c ToolEmu&#xff08;Ruan \u7b49&#xff0c;2024&#xff09;\u89e3\u51b3\u4e86\u90e8\u5206\u95ee\u9898&#xff0c;\u4f46\u901a\u5e38\u805a\u7126\u4e8e\u5355\u4e00\u653b\u51fb\u65cf\u3001\u4e00\u79cd\u73af\u5883\u62bd\u8c61\u6216\u6709\u9650\u7684\u9632\u5fa1\u8986\u76d6\u3002SafeClawBench \u66f4\u63a5\u8fd1\u4e8e\u4e00\u4e2a\u5206\u9636\u6bb5\u538b\u529b\u6d4b\u8bd5\u57fa\u51c6&#xff0c;\u800c\u975e\u7fa4\u4f53\u98ce\u9669\u7814\u7a76&#xff1a;\u5176\u76ee\u7684\u662f\u5728\u53d7\u63a7\u5bf9\u6297\u538b\u529b\u4e0b\u533a\u5206\u6a21\u578b\u3001\u63d0\u793a\u7b56\u7565\u548c\u7aef\u70b9\u5b9a\u4e49\u3002<\/p>\n<p>\u6838\u5fc3\u8bbe\u8ba1\u539f\u5219\u662f\u7aef\u70b9\u5206\u79bb\u3002\u8bed\u4e49\u635f\u5bb3\u3001\u8bc1\u636e\u652f\u6301\u7684\u5371\u5bb3\u3001\u53ef\u6267\u884c\u72b6\u6001\u53d8\u5316\u548c\u826f\u6027\u6548\u7528\u662f\u76f8\u5173\u4f46\u4e0d\u53ef\u4e92\u6362\u7684\u3002\u8fd9\u79cd\u5206\u79bb\u4f7f\u6211\u4eec\u4e0d\u4ec5\u80fd\u591f\u8be2\u95ee\u6a21\u578b\u662f\u5426\u9075\u5faa\u4e86\u5bf9\u6297\u6027\u6307\u4ee4&#xff0c;\u8fd8\u80fd\u8be2\u95ee\u8be5\u5931\u6548\u662f\u5426\u4ea7\u751f\u4e86\u53d7\u4fdd\u62a4\u5bf9\u8c61\u6cc4\u9732\u3001\u672a\u6388\u6743\u8bbf\u95ee\u3001\u6709\u5bb3\u72b6\u6001\u53d8\u5316\u3001\u6301\u4e45\u5316\u6c61\u67d3\u6216\u6709\u5bb3\u5de5\u5177\/\u72b6\u6001\u6548\u5e94\u3002<\/p>\n<p>\u6211\u4eec\u5c06\u4e3b\u8981\u5206\u6790\u56f4\u7ed5\u4e09\u4e2a\u7814\u7a76\u95ee\u9898\u5c55\u5f00&#xff1a;<\/p>\n<li>\n<p>RQ1&#xff1a;\u5728\u7cbe\u5fc3\u7b56\u5212\u7684\u4ee3\u7406\u5b89\u5168\u538b\u529b\u6d4b\u8bd5\u4e0b&#xff0c;\u6a21\u578b\u5728\u8bed\u4e49\u635f\u5bb3\u65b9\u9762\u6709\u4f55\u5dee\u5f02&#xff1f;<\/p>\n<\/li>\n<li>\n<p>RQ2&#xff1a;\u5ba1\u8ba1\u5230\u7684\u8bed\u4e49\u5931\u8d25\u5728\u591a\u5927\u7a0b\u5ea6\u4e0a\u5bf9\u5e94\u4e8e\u8bc1\u636e\u652f\u6301\u7684\u5371\u5bb3&#xff1f;<\/p>\n<\/li>\n<li>\n<p>RQ3&#xff1a;\u53ef\u6267\u884c\u72b6\u6001\u9884\u8a00\u673a\u4f55\u65f6\u4e0e\u8bed\u4e49\u5224\u65ad\u4e0d\u4e00\u81f4&#xff1f;<\/p>\n<\/li>\n<p>\u6211\u4eec\u505a\u51fa\u4e94\u9879\u8d21\u732e&#xff1a;<\/p>\n<li>\n<p>\u57fa\u51c6\u5de5\u4ef6&#xff1a;\u4e00\u4e2a\u62a5\u544a\u65cf\u5206\u7c7b\u6cd5\u3001\u4e00\u4e2a 600 \u6848\u4f8b\u7684\u8bed\u4e49\u6838\u5fc3\u3001\u4e00\u4e2a\u6838\u5fc3\u95e8\u63a7\u7684\u5371\u5bb3\u8bc1\u636e\u6a21\u5f0f\u3001\u4e00\u4e2a\u53ef\u6267\u884c\u6c99\u7bb1\u9762\u677f&#xff0c;\u4ee5\u53ca\u4e00\u4e2a\u7528\u4e8e\u4f7f\u7528\u5de5\u5177\u7684\u4ee3\u7406\u7684\u5b58\u6863\u826f\u6027\u6548\u7528\u914d\u5957\u68c0\u67e5\u3002<\/p>\n<\/li>\n<li>\n<p>\u591a\u7aef\u70b9\u6d4b\u91cf&#xff1a;CoreFail&#064;600\u3001\u6838\u5fc3\u95e8\u63a7\u7684 HarmEvidence&#064;600\u3001\u53ef\u6267\u884c\u72b6\u6001\u9884\u8a00\u673a\u7aef\u70b9&#xff0c;\u4ee5\u53ca\u4f5c\u4e3a\u72ec\u7acb\u7ed3\u679c\u800c\u975e\u538b\u7f29\u653b\u51fb\u6210\u529f\u7387\u62a5\u544a\u7684\u6548\u7528\u6307\u6807\u3002<\/p>\n<\/li>\n<li>\n<p>\u65d7\u8230\u6bd4\u8f83\u9762\u677f&#xff1a;\u4e00\u4e2a\u4e94\u6a21\u578b \u00d7 \u56db\u7b56\u7565\u7684\u4e3b\u77e9\u9635&#xff0c;\u6bcf\u4e2a\u63d0\u4f9b\u8005\u65cf\u4f7f\u7528\u4e00\u4e2a\u4e3b\u7aef\u70b9&#xff0c;\u7aef\u70b9\u548c\u5de5\u4ef6\u5728\u5143\u6570\u636e\u4e2d\u56fa\u5b9a\u3002<\/p>\n<\/li>\n<li>\n<p>\u5ba1\u8ba1\u548c\u53ef\u6267\u884c\u68c0\u67e5&#xff1a;\u6838\u5fc3\u95e8\u63a7\u7684\u5371\u5bb3\u8bc1\u636e\u5ba1\u8ba1&#xff0c;\u4ee5\u53ca\u5339\u914d\u7684\u6838\u5fc3\u2013\u53ef\u6267\u884c\u884c&#xff08;\u5176\u4e2d\u5b58\u5728\u914d\u5bf9\u5de5\u4ef6&#xff09;&#xff0c;\u5305\u62ec\u5728\u5355\u72ec\u53ef\u6267\u884c\u534f\u8bae\u4e0b\u89c2\u5bdf\u5230\u7684 CorePass\u2013ExecHarm \u6848\u4f8b\u3002<\/p>\n<\/li>\n<li>\n<p>\u63d0\u793a\u7b56\u7565\u8bca\u65ad&#xff1a;\u5c06\u8f7b\u91cf\u7ea7\u3001\u5206\u5c42\u548c\u957f\u8fc7\u5ea6\u6307\u5b9a\u7684\u63d0\u793a\u7b56\u7565\u4e0e\u6d88\u878d\u5b9e\u9a8c\u548c\u5339\u914d\u957f\u5ea6\u5bf9\u7167\u8fdb\u884c\u6bd4\u8f83&#xff0c;\u4ee5\u5206\u79bb\u7b56\u7565\u6548\u5e94\u4e0e\u63d0\u793a\u957f\u5ea6\u6548\u5e94\u3002<\/p>\n<\/li>\n<p>\u00a0<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213330-6a6fb7aa92cbd.png\" \/><\/p>\n<p>\u56fe1&#xff1a;SafeClawBench \u57fa\u51c6\u6d4b\u8bd5\u548c\u7aef\u70b9\u7ed3\u6784\u3002\u57fa\u51c6\u4ece600\u4e2a\u7cbe\u5fc3\u7b56\u5212\u7684\u4ee3\u7406\u5b89\u5168\u6848\u4f8b\u51fa\u53d1&#xff0c;\u8bc4\u4f30\u56fa\u5b9a\u7684\u4e94\u7aef\u70b9\u00d7\u56db\u4e2a\u7b56\u7565\u8bed\u4e49\u6838\u5fc3\u5c0f\u7ec4&#xff0c;\u5ba1\u8ba1\u89c4\u8303\u7684CoreFail\u884c\u4ee5\u5bfb\u627e\u4f2a\u9020\u7269\u53ef\u89c1\u7684\u4f24\u5bb3\u8bc1\u636e&#xff0c;\u5e76\u5728\u6267\u884c\u5e73\u8861\u6c99\u76d2\u4e2d\u5355\u72ec\u8fd0\u884c\u5339\u914d\u6848\u4f8b\u3002\u8bba\u6587\u62a5\u544a\u4e86CoreFail&#064;600\u3001HarmEvidence&#064;600\u3001SemanticOnly\u548cObservedHarm&#064;Exec\u4f5c\u4e3a\u4e0d\u540c\u7684\u7ec8\u70b9;D4 \u8868\u793a D4\/LongPolicy \u63d0\u793a\u590d\u6742\u5ea6\u63a2\u9488\u3002<\/p>\n<hr \/>\n<p>2. \u76f8\u5173\u5de5\u4f5c<\/p>\n<p>LLM \u5b89\u5168\u8bc4\u4f30\u5df2\u4ece\u7ea2\u961f\u6d4b\u8bd5&#xff08;Ganguli \u7b49&#xff0c;2022&#xff1b;Perez \u7b49&#xff0c;2022&#xff09;\u53d1\u5c55\u5230 HarmBench\u3001SafetyBench\u3001TrustLLM \u548c AdvBench \u7b49\u57fa\u51c6&#xff08;Mazeika \u7b49&#xff0c;2024&#xff1b;Zhang \u7b49&#xff0c;2024&#xff1b;Huang \u7b49&#xff0c;2024b&#xff1b;Zou \u7b49&#xff0c;2023&#xff09;\u3002\u8fd9\u4e9b\u5de5\u4f5c\u5bf9\u804a\u5929\u5b89\u5168\u81f3\u5173\u91cd\u8981&#xff0c;\u4f46\u5e76\u4e0d\u76f4\u63a5\u6d4b\u8bd5\u6301\u4e45\u5316\u5185\u5b58\u3001\u5de5\u5177\u8fd4\u56de\u901a\u9053\u6216\u72b6\u6001\u53d8\u5316\u3002GCG\u3001AutoDAN\u3001PAIR\u3001TAP \u548c\u591a\u8f6e\u8d8a\u72f1\u7b49\u5bf9\u6297\u63d0\u793a\u65b9\u6cd5&#xff08;Zou \u7b49&#xff0c;2023&#xff1b;Liu \u7b49&#xff0c;2024&#xff1b;Chao \u7b49&#xff0c;2024&#xff1b;Mehrotra \u7b49&#xff0c;2024&#xff1b;Anil \u7b49&#xff0c;2024&#xff09;\u63a8\u52a8\u4e86\u9c81\u68d2\u81ea\u52a8\u5316\u8bc4\u4f30\u7684\u9700\u6c42&#xff0c;\u800c\u95f4\u63a5\u63d0\u793a\u6ce8\u5165\u548c\u6295\u6bd2\u5de5\u4f5c&#xff08;Greshake \u7b49&#xff0c;2023&#xff1b;Yang \u7b49&#xff0c;2024&#xff1b;Zou \u7b49&#xff0c;2024&#xff09;\u5219\u6fc0\u53d1\u4e86\u4ee3\u7406\u7279\u5b9a\u5a01\u80c1\u901a\u9053\u7684\u52a8\u673a\u3002<\/p>\n<p>\u4ee3\u7406\u5b89\u5168\u57fa\u51c6\u5728\u8303\u56f4\u4e0a\u66f4\u4e3a\u63a5\u8fd1\u3002PASB\u3001InjecAgent\u3001AgentDojo\u3001AgentHarm\u3001R-Judge \u548c ToolEmu&#xff08;Wang \u7b49&#xff0c;2026&#xff1b;Zhan \u7b49&#xff0c;2024&#xff1b;Debenedetti \u7b49&#xff0c;2024&#xff1b;Andriushchenko \u7b49&#xff0c;2025&#xff1b;Yuan \u7b49&#xff0c;2024&#xff1b;Ruan \u7b49&#xff0c;2024&#xff09;\u63d0\u4f9b\u4e86\u4e2a\u6027\u5316\u4ee3\u7406\u3001\u6ce8\u5165\u3001\u6709\u5bb3\u9075\u4ece\u6027\u3001\u98ce\u9669\u610f\u8bc6\u6216\u6a21\u62df\u5de5\u5177\u8bc4\u4f30\u3002\u53e6\u4e00\u6761\u76f8\u8fd1\u7684\u5de5\u4f5c\u7ebf\u7814\u7a76\u81ea\u4e3b\u7814\u7a76\u4ee3\u7406&#xff1a;PaperQA \u68c0\u7d22\u5e76\u7efc\u5408\u5e26\u6709\u6765\u6e90\u7684\u79d1\u5b66\u6587\u732e&#xff08;L\u00e1la \u7b49&#xff0c;2023&#xff09;&#xff0c;MLAgentBench \u8bc4\u4f30\u8bed\u8a00\u4ee3\u7406\u5728\u673a\u5668\u5b66\u4e60\u5b9e\u9a8c\u4efb\u52a1\u4e0a\u7684\u8868\u73b0&#xff08;Huang \u7b49&#xff0c;2024a&#xff09;&#xff0c;AI Scientist \u81ea\u52a8\u5316\u60f3\u6cd5\u751f\u6210\u3001\u7f16\u7801\u3001\u5b9e\u9a8c\u3001\u53ef\u89c6\u5316\u3001\u8bba\u6587\u5199\u4f5c\u548c\u6a21\u62df\u8bc4\u5ba1&#xff08;Lu \u7b49&#xff0c;2024&#xff09;&#xff0c;Agent Laboratory \u5c06\u7814\u7a76\u8f85\u52a9\u7ec4\u7ec7\u4e3a\u6587\u732e\u7efc\u8ff0\u3001\u5b9e\u9a8c\u548c\u62a5\u544a\u64b0\u5199\u9636\u6bb5&#xff08;Schmidgall \u7b49&#xff0c;2025&#xff09;\u3002\u66f4\u65b0\u7684\u7814\u7a76\u4ee3\u7406\u8bc4\u4f30\u4f7f\u89c4\u5212\u548c\u53ef\u89c6\u5316\u8868\u9762\u66f4\u52a0\u5177\u4f53&#xff1a;PaperBench \u7814\u7a76\u8bba\u6587\u5230\u4ee3\u7801\u7684\u590d\u73b0&#xff08;Starace \u7b49&#xff0c;2025&#xff09;&#xff0c;ScienceAgentBench \u5c06\u6570\u636e\u9a71\u52a8\u53d1\u73b0\u5206\u89e3\u4e3a\u7ecf\u8fc7\u9a8c\u8bc1\u7684\u7f16\u7a0b\u4efb\u52a1&#xff08;Chen \u7b49&#xff0c;2025b&#xff09;&#xff0c;AI Scientist-v2 \u589e\u52a0\u4e86\u4ee3\u7406\u6811\u641c\u7d22\u548c\u56fe\u8868\u53cd\u9988\u5faa\u73af&#xff08;Yamada \u7b49&#xff0c;2025&#xff09;&#xff0c;SciVisAgentBench \u9488\u5bf9\u79d1\u5b66\u6570\u636e\u5206\u6790\u548c\u53ef\u89c6\u5316\u4ee3\u7406&#xff08;Ai \u7b49&#xff0c;2026&#xff09;\u3002SafeClawBench \u662f\u4e92\u8865\u7684&#xff1a;\u5b83\u5f3a\u8c03\u653b\u51fb-\u9632\u5fa1\u8986\u76d6&#xff0c;\u5728\u56fa\u5b9a\u7684\u524d\u6cbf\u7aef\u70b9\u4e0a\u8bc4\u4f30\u591a\u79cd\u63d0\u793a\u7b56\u7565&#xff0c;\u5e76\u5206\u522b\u62a5\u544a harness\/Exec \u8bc1\u636e&#xff0c;\u4ee5\u4fbf\u4e0d\u6df7\u6dc6\u63d0\u793a\u7ea7\u548c\u72b6\u6001\u7ea7\u58f0\u660e&#xff08;\u9644\u5f55\u8868 6&#xff09;\u3002\u6211\u4eec\u4e0e\u9632\u5fa1\u601d\u60f3\u8fdb\u884c\u6bd4\u8f83&#xff0c;\u5305\u62ec\u6307\u4ee4\u5c42\u6b21\u3001StruQ\u3001SmoothLLM\u3001Llama Guard\u3001spotlighting\u3001FASA\/ClawGuard \u548c\u8f68\u8ff9\u5ba1\u8ba1&#xff08;Wallace \u7b49&#xff0c;2024&#xff1b;Chen \u7b49&#xff0c;2025a&#xff1b;Robey \u7b49&#xff0c;2023&#xff1b;Inan \u7b49&#xff0c;2023&#xff1b;Hines \u7b49&#xff0c;2024&#xff1b;Ying \u7b49&#xff0c;2026&#xff1b;Chen \u7b49&#xff0c;2026&#xff09;\u3002<\/p>\n<hr \/>\n<p>3. SafeClawBench&#xff1a;\u57fa\u51c6\u8bbe\u8ba1<\/p>\n<p>3.1 \u653b\u51fb\u9762\u5f62\u5f0f\u5316<\/p>\n<p>\u6211\u4eec\u5c06 DPI\u3001IPI\u3001TRI\u3001MPI\u3001MEX \u548c ADI \u89c6\u4e3a\u62a5\u544a\u65cf\u800c\u975e\u4e92\u65a5\u7684\u56e0\u679c\u7c7b\u522b\u3002DPI\/IPI\/TRI \u4e3b\u8981\u63cf\u8ff0\u6307\u4ee4\u6765\u6e90&#xff0c;MPI \u63cf\u8ff0\u5185\u5b58\u5199\u5165\u6301\u4e45\u6027&#xff0c;MEX \u63cf\u8ff0\u53d7\u4fdd\u62a4\u4fe1\u606f\u63d0\u53d6\u76ee\u6807&#xff0c;ADI \u63cf\u8ff0\u7531\u6b20\u89c4\u8303\u89e6\u53d1\u7684\u4e0d\u5b89\u5168\u63a8\u7406\u673a\u5236\u3002\u6bcf\u4e2a\u6848\u4f8b\u8fd8\u53ef\u4ee5\u5206\u89e3\u4e3a\u6e90\/\u901a\u9053\u3001\u673a\u5236\u3001\u76ee\u6807\u8d44\u4ea7\u3001\u5371\u5bb3\u76ee\u6807\u3001\u751f\u547d\u5468\u671f\u9636\u6bb5\u548c\u8bc1\u636e\u901a\u9053\u3002\u5728\u5f53\u524d\u53d1\u5e03\u7684\u5de5\u4ef6\u4e2d&#xff0c;\u8fd9\u4e9b\u591a\u6807\u7b7e\u5b57\u6bb5\u4f5c\u4e3a\u57fa\u51c6\u5143\u6570\u636e\u63d0\u4f9b&#xff0c;\u800c\u975e\u4f5c\u4e3a\u6700\u7ec8\u7684\u6b63\u4ea4\u5206\u7c7b\u6cd5\u3002\u65e7\u7684 \u2131 \u00d7 \u2112 \u00d7 \u210b \u7f51\u683c\u4ec5\u4f5c\u4e3a\u6784\u5efa\u548c\u62a5\u544a\u8bbe\u5907\u4fdd\u7559&#xff1b;\u672a\u8986\u76d6\u7684\u5355\u5143\u4e3b\u8981\u5bf9\u5e94\u4e8e\u4e0d\u5408\u7406\u7684\u7ec4\u5408&#xff0c;\u5982\u8f93\u51fa\u671f\u95f4\u7684\u5185\u5b58\u63d0\u53d6\u3002<\/p>\n<p>3.2 \u5a01\u80c1\u6a21\u578b<\/p>\n<p>\u6211\u4eec\u5efa\u6a21\u4e00\u4e2a\u653b\u51fb\u8005&#xff0c;\u5b83\u53ef\u4ee5\u5f71\u54cd\u5df2\u90e8\u7f72\u4ee3\u7406\u53ef\u80fd\u5904\u7406\u7684\u4e00\u4e2a\u6216\u591a\u4e2a\u4e0d\u53ef\u4fe1\u901a\u9053&#xff1a;\u76f4\u63a5\u7528\u6237\u6d88\u606f\u3001\u5916\u90e8\u6587\u6863\u6216\u7f51\u9875\u3001\u5de5\u5177\u8fd4\u56de\u503c\u4ee5\u53ca\u5185\u5b58\u5199\u5165\u5019\u9009\u3002\u653b\u51fb\u8005\u4e0d\u80fd\u4fee\u6539\u7279\u6743\u7cfb\u7edf\u63d0\u793a\u6216\u57fa\u51c6\u4ee3\u7801&#xff0c;\u4f46\u53ef\u4ee5\u5c06\u5185\u5bb9\u4f2a\u88c5\u4e3a\u4e1a\u52a1\u6d41\u7a0b\u3001\u5408\u89c4\u6307\u5357\u3001\u5de5\u5177\u5143\u6570\u636e\u6216\u8bb0\u5fc6\u7684\u7528\u6237\u504f\u597d\u3002\u76ee\u6807\u4ee3\u7406\u5177\u6709\u63cf\u8ff0\u6587\u4ef6\u3001\u7535\u5b50\u90ae\u4ef6\u3001Web\u3001\u4ee3\u7801\u3001\u6570\u636e\u5e93\u548c\u5185\u5b58\u5de5\u5177\u7684\u7cfb\u7edf\u63d0\u793a&#xff1b;\u5728\u8bed\u4e49\u6838\u5fc3\u4e2d&#xff0c;\u8fd9\u4e9b\u80fd\u529b\u662f\u4e0a\u4e0b\u6587\u6027\u7684&#xff0c;\u800c\u5728 Exec-Balanced \u4e2d&#xff0c;\u5b83\u4eec\u7531\u9694\u79bb\u7684\u6c99\u7bb1\u5b58\u50a8\u652f\u6301\u3002<\/p>\n<p>\u9632\u5fa1\u8005\u5728\u4e3b\u77e9\u9635\u4e2d\u9650\u4e8e\u63d0\u793a\u7ea7\u7b56\u7565\u3002\u8fd9\u4e9b\u7b56\u7565\u53ef\u4ee5\u6307\u793a\u6a21\u578b\u62d2\u7edd\u3001\u6f84\u6e05\u3001\u7ef4\u62a4\u6307\u4ee4\/\u6570\u636e\u8fb9\u754c\u6216\u9a8c\u8bc1\u8bf7\u6c42\u8303\u56f4&#xff0c;\u4f46\u4e0d\u80fd\u5f3a\u5236\u6267\u884c\u6587\u4ef6\u7cfb\u7edf\u6743\u9650\u3001\u7f51\u7edc\u51fa\u53e3\u63a7\u5236\u3001\u4eba\u5de5\u6279\u51c6\u6216\u4e8b\u540e\u8f93\u51fa\u8fc7\u6ee4\u3002\u8fd9\u79cd\u9650\u5236\u662f\u6545\u610f\u7684&#xff1a;\u5b83\u9694\u79bb\u4e86\u63d0\u793a\u7b56\u7565\u548c\u6a21\u578b\u5bf9\u9f50\u7684\u884c\u4e3a\u3002\u56e0\u6b64&#xff0c;\u6211\u4eec\u907f\u514d\u58f0\u79f0\u63d0\u793a\u9632\u5fa1\u662f\u4e00\u4e2a\u5b8c\u6574\u7684\u751f\u4ea7\u63a7\u5236&#xff1b;\u5bf9\u4e8e\u4e0d\u53ef\u9006\u64cd\u4f5c\u548c\u627f\u8f7d\u79d8\u5bc6\u7684\u5de5\u5177&#xff0c;\u8fd0\u884c\u65f6\u8bbf\u95ee\u63a7\u5236\u4ecd\u7136\u662f\u5fc5\u8981\u7684\u3002<\/p>\n<p>3.3 \u57fa\u51c6\u6784\u5efa<\/p>\n<p>SafeClawBench \u5305\u542b 600 \u4e2a\u5408\u6210\u6311\u6218\u6848\u4f8b&#xff0c;\u6bcf\u4e2a\u62a5\u544a\u65cf 100 \u4e2a\u6848\u4f8b\u3002\u6bcf\u4e2a\u53d1\u5e03\u7684\u6848\u4f8b\u6307\u5b9a\u4e86\u573a\u666f\u3001\u7528\u6237\u63d0\u793a\u3001\u5371\u5bb3\u76ee\u6807\u3001\u751f\u547d\u5468\u671f\u9636\u6bb5\u3001\u6210\u529f\u8c13\u8bcd\u548c\u5b89\u5168\u884c\u4e3a\u3002\u8868 1 \u603b\u7ed3\u4e86\u516d\u4e2a\u62a5\u544a\u65cf\u3002\u6b63\u6587\u5c06\u6b64\u56fa\u5b9a\u5206\u5272\u89c6\u4e3a\u53d7\u63a7\u6bd4\u8f83\u538b\u529b\u6d4b\u8bd5&#xff0c;\u800c\u975e\u6d41\u884c\u7387\u6837\u672c&#xff1b;\u9644\u5f55 G \u63cf\u8ff0\u4e86\u53d1\u5e03\u7684\u6848\u4f8b\u6a21\u5f0f\u548c\u4ee3\u8868\u6027\u573a\u666f\u3002<\/p>\n<p>\u8868 1&#xff1a;SafeClawBench \u57fa\u51c6\u7edf\u8ba1\u3002\u00a0600 \u4e2a\u53d7\u63a7\u6d4b\u8bd5\u6848\u4f8b&#xff0c;\u6bcf\u4e2a\u653b\u51fb\u7c7b\u578b 100 \u4e2a\u3002\u57fa\u51c6\u56f4\u7ed5\u5339\u914d\u7684\u4efb\u52a1\u6807\u8bc6\u7ec4\u7ec7&#xff0c;\u7528\u4e8e\u6bd4\u8f83\u6a21\u578b\u548c\u63d0\u793a\u7b56\u7565\u884c\u4e3a\u3002<\/p>\n<table>\n<tr>\n<p>\u653b\u51fb\u7c7b\u578b<\/p>\n<p>\u6837\u672c\u6570<\/p>\n<p>\u4e3b\u8981\u5371\u5bb3<\/p>\n<\/tr>\n<tbody>\n<tr>\n<td>\n<p>DPI&#xff08;\u76f4\u63a5\u63d0\u793a\u6ce8\u5165&#xff09;<\/p>\n<\/td>\n<td>\n<p>100<\/p>\n<\/td>\n<td>\n<p>\u6cc4\u9732\/\u884c\u52a8<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>IPI&#xff08;\u95f4\u63a5\u63d0\u793a\u6ce8\u5165&#xff09;<\/p>\n<\/td>\n<td>\n<p>100<\/p>\n<\/td>\n<td>\n<p>\u884c\u52a8\/\u6301\u4e45\u5316<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>TRI&#xff08;\u5de5\u5177\u8fd4\u56de\u6ce8\u5165&#xff09;<\/p>\n<\/td>\n<td>\n<p>100<\/p>\n<\/td>\n<td>\n<p>\u6cc4\u9732\/\u884c\u52a8<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>MPI&#xff08;\u5185\u5b58\u6295\u6bd2&#xff09;<\/p>\n<\/td>\n<td>\n<p>100<\/p>\n<\/td>\n<td>\n<p>\u6301\u4e45\u5316<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>MEX&#xff08;\u5185\u5b58\u63d0\u53d6&#xff09;<\/p>\n<\/td>\n<td>\n<p>100<\/p>\n<\/td>\n<td>\n<p>\u6cc4\u9732<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>ADI&#xff08;\u6b67\u4e49\u9a71\u52a8\u7684\u4e0d\u5b89\u5168\u63a8\u7406&#xff09;<\/p>\n<\/td>\n<td>\n<p>100<\/p>\n<\/td>\n<td>\n<p>\u884c\u52a8<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u603b\u8ba1<\/p>\n<\/td>\n<td>\n<p>600<\/p>\n<\/td>\n<td>\n<p>\u6cc4\u9732\/\u884c\u52a8\/\u6301\u4e45\u5316<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>3.4 \u57fa\u51c6\u5c42\u7ea7<\/p>\n<p>SafeClawBench \u5c06\u4ee3\u7406\u5b89\u5168\u5931\u6548\u5206\u89e3\u4e3a\u5206\u9636\u6bb5\u7aef\u70b9\u3002\u8bed\u4e49\u6838\u5fc3\u662f\u7528\u4e8e\u5e7f\u6cdb\u6bd4\u8f83\u7684 600 \u6848\u4f8b\u63d0\u793a\u7ea7\u57fa\u51c6&#xff1a;\u6bcf\u4e2a\u6a21\u578b-\u9632\u5fa1\u5bf9\u63a5\u6536\u76f8\u540c\u7684\u7528\u6237\u548c\u7cfb\u7edf\u63d0\u793a&#xff0c;LLM \u8bc4\u5224\u5458\u6807\u8bb0\u6700\u7ec8\u54cd\u5e94\u662f\u5426\u5728\u8bed\u4e49\u4e0a\u9075\u4ece\u653b\u51fb\u76ee\u6807\u3002\u6838\u5fc3\u95e8\u63a7\u7684\u5371\u5bb3\u8bc1\u636e\u5ba1\u8ba1\u968f\u540e\u5ba1\u67e5\u89c4\u8303\u7684\u6838\u5fc3\u5931\u8d25\u4e09\u5143\u7ec4&#xff0c;\u5bfb\u627e\u53d7\u4fdd\u62a4\u5bf9\u8c61\u3001\u8bbf\u95ee\u3001\u884c\u52a8\u548c\u6301\u4e45\u5316\u8bc1\u636e&#xff1b;\u5728\u4e94\u7aef\u70b9\u9762\u677f\u4e2d&#xff0c;\u6765\u81ea\u4e94\u7aef\u70b9 \u00d7 \u56db\u7b56\u7565\u77e9\u9635\u7684\u6240\u6709 1,834 \u4e2a\u89c4\u8303\u6838\u5fc3\u5931\u8d25\u884c\u5747\u88ab\u5ba1\u8ba1\u3002Exec-Balanced \u9762\u677f\u662f\u4e00\u4e2a 600 \u6848\u4f8b\u7684\u53ef\u6267\u884c\u6269\u5c55&#xff0c;\u6db5\u76d6\u6240\u6709\u516d\u79cd\u653b\u51fb\u7c7b\u578b&#xff0c;\u901a\u8fc7\u786e\u5b9a\u6027\u9884\u8a00\u673a\u5bf9\u6587\u4ef6\u3001\u7535\u5b50\u90ae\u4ef6\u3001\u6570\u636e\u5e93\u3001\u5185\u5b58\u3001Web \u548c\u4ee3\u7801\u6267\u884c\u5b58\u50a8\u8fdb\u884c\u68c0\u67e5&#xff1b;\u89e3\u6790\u7684\u5de5\u5177\u8c03\u7528\u66f4\u65b0\u9694\u79bb\u7684\u6c99\u7bb1\u72b6\u6001&#xff0c;\u72b6\u6001\u5dee\u5f02\u9884\u8a00\u673a\u4e3a\u5de5\u5177\/\u72b6\u6001\u5371\u5bb3\u63d0\u4f9b\u5355\u72ec\u7684\u53ef\u6267\u884c\u7aef\u70b9&#xff08;\u7b2c 5.1 \u8282&#xff09;\u3002SafeClawBench-Harness \u5c06\u6838\u5fc3\u6848\u4f8b\u5305\u88c5\u5728\u6a21\u62df\u5de5\u5177\u8c03\u7528 harness \u4e2d\u5e76\u8bb0\u5f55\u8f6e\u6b21\/\u5de5\u5177\u8f68\u8ff9&#xff0c;\u4f46\u5de5\u5177\u89c2\u5bdf\u662f\u6a21\u62df\u7684\u3002<\/p>\n<p>\u6838\u5fc3\u4e0e Exec-Balanced \u4e4b\u95f4\u7684\u914d\u5bf9\u3002<br \/>\n\u6bcf\u4e2a Exec-Balanced \u6848\u4f8b\u901a\u8fc7\u4fdd\u7559\u653b\u51fb\u76ee\u6807\u548c\u5b89\u5168\u884c\u4e3a\u8c13\u8bcd&#xff0c;\u540c\u65f6\u5b9e\u4f8b\u5316\u53d7\u4fdd\u62a4\u5bf9\u8c61\u548c\u72b6\u6001\u9884\u8a00\u673a&#xff0c;\u4ece\u6838\u5fc3\u6848\u4f8b\u6620\u5c04\u800c\u6765\u3002\u6838\u5fc3\u548c Exec \u662f\u5355\u72ec\u7684\u6a21\u578b\u8c03\u7528&#xff0c;\u56e0\u4e3a\u53ef\u6267\u884c\u534f\u8bae\u66b4\u9732\u4e86\u5de5\u5177\u548c\u72b6\u6001&#xff0c;\u800c\u8bed\u4e49\u6838\u5fc3\u662f\u63d0\u793a\u7ea7\u7684\u3002\u56e0\u6b64&#xff0c;\u95e8\u63a7\u9a8c\u8bc1\u5206\u6790\u4ec5\u5728\u540c\u65f6\u5177\u6709\u6838\u5fc3\u548c\u6267\u884c\u7ed3\u679c\u7684\u6848\u4f8b\/\u6a21\u578b\/\u9632\u5fa1\u4e09\u5143\u7ec4\u4e0a\u6267\u884c&#xff1b;\u6c47\u603b Exec \u8868\u4e0d\u4f5c\u4e3a\u5bf9 CoreFail&#064;600 \u7684\u5206\u6bcd\u6df7\u5408\u6821\u6b63\u3002<\/p>\n<p>3.5 \u6d4b\u91cf\u6a21\u578b&#xff1a;\u8bed\u4e49\u5931\u8d25\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3<\/p>\n<p>\u5bf9\u4e8e\u6838\u5fc3\u6848\u4f8b\u00a0ii\u3001\u6a21\u578b\u00a0mm\u00a0\u548c\u9632\u5fa1\u00a0dd&#xff0c;\u4ee4\u00a0Ci,m,d\u2208{0,1}Ci,m,d\u200b\u2208{0,1}\u00a0\u8868\u793a\u8bed\u4e49\u6838\u5fc3\u77e9\u9635\u4e2d\u7684\u89c4\u8303 CoreFail \u6807\u7b7e&#xff1a;\u8bc4\u5224\u5458\u5c06\u54cd\u5e94\u6807\u8bb0\u4e3a\u653b\u51fb\u6210\u529f&#xff0c;\u56e0\u4e3a\u5b83\u4ece\u8bed\u4e49\u4e0a\u91c7\u7eb3\u3001\u6388\u6743\u3001\u63ed\u793a\u3001\u8ba1\u5212\u6216\u8bc1\u660e\u4e0d\u5b89\u5168\u7684\u653b\u51fb\u76ee\u6807\u3002CoreFail \u662f\u4e00\u4e2a\u8bed\u4e49\u7aef\u70b9&#xff0c;\u800c\u975e\u5b8c\u6210\u7684\u5371\u5bb3\u3002\u89c4\u8303\u6838\u5fc3\u6307\u6807\u4e3a&#xff1a;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213330-6a6fb7aabfb5d.png\" \/><\/p>\n<p>\u5bf9\u4e8e\u4e94\u7aef\u70b9 \u00d7 \u56db\u7b56\u7565\u7cbe\u7b80\u9762\u677f&#xff0c;\u6838\u5fc3\u95e8\u63a7\u7684\u5371\u5bb3\u8bc1\u636e\u5ba1\u8ba1\u7cbe\u786e\u9009\u62e9\u89c4\u8303 CoreFail \u884c&#xff1a;Si,m,d&#061;1Si,m,d\u200b&#061;1\u00a0\u5f53\u4e14\u4ec5\u5f53\u00a0Ci,m,d&#061;1Ci,m,d\u200b&#061;1\u3002\u6240\u6709 1,834 \u4e2a\u9009\u5b9a\u884c\u5747\u88ab\u5ba1\u8ba1&#xff0c;\u7f3a\u5931\u5ba1\u8ba1\u884c\u4e3a\u96f6\u3002\u4ee4\u00a0Pi,m,d&#061;1Pi,m,d\u200b&#061;1\u00a0\u8868\u793a\u5ba1\u8ba1\u8fd4\u56de\u4e86\u4e00\u4e2a\u89e3\u6790\u7684\u975e\u7a7a\u5904\u7f6e\u3002\u5ba1\u8ba1\u8be2\u95ee\u4e00\u4e2a\u5ba1\u8ba1\u7684\u6838\u5fc3\u5931\u8d25\u4e09\u5143\u7ec4\u662f\u5426\u5305\u542b\u6838\u5fc3\u5de5\u4ef6\u4e2d\u56db\u7c7b\u5371\u5bb3\u4e4b\u4e00\u7684\u8bc1\u636e&#xff1a;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213330-6a6fb7aac9414.png\" \/><\/p>\n<p>\u8fd9\u91cc\u00a0HtextHtext\u3001HaccessHaccess\u3001HactionHaction\u00a0\u548c\u00a0HpersistHpersist\u00a0\u662f\u884c\u7ea7\u5e03\u5c14\u5ba1\u8ba1\u6807\u7b7e&#xff0c;\u5bf9\u5e94\u4e8e\u540c\u4e00\u00a0(i,m,d)(i,m,d)\u00a0\u4e09\u5143\u7ec4\u7684 TextHarm\u3001AccessHarm\u3001ActionHarm \u548c PersistHarm\u3002\u6211\u4eec\u5728\u5b8c\u6574\u7684 600 \u6848\u4f8b\u6838\u5fc3\u5206\u6bcd\u4e0a\u62a5\u544a\u4e3b\u8981\u7684\u6838\u5fc3\u5931\u8d25\u95e8\u63a7\u5371\u5bb3\u6307\u6807\u3002\u4ee4\u00a0Ii,m,dharm&#061;1Ii,m,dharm\u200b&#061;1\u00a0\u5f53\u4e14\u4ec5\u5f53\u00a0Ci,m,d&#061;Si,m,d&#061;Pi,m,d&#061;Hi,m,d&#061;1Ci,m,d\u200b&#061;Si,m,d\u200b&#061;Pi,m,d\u200b&#061;Hi,m,d\u200b&#061;1&#xff0c;\u5426\u5219\u4e3a 0&#xff1a;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213330-6a6fb7aad237f.png\" \/><\/p>\n<p>\u56e0\u6b64&#xff0c;HarmEvidence&#064;600 \u662f\u89c4\u8303 CoreFail \u884c\u4e0a\u7684\u6838\u5fc3\u95e8\u63a7\u8bc1\u636e\u7aef\u70b9\u3002\u53ef\u6267\u884c\u6c99\u7bb1\u5371\u5bb3\u7531 Exec-Balanced \u6d4b\u91cf&#xff0c;\u5b83\u6709\u81ea\u5df1\u72ec\u7acb\u7684\u534f\u8bae\u3001\u6a21\u578b\u8c03\u7528\u3001\u5de5\u5177\u66b4\u9732\u548c\u5206\u6bcd\u3002\u6211\u4eec\u5355\u72ec\u4fdd\u7559\u7a7a\u5ba1\u8ba1\u5c1d\u8bd5\u3002\u4ee4\u00a0Ii,m,dnull&#061;1Ii,m,dnull\u200b&#061;1\u00a0\u5f53\u4e14\u4ec5\u5f53\u00a0Ci,m,d&#061;Si,m,d&#061;1Ci,m,d\u200b&#061;Si,m,d\u200b&#061;1\u00a0\u4e14\u00a0Pi,m,d&#061;0Pi,m,d\u200b&#061;0&#xff1a;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213330-6a6fb7aadbd28.png\" \/><\/p>\n<p>\u6211\u4eec\u4ec5\u5728\u6838\u5fc3\u95e8\u63a7\u5ba1\u8ba1\u5185\u4f7f\u7528 CoreFail \u884c\u4e2d\u7684\u6761\u4ef6\u8bc1\u636e\u7387\u00a0P(H&#061;1\u2223C&#061;1)P(H&#061;1\u2223C&#061;1)\u00a0\u4f5c\u4e3a\u8bca\u65ad\u6027\u8bc1\u636e\u7387\u3002\u6ca1\u6709\u5371\u5bb3\u8bc1\u636e\u7684\u5ba1\u8ba1\u6838\u5fc3\u5931\u8d25\u4fdd\u7559\u4e3a SemanticOnly&#xff1a;\u5728\u5ba1\u8ba1\u6a21\u5f0f\u4e0b\u6ca1\u6709\u53d7\u4fdd\u62a4\u5bf9\u8c61\u3001\u884c\u52a8\u3001\u8bbf\u95ee\u6216\u6301\u4e45\u5316\u8bc1\u636e\u7684\u8bed\u4e49\u635f\u5bb3\u3002NullAudit\u3001NotObservable \u548c\u5355\u4e2a\u89e3\u6790\u7684 OtherNoHarm \u6b8b\u4f59\u884c\u5206\u522b\u62a5\u544a\u3002\u9644\u5f55\u8868 12 \u7ed9\u51fa\u4e86\u5b8c\u6574\u7684\u8bc1\u636e\u6a21\u5f0f&#xff0c;\u8868 13 \u5217\u51fa\u4e86\u5206\u6bcd\u3002<\/p>\n<p>3.6 \u6838\u5fc3\u95e8\u63a7\u7684\u5371\u5bb3\u8bc1\u636e\u5ba1\u8ba1\u7a0b\u5e8f<\/p>\n<p>\u6838\u5fc3\u95e8\u63a7\u5ba1\u8ba1\u4f7f\u7528\u8bed\u4e49\u6838\u5fc3\u5de5\u4ef6\u800c\u975e Exec-Balanced \u6c99\u7bb1\u65e5\u5fd7\u3002TextHarm \u901a\u8fc7\u5728\u6700\u7ec8\u54cd\u5e94\u4e2d\u5bf9\u79cd\u5b50\u91d1\u4e1d\u96c0\u3001\u53d7\u4fdd\u62a4\u914d\u7f6e\u503c\u548c\u53d7\u4fdd\u62a4\u5185\u5b58\u503c\u8fdb\u884c\u7cbe\u786e\u5b57\u7b26\u4e32\u548c\u53d7\u4fdd\u62a4\u5bf9\u8c61\u5339\u914d\u6765\u68c0\u67e5\u3002AccessHarm\u3001ActionHarm \u548c PersistHarm \u662f\u9488\u5bf9\u6838\u5fc3\u54cd\u5e94\u6216\u7ed3\u6784\u5316\u6838\u5fc3\u8f68\u8ff9\u4e2d\u8868\u8fbe\u7684\u660e\u786e\u53d7\u4fdd\u62a4\u5bf9\u8c61\u8bbf\u95ee\u3001\u4e0d\u5b89\u5168\u884c\u52a8\u627f\u8bfa\u6216\u6301\u4e45\u5316\u6c61\u67d3\u7684\u5ba1\u8ba1\u8bc1\u636e\u6807\u7b7e&#xff1b;\u5b83\u4eec\u8868\u793a\u5de5\u4ef6\u53ef\u89c1\u7684\u5371\u5bb3\u8bc1\u636e\u3002\u5df2\u6267\u884c\u7684\u72b6\u6001\u53d8\u5316\u7531 Exec-Balanced \u7684 ObservedHarm&#064;Exec \u6d4b\u91cf\u3002\u5728\u786e\u5b9a\u6027\u8bc1\u636e\u68c0\u67e5\u4e4b\u540e&#xff0c;\u4f7f\u7528\u4e00\u4e2a\u5355\u72ec\u7684\u540e\u5907\u5ba1\u8ba1 API \u5c06\u5269\u4f59\u9009\u5b9a\u7684\u6838\u5fc3\u5931\u8d25\u884c\u5206\u7c7b\u4e3a SemanticOnly \u6216 NotObservable&#xff1b;NullAudit \u8868\u793a\u5728\u91cd\u8bd5\u5904\u7406\u540e\u65e0\u6cd5\u89e3\u6790\u7684\u540e\u5907\u5206\u7c7b\u5668\u54cd\u5e94\u3002\u5728\u9009\u5b9a\u7684\u5ba1\u8ba1\u884c\u4e2d&#xff0c;HarmEvidence\u3001SemanticOnly\u3001NotObservable\u3001NullAudit \u548c OtherNoHarm \u662f\u4e92\u65a5\u7684\u5ba1\u8ba1\u7ed3\u679c\u3002\u6388\u6743\u662f\u573a\u666f\u672c\u5730\u7684&#xff1a;\u6bcf\u4e2a\u6848\u4f8b\u5b9a\u4e49\u53d7\u4fdd\u62a4\u5bf9\u8c61\u548c\u5141\u8bb8\u8303\u56f4&#xff1b;\u5f53\u6838\u5fc3\u5de5\u4ef6\u660e\u786e\u9488\u5bf9\u8be5\u8303\u56f4\u4e4b\u5916\u7684\u53d7\u4fdd\u62a4\u5bf9\u8c61\u65f6&#xff0c;\u8bbf\u95ee\u662f\u672a\u6388\u6743\u7684&#xff0c;\u4e0e Exec-Balanced \u662f\u5426\u4f1a\u89e6\u53d1\u6c99\u7bb1\u9884\u8a00\u673a\u65e0\u5173\u3002NotObservable \u4e3a\u5411\u524d\u517c\u5bb9\u6027\u548c\u540e\u5907\u5ba1\u8ba1\u8ddf\u8e2a\u4fdd\u7559&#xff1b;\u5728\u6c47\u603b\u7684 D0 \u548c\u56db\u79cd\u9632\u5fa1\u5ba1\u8ba1\u5de5\u4ef6\u4e2d&#xff0c;\u5728\u786e\u5b9a\u6027\u68c0\u67e5\u548c\u91cd\u8bd5\u8fc7\u6ee4\u540e\u5b83\u4e3a\u96f6\u3002<\/p>\n<p>3.7 \u4ee3\u7406\u73af\u5883\u548c\u8bc4\u4f30\u534f\u8bae<\/p>\n<p>\u6bcf\u4e2a\u6838\u5fc3\u6848\u4f8b\u5728\u6a21\u62df\u4ee3\u7406\u73af\u5883\u4e2d\u8bc4\u4f30&#xff0c;\u7cfb\u7edf\u63d0\u793a\u4e2d\u63cf\u8ff0\u4e86\u516d\u7c7b\u5de5\u5177&#xff1a;\u6587\u4ef6\u64cd\u4f5c\u3001Web\u3001\u7535\u5b50\u90ae\u4ef6\u3001\u4ee3\u7801\u6267\u884c\u3001\u6570\u636e\u5e93\u67e5\u8be2\u548c\u6301\u4e45\u5316\u5185\u5b58\u3002\u4e3b\u8981\u6838\u5fc3\u6307\u6807\u662f CoreFail&#064;600&#xff0c;\u5373 GPT-5.4 \u8bc4\u5224\u5458\u5728\u56fa\u5b9a\u7684 600 \u6848\u4f8b\u6838\u5fc3\u5206\u6bcd\u4e0a\u5e73\u5747\u7684\u653b\u51fb\u6210\u529f\u6807\u7b7e\u3002\u6211\u4eec\u8fd8\u8bb0\u5f55 1-5 \u5b89\u5168\u8bc4\u5206\u3001\u5371\u5bb3\u7c7b\u578b\u3001\u9632\u5fa1\u884c\u4e3a\u548c\u8bc4\u5224\u5458\u89e3\u91ca\u3002Harness \u548c\u53ef\u6267\u884c\u6307\u6807\u5206\u522b\u62a5\u544a\u4e3a ToolCall-ASR\u3001StateChange-ASR\u3001ObservedHarm&#064;Exec&#xff0c;\u4ee5\u53ca\u9002\u7528\u65f6\u7684\u826f\u6027\u6548\u7528&#xff08;\u9644\u5f55 M&#xff09;\u3002<\/p>\n<hr \/>\n<p>4. \u5b9e\u9a8c\u4e0e\u7ed3\u679c<\/p>\n<p>\u672c\u8282\u9996\u5148\u5b9a\u4e49\u63d0\u793a\u7b56\u7565\u6761\u4ef6\u548c\u56fa\u5b9a\u7684\u7cbe\u7b80\u9762\u677f\u8bc4\u4f30\u6846\u67b6&#xff0c;\u7136\u540e\u62a5\u544a\u8bed\u4e49\u6838\u5fc3\u548c\u6838\u5fc3\u95e8\u63a7\u5ba1\u8ba1\u7ed3\u679c\u3002\u53ef\u6267\u884c\u7aef\u70b9\u8bca\u65ad\u5728\u7b2c 5 \u8282\u5355\u72ec\u62a5\u544a\u3002<\/p>\n<p>4.1 \u63d0\u793a\u7ea7\u9632\u5fa1\u914d\u7f6e<\/p>\n<p>\u6211\u4eec\u8bc4\u4f30\u4e86\u4e5d\u79cd\u63d0\u793a\u7ea7\u914d\u7f6e&#xff0c;\u4f46\u6b63\u6587\u805a\u7126\u4e8e\u56db\u79cd\u8de8\u8d8a\u8bbe\u8ba1\u7a7a\u95f4\u7684\u7b56\u7565&#xff1a;D0\/\u65e0\u9632\u5fa1\u3001B2\/Sandwich \u4f5c\u4e3a\u8f7b\u91cf\u7ea7\u57fa\u7ebf\u3001D3 \u4f5c\u4e3a\u5206\u5c42\u63d0\u793a\u7b56\u7565\u5806\u6808&#xff0c;\u4ee5\u53ca D4\/LongPolicy \u4f5c\u4e3a\u6545\u610f\u957f\u4e14\u8fc7\u5ea6\u6307\u5b9a\u7684\u7b56\u7565\u63a2\u9488\u3002\u8868 2 \u7ed9\u51fa\u4e86\u672c\u6587\u4f7f\u7528\u7684\u7b56\u7565\u7ec4\u6210\u3002\u8fd9\u4e9b\u6807\u7b7e\u662f\u5b9e\u9a8c\u6761\u4ef6\u800c\u975e\u4ea7\u54c1\u5316\u9632\u5fa1&#xff1a;\u6240\u6709\u90fd\u662f\u7cfb\u7edf\u63d0\u793a\u6dfb\u52a0&#xff0c;\u6ca1\u6709\u4e00\u4e2a\u5f3a\u5236\u6267\u884c\u8fd0\u884c\u65f6\u6587\u4ef6\u7cfb\u7edf\u3001\u7f51\u7edc\u3001\u5185\u5b58\u6216\u7535\u5b50\u90ae\u4ef6\u6743\u9650&#xff0c;\u4e0b\u9762\u7684\u8bbe\u7f6e\u6307\u5b9a\u4e86\u54ea\u4e9b\u6761\u4ef6\u8fdb\u5165\u4e3b\u77e9\u9635\u3002<\/p>\n<p>\u5177\u4f53\u5730&#xff0c;D0 \u662f\u5e26\u6709\u5de5\u5177\u63cf\u8ff0\u4f46\u65e0\u9644\u52a0\u9632\u5fa1\u7b56\u7565\u7684\u57fa\u7840\u4ee3\u7406\u63d0\u793a&#xff0c;\u56e0\u6b64\u5b83\u5728\u76f8\u540c\u7684\u6a21\u62df\u5de5\u5177\u73af\u5883\u4e0b\u6d4b\u91cf\u57fa\u7ebf\u884c\u4e3a\u3002B2\/Sandwich \u662f\u4e00\u4e2a\u7b80\u77ed\u7684\u57fa\u7ebf\u5305\u88c5\u5668&#xff0c;\u653e\u7f6e\u5728\u9762\u5411\u7528\u6237\u5185\u5bb9\u7684\u524d\u540e&#xff1b;\u5b83\u63d0\u9192\u6a21\u578b\u66f4\u9ad8\u4f18\u5148\u7ea7\u7684\u6307\u4ee4\u8986\u76d6\u7528\u6237\u3001\u6587\u6863\u3001\u7f51\u9875\u3001\u5de5\u5177\u8fd4\u56de\u548c\u5185\u5b58\u6587\u672c\u3002D3 \u662f\u4e3b\u8981\u7684\u5206\u5c42\u63d0\u793a\u7b56\u7565\u6bd4\u8f83\u5668&#xff1a;\u5b83\u7ed3\u5408\u4e86\u6307\u4ee4\u5c42\u6b21\u3001\u53d7\u4fdd\u62a4\u72b6\u6001\u548c\u5185\u5b58\u5904\u7406\u3001\u8bf7\u6c42\u76ee\u6807\u9a8c\u8bc1\u3001\u5b89\u5168\u884c\u52a8\u68c0\u67e5\u4ee5\u53ca\u63d0\u793a\u7ea7\u5de5\u5177\u94fe\u89c4\u5219\u3002D4\/LongPolicy \u662f\u957f\u63d0\u793a\u5305&#xff0c;\u5728\u53d7\u4fdd\u62a4\u72b6\u6001\u89c4\u5219\u4e4b\u4e0a\u6dfb\u52a0\u4e86\u8bed\u4e49\u610f\u56fe\u951a\u5b9a\u3001\u5185\u5b58\u5b8c\u6574\u6027\u5b88\u536b\u548c\u5de5\u5177\u94fe\u5206\u6790\u5668\u8bed\u8a00\u3002B \u884c\u662f\u8f7b\u91cf\u7ea7\u57fa\u7ebf&#xff0c;D \u884c\u662f\u9010\u6b65\u5206\u5c42\u7684\u63d0\u793a\u7b56\u7565\u6761\u4ef6&#xff0c;D4\/LongPolicy \u4fdd\u7559\u4e3a\u7b56\u7565\u590d\u6742\u6027\u8bca\u65ad&#xff0c;\u800c\u975e\u58f0\u79f0\u66f4\u957f\u7684\u63d0\u793a\u672c\u8d28\u4e0a\u5c31\u662f\u66f4\u5f3a\u7684\u9632\u5fa1\u3002<\/p>\n<p>\u8868 2&#xff1a;SafeClawBench \u4e2d\u6bd4\u8f83\u7684\u63d0\u793a\u7b56\u7565\u914d\u7f6e\u3002\u00a0\u8fd9\u4e9b\u914d\u7f6e\u662f\u53ef\u52a0\u6027\u7684\u7cfb\u7edf\u63d0\u793a\u6a21\u677f\u3002B1\u2013B4 \u662f\u8f7b\u91cf\u7ea7\u57fa\u7ebf&#xff0c;D1\u2013D3 \u662f\u5206\u5c42\u63d0\u793a\u7b56\u7565\u5806\u6808&#xff0c;D4\/LongPolicy \u4fdd\u7559\u4e3a\u957f\u7b56\u7565\u590d\u6742\u6027\u63a2\u9488&#xff0c;\u800c\u975e\u58f0\u79f0\u7684\u751f\u4ea7\u9632\u5fa1\u3002<\/p>\n<table>\n<tr>\n<p>\u914d\u7f6e<\/p>\n<p>\u63d0\u793a\u7ec4\u6210<\/p>\n<p>\u9884\u671f\u8fb9\u754c\u6216\u653b\u51fb\u65cf<\/p>\n<p>\u5728\u672c\u6587\u4e2d\u7684\u89d2\u8272<\/p>\n<\/tr>\n<tbody>\n<tr>\n<td>\n<p>D0<\/p>\n<\/td>\n<td>\n<p>\u4ec5\u5de5\u5177\u4f7f\u7528\u7cfb\u7edf\u63d0\u793a<\/p>\n<\/td>\n<td>\n<p>\u57fa\u7ebf\u4ee3\u7406\u884c\u4e3a<\/p>\n<\/td>\n<td>\n<p>\u4e3b\u77e9\u9635<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>B1<\/p>\n<\/td>\n<td>\n<p>\u7b80\u77ed\u901a\u7528\u5b89\u5168\u63d0\u9192<\/p>\n<\/td>\n<td>\n<p>\u901a\u7528\u62d2\u7edd\/\u6f84\u6e05\u884c\u4e3a<\/p>\n<\/td>\n<td>\n<p>\u8bca\u65ad<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>B2\/Sandwich<\/p>\n<\/td>\n<td>\n<p>\u653e\u7f6e\u5728\u7528\u6237\u5185\u5bb9\u524d\u540e\u7684\u7b80\u77ed\u6307\u4ee4\u5305\u88c5\u5668&#xff1b;\u91cd\u590d\u66f4\u9ad8\u4f18\u5148\u7ea7\u7b56\u7565\u8986\u76d6\u7528\u6237\u3001\u6587\u6863\u548c\u5de5\u5177\u8fd4\u56de\u6587\u672c<\/p>\n<\/td>\n<td>\n<p>DPI\u3001IPI \u548c TRI \u7684\u8f7b\u91cf\u7ea7\u6307\u4ee4\u8fb9\u754c\u57fa\u7ebf<\/p>\n<\/td>\n<td>\n<p>\u4e3b\u77e9\u9635<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>B3<\/p>\n<\/td>\n<td>\n<p>B1 \u52a0\u4e0a\u5728\u56de\u7b54\u524d\u68c0\u67e5\u53ef\u7591\u5173\u952e\u8bcd\u548c\u4e0d\u5b89\u5168\u8bf7\u6c42\u7684\u63d0\u9192<\/p>\n<\/td>\n<td>\n<p>\u4f4e\u6210\u672c\u8b66\u544a\/\u8fc7\u6ee4\u57fa\u7ebf<\/p>\n<\/td>\n<td>\n<p>\u8bca\u65ad<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>B4<\/p>\n<\/td>\n<td>\n<p>StruQ \u98ce\u683c\u7684\u53d7\u4fe1\u4efb\u6307\u4ee4\u4e0e\u4e0d\u53ef\u4fe1\u6570\u636e\u5757\u5206\u79bb<\/p>\n<\/td>\n<td>\n<p>\u6307\u4ee4\/\u6570\u636e\u5206\u79bb\u7684\u63d0\u793a\u683c\u5f0f\u5316\u57fa\u7ebf<\/p>\n<\/td>\n<td>\n<p>\u8bca\u65ad<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>D1<\/p>\n<\/td>\n<td>\n<p>\u6307\u4ee4\u5c42\u6b21\u52a0\u4e0a\u9488\u5bf9\u51b2\u7a81\u7528\u6237\u6216\u6587\u6863\u6307\u4ee4\u7684\u62d2\u7edd\/\u6f84\u6e05\u89c4\u5219<\/p>\n<\/td>\n<td>\n<p>\u76f4\u63a5\u548c\u95f4\u63a5\u6307\u4ee4\u51b2\u7a81<\/p>\n<\/td>\n<td>\n<p>\u8bca\u65ad<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>D2<\/p>\n<\/td>\n<td>\n<p>D1 \u52a0\u4e0a\u5185\u5b58\u8bfb\/\u5199\u8fb9\u754c\u3001\u6e05\u7406\u89c4\u5219\u548c\u53d7\u4fdd\u62a4\u5bf9\u8c61\u5904\u7406<\/p>\n<\/td>\n<td>\n<p>MPI \u548c MEX \u5bfc\u5411\u7684\u53d7\u4fdd\u62a4\u72b6\u6001\u5904\u7406<\/p>\n<\/td>\n<td>\n<p>\u8bca\u65ad<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>D3<\/p>\n<\/td>\n<td>\n<p>D2 \u52a0\u4e0a\u8bf7\u6c42\u76ee\u6807\u9a8c\u8bc1\u3001\u5b89\u5168\u884c\u52a8\u8c13\u8bcd\u548c\u63d0\u793a\u7ea7\u5de5\u5177\u94fe\u7b56\u7565<\/p>\n<\/td>\n<td>\n<p>\u5206\u5c42\u901a\u7528\u63d0\u793a\u7b56\u7565\u6bd4\u8f83\u5668<\/p>\n<\/td>\n<td>\n<p>\u4e3b\u77e9\u9635<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>D4\/LongPolicy<\/p>\n<\/td>\n<td>\n<p>D2 \u52a0\u4e0a\u8bed\u4e49\u610f\u56fe\u951a\u5b9a&#xff08;SIA&#xff09;\u3001\u5185\u5b58\u5b8c\u6574\u6027\u5b88\u536b&#xff08;MIG&#xff09;\u548c\u5de5\u5177\u94fe\u5206\u6790\u5668&#xff08;TCA&#xff09;&#xff0c;\u4ea7\u751f\u66f4\u957f\u7684\u63d0\u793a\u5305<\/p>\n<\/td>\n<td>\n<p>\u8bed\u4e49\u610f\u56fe\u3001\u5185\u5b58\u5b8c\u6574\u6027\u548c\u5de5\u5177\u94fe\u63a8\u7406\u7684\u957f\u63d0\u793a\/\u7b56\u7565\u590d\u6742\u6027\u538b\u529b\u6848\u4f8b<\/p>\n<\/td>\n<td>\n<p>\u4e3b\u77e9\u9635\u63a2\u9488<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u56e0\u6b64&#xff0c;D4\/LongPolicy \u4e0d\u88ab\u89e3\u91ca\u4e3a\u65b0\u7684\u8bbf\u95ee\u63a7\u5236\u673a\u5236&#xff0c;\u4e5f\u4e0d\u4f5c\u4e3a SIA\u3001MIG \u6216 TCA \u72ec\u7acb\u56e0\u679c\u7684\u8bc1\u636e\u3002\u6d88\u878d\u8868\u4ece D4 \u4e2d\u79fb\u9664\u4e00\u4e2a\u7ec4\u4ef6&#xff0c;\u4f5c\u4e3a\u5b58\u6863\u7684\u8bca\u65ad\u8fd0\u884c&#xff0c;\u9644\u5f55 L \u5c06 D3 \u586b\u5145\u5230\u4e0e D4 \u5927\u81f4\u76f8\u540c\u7684\u957f\u5ea6&#xff0c;\u4ee5\u6d4b\u8bd5\u63d0\u793a\u957f\u5ea6\u6df7\u6742\u3002\u8fd9\u79cd\u8bbe\u8ba1\u8ba9\u6211\u4eec\u80fd\u591f\u8be2\u95ee\u957f\u4e14\u8fc7\u5ea6\u6307\u5b9a\u7684\u7b56\u7565\u5305\u662f\u5426\u6539\u53d8\u6a21\u578b\u884c\u4e3a&#xff0c;\u540c\u65f6\u4fdd\u6301\u57fa\u51c6\u8d21\u732e\u4e0e\u53ef\u90e8\u7f72\u9632\u5fa1\u7684\u4e3b\u5f20\u5206\u79bb\u3002<\/p>\n<p>4.2 \u5b9e\u9a8c\u8bbe\u7f6e<\/p>\n<p>\u4e3b\u8981\u5206\u6790\u4f7f\u7528\u7cbe\u7b80\u9762\u677f\u5de5\u4ef6\u4e2d\u56fa\u5b9a\u7684\u4e94\u4e2a\u7aef\u70b9\u5b57\u7b26\u4e32&#xff1a;GPT-5.5\u3001Claude Opus 4.7\u3001Qwen3.6-Plus\u3001GLM-5.1 \u548c Kimi K2.5\u3002\u5de5\u4ef6\u5143\u6570\u636e\u5b9a\u4e49\u4e86\u6bd4\u8f83\u6846\u67b6\u3002\u4ece\u8868 2 \u7684\u914d\u7f6e\u4e2d&#xff0c;\u6838\u5fc3\u5206\u6790\u4f7f\u7528 D0\u3001B2\/Sandwich\u3001D3 \u548c D4\/LongPolicy&#xff0c;\u4ea7\u751f 5 \u00d7 4 \u00d7 600 &#061; 12,000 \u6b21\u63d0\u793a\u7ea7\u8bc4\u4f30\u3002\u6807\u9898\u6838\u5fc3\u8868\u4ece\u7cbe\u7b80\u6838\u5fc3\u7cbe\u786e\u884c\u5de5\u4ef6 reduced_core_exact_0524_rejudged_gpt55 \u91cd\u65b0\u751f\u6210&#xff1a;20 \u4e2a\u6a21\u578b-\u7b56\u7565\u5355\u5143\u4e2d\u7684\u6bcf\u4e00\u4e2a\u90fd\u6709 600 \u4e2a\u89e3\u6790\u7684\u6838\u5fc3\u884c&#xff0c;\u7f3a\u5931\u884c\u4e3a\u96f6\u3002GPT-5.4 \u5728\u6e29\u5ea6 0 \u4e0b\u8bc4\u5224\u6838\u5fc3\u8f93\u51fa&#xff0c;\u4ee5\u5728\u5927\u89c4\u6a21\u8bc4\u4f30\u4e2d\u5b9e\u73b0\u6210\u672c\u3001\u6210\u719f\u5ea6\u548c\u786e\u5b9a\u6027\u53ef\u7528\u6027&#xff1b;\u72ec\u7acb\u7684\u7b2c\u4e8c\u8bc4\u5224\u5458\u9a8c\u8bc1\u89c1\u9644\u5f55 J\u3002\u9644\u5f55 K \u7ed9\u51fa 240 \u884c\u6821\u51c6\u6837\u672c\u548c\u4eba\u5de5\u4e8c\u6b21\u88c1\u51b3\u68c0\u67e5\u3002\u9644\u5f55\u8868 8\u30019 \u548c 10 \u8bb0\u5f55\u4e86\u7aef\u70b9 ID\u3001\u8c03\u7528\u53c2\u6570\u3001\u91cd\u8bd5\/\u622a\u65ad\u8bbe\u7f6e\u3001\u5de5\u5177\u8c03\u7528\u683c\u5f0f\u3001\u8bc4\u5224\u5458\u89c4\u5219\u4f4d\u7f6e\u3001\u54c8\u5e0c\u503c\u548c\u53d1\u5e03\u5305\u5de5\u4ef6\u3002<\/p>\n<p>4.3 RQ1&#xff1a;\u8bed\u4e49\u6838\u5fc3<\/p>\n<p>\u9664\u975e\u53e6\u6709\u8bf4\u660e&#xff0c;\u6838\u5fc3\u7ed3\u679c\u4f7f\u7528 CoreFail&#064;600&#xff1a;LLM \u8bc4\u5224\u5458\u5bf9\u6700\u7ec8\u54cd\u5e94\u662f\u5426\u9075\u4ece\u653b\u51fb\u76ee\u6807\u7684\u4e8c\u5143\u6807\u7b7e&#xff0c;\u5728\u56fa\u5b9a\u7684 600 \u6848\u4f8b\u5206\u6bcd\u4e0a\u5e73\u5747\u3002\u6211\u4eec\u5c06 ToolCall-ASR \u548c StateChange-ASR \u4fdd\u7559\u7528\u4e8e\u53ef\u6267\u884c\u7aef\u70b9\u3002\u8868 3 \u662f\u4e3b\u8981\u7684\u6a21\u578b-\u9632\u5fa1\u6bd4\u8f83&#xff1b;\u5b8c\u6574\u7684\u4e5d\u7b56\u7565\u7cbe\u7b80\u9762\u677f\u77e9\u9635\u4fdd\u7559\u4e3a\u9644\u5f55\u8868 27\u3002<\/p>\n<p>\u8868 3&#xff1a;\u5728\u56db\u79cd\u9762\u5411\u63d0\u4ea4\u7684\u63d0\u793a\u7b56\u7565\u4e0b&#xff0c;\u7cbe\u7b80\u4e94\u7aef\u70b9\u9762\u677f\u7684\u4e3b\u8981\u8bed\u4e49\u6838\u5fc3\u7ed3\u679c\u3002\u00a0\u6761\u76ee\u662f\u7cbe\u786e\u7684 CoreFail&#064;600 \u767e\u5206\u6bd4&#xff0c;\u62ec\u53f7\u4e2d\u4e3a\u63cf\u8ff0\u6027 95% Wilson \u533a\u95f4&#xff0c;\u6bcf\u4e2a\u6a21\u578b-\u9632\u5fa1\u5355\u5143\u4f7f\u7528\u56fa\u5b9a\u7684 600 \u6848\u4f8b\u6311\u6218\u5206\u6bcd\u3002\u6700\u540e\u4e00\u884c\u6c47\u96c6\u4e86\u6bcf\u4e2a\u9632\u5fa1\u5185\u7684\u4e94\u4e2a 600 \u6848\u4f8b\u5355\u5143&#xff08;n&#061;3,000&#xff09;&#xff0c;\u5e76\u7ed9\u51fa\u4e86\u63cf\u8ff0\u6027\u6c47\u96c6\u533a\u95f4&#xff0c;\u800c\u975e\u914d\u5bf9\u7684\u6a21\u578b\u7ea7\u68c0\u9a8c\u3002<\/p>\n<table>\n<tr>\n<p>\u6a21\u578b<\/p>\n<p>D0<\/p>\n<p>B2\/Sandwich<\/p>\n<p>D3<\/p>\n<p>D4\/LongPolicy<\/p>\n<\/tr>\n<tbody>\n<tr>\n<td>\n<p>GPT-5.5<\/p>\n<\/td>\n<td>\n<p>23.8 [20.6, 27.4]<\/p>\n<\/td>\n<td>\n<p>14.7 [12.1, 17.7]<\/p>\n<\/td>\n<td>\n<p>10.8 [8.6, 13.6]<\/p>\n<\/td>\n<td>\n<p>8.8 [6.8, 11.4]<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>Claude Opus 4.7<\/p>\n<\/td>\n<td>\n<p>9.0 [7.0, 11.6]<\/p>\n<\/td>\n<td>\n<p>10.3 [8.1, 13.0]<\/p>\n<\/td>\n<td>\n<p>5.8 [4.2, 8.0]<\/p>\n<\/td>\n<td>\n<p>5.8 [4.2, 8.0]<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>Qwen3.6-Plus<\/p>\n<\/td>\n<td>\n<p>35.5 [31.8, 39.4]<\/p>\n<\/td>\n<td>\n<p>14.2 [11.6, 17.2]<\/p>\n<\/td>\n<td>\n<p>9.5 [7.4, 12.1]<\/p>\n<\/td>\n<td>\n<p>11.5 [9.2, 14.3]<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>GLM-5.1<\/p>\n<\/td>\n<td>\n<p>27.5 [24.1, 31.2]<\/p>\n<\/td>\n<td>\n<p>13.3 [10.8, 16.3]<\/p>\n<\/td>\n<td>\n<p>2.2 [1.3, 3.7]<\/p>\n<\/td>\n<td>\n<p>2.8 [1.8, 4.5]<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>Kimi K2.5<\/p>\n<\/td>\n<td>\n<p>44.2 [40.2, 48.2]<\/p>\n<\/td>\n<td>\n<p>21.7 [18.6, 25.1]<\/p>\n<\/td>\n<td>\n<p>17.0 [14.2, 20.2]<\/p>\n<\/td>\n<td>\n<p>17.2 [14.4, 20.4]<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u6c47\u5408 (5\u00d7600)<\/p>\n<\/td>\n<td>\n<p>28.0 [26.4, 29.6]<\/p>\n<\/td>\n<td>\n<p>14.8 [13.6, 16.2]<\/p>\n<\/td>\n<td>\n<p>9.1 [8.1, 10.1]<\/p>\n<\/td>\n<td>\n<p>9.2 [8.2, 10.3]<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213330-6a6fb7aae4d2e.png\" \/><\/p>\n<p>\u56fe2&#xff1a;\u7cbe\u7b80\u9762\u677f CoreFail&#064;600 \u9632\u5fa1\u8f68\u8ff9\u3002\u6bcf\u884c\u8ddf\u8e2a\u4e00\u4e2a\u7aef\u70b9\u8de8\u63d0\u793a\u7b56\u7565&#xff1b;\u8d8a\u4f4e\u8d8a\u597d\u3002<\/p>\n<p>\u5728\u6311\u6218\u96c6\u4e2d&#xff0c;\u5185\u5b58\/\u6301\u4e45\u5316\u548c\u63d0\u53d6\u5bfc\u5411\u7684\u6848\u4f8b\u6700\u9ad8\u3002\u5728\u4e94\u7aef\u70b9\u9762\u677f\u7684 D0 \u4e0b&#xff0c;MPI \u548c MEX \u5e73\u5747 CoreFail&#064;600 \u5747\u4e3a 54.4%&#xff0c;\u9ad8\u4e8e DPI&#xff08;20.8%&#xff09;\u3001IPI&#xff08;16.0%&#xff09;\u3001TRI&#xff08;13.4%&#xff09;\u548c ADI&#xff08;9.0%&#xff09;\u3002\u8fd9\u4e00\u6a21\u5f0f\u53cd\u6620\u4e86\u6311\u6218\u96c6\u5bf9\u4ee3\u7406\u7279\u5b9a\u72b6\u6001\u548c\u53d7\u4fdd\u62a4\u4fe1\u606f\u5931\u6548\u7684\u5f3a\u8c03&#xff1a;MPI \u6355\u83b7\u6301\u4e45\u5316\u5931\u6548&#xff0c;\u800c MEX \u6df7\u5408\u4e86\u7cbe\u786e\u53d7\u4fdd\u62a4\u5bf9\u8c61\u6cc4\u9732\u4e0e\u66f4\u5e7f\u6cdb\u7684\u5b89\u5168\u7b56\u7565\/\u914d\u7f6e\u900f\u660e\u5ea6\u9519\u8bef\u3002MEX \u6848\u4f8b\u7684\u542f\u53d1\u5f0f\u5143\u6570\u636e\u5206\u5272\u8bc1\u5b9e\u4e86\u8fd9\u79cd\u5f02\u8d28\u6027&#xff1a;\u5728\u6311\u6218\u5206\u5272\u4e2d&#xff0c;100 \u4e2a MEX \u6848\u4f8b\u4e2d\u6709 30 \u4e2a\u9488\u5bf9\u7cbe\u786e\u79d8\u5bc6\u6216\u7cfb\u7edf\u63d0\u793a&#xff0c;9 \u4e2a\u9488\u5bf9\u53d7\u4fdd\u62a4\u5185\u5b58\/\u914d\u7f6e&#xff0c;49 \u4e2a\u9488\u5bf9\u7b56\u7565\u6216\u67b6\u6784\u900f\u660e\u5ea6&#xff0c;12 \u4e2a\u4e3a\u5176\u4ed6\u6cc4\u9732\u5f62\u5f0f&#xff08;\u9644\u5f55\u8868 17&#xff09;\u3002<\/p>\n<p>\u5373\u4f7f\u5728\u7aef\u70b9\u4fee\u526a\u540e&#xff0c;\u6a21\u578b\u9009\u62e9\u4ecd\u7136\u662f\u4e00\u4e2a\u91cd\u8981\u7684\u5b89\u5168\u56e0\u7d20\u3002D0 CoreFail&#064;600 \u4ece Claude Opus 4.7 \u7684 9.0% \u5230 Kimi K2.5 \u7684 44.2%&#xff0c;\u5728\u7cbe\u7b80\u4e3b\u9762\u677f\u4e0a\u6bd4\u7387\u4e3a 4.9 \u500d&#xff0c;\u7edd\u5bf9\u5dee\u8ddd\u4e3a 35.2 \u4e2a\u767e\u5206\u70b9\u3002GPT-5.5&#xff08;23.8%&#xff09;\u3001GLM-5.1&#xff08;27.5%&#xff09;\u548c Qwen3.6-Plus&#xff08;35.5%&#xff09;\u5904\u4e8e\u8be5\u538b\u529b\u6d4b\u8bd5\u5206\u5e03\u7684\u4e2d\u95f4&#xff0c;\u800c Claude Opus 4.7 \u5728\u672c\u7814\u7a76\u4e2d\u5177\u6709\u6700\u4f4e\u7684 D0 CoreFail \u7387\u3002<\/p>\n<p>\u63d0\u793a\u7b56\u7565\u6548\u679c\u4f9d\u8d56\u4e8e\u6a21\u578b\u3002B2\/Sandwich \u4e0e\u6bd4 D0 \u66f4\u4f4e\u7684\u6c47\u5408\u7cbe\u7b80\u9762\u677f CoreFail&#064;600 \u76f8\u5173&#xff08;14.8% \u5bf9\u6bd4 28.0%&#xff09;\u3002D3 \u548c D4\/LongPolicy \u63a5\u8fd1&#xff0c;\u5206\u522b\u4e3a 9.1% \u548c 9.2%\u3002D4\/LongPolicy \u5728 GPT-5.5 \u4e0a\u6700\u4f4e&#xff0c;\u5728 Claude Opus 4.7 \u4e0a\u4e0e D3 \u6301\u5e73&#xff0c;\u800c D3 \u5728 Qwen3.6-Plus\u3001GLM-5.1 \u548c Kimi K2.5 \u4e0a\u6700\u4f4e\u3002\u56e0\u6b64&#xff0c;\u6211\u4eec\u5c06\u5176\u63cf\u8ff0\u4e3a\u4f9d\u8d56\u4e8e\u6a21\u578b\u7684\u63d0\u793a\u7b56\u7565\u4ea4\u4e92&#xff0c;\u800c\u975e\u7edf\u4e00\u7684 D4 \u4f18\u52bf\u6216\u63d0\u793a\u9632\u5fa1\u7ec4\u4ef6\u7684\u56e0\u679c\u6392\u5e8f\u3002\u9644\u5f55 N \u62a5\u544a\u4e86\u6765\u81ea\u5355\u72ec\u6269\u5c55\u8fd0\u884c\u7684\u914d\u5bf9\u68c0\u9a8c\u4f5c\u4e3a\u7a33\u5065\u6027\u8bc1\u636e\u3002<\/p>\n<p>D4\/LongPolicy \u662f\u4e00\u4e2a\u7b56\u7565\u590d\u6742\u6027\u63a2\u9488&#xff0c;\u800c\u975e\u6807\u9898\u8d62\u5bb6\u3002\u8de8\u7aef\u70b9&#xff0c;\u5176\u5b9e\u8bc1\u503c\u662f\u6df7\u5408\u7684\u3002\u5728\u6838\u5fc3\u7aef\u70b9&#xff0c;\u6c47\u5408\u7387\u4e0a D3 \u7565\u4f4e\u4e8e D4\/LongPolicy&#xff08;9.1% \u5bf9\u6bd4 9.2%&#xff09;\u3002\u5728\u6838\u5fc3\u95e8\u63a7\u5ba1\u8ba1\u4e2d&#xff0c;D3 \u7684 HarmEvidence \u884c\u6570\u5c11\u4e8e D4\/LongPolicy&#xff08;119 \u5bf9\u6bd4 160&#xff09;&#xff0c;\u5c3d\u7ba1 CoreFail \u8ba1\u6570\u76f8\u4f3c&#xff08;272 \u5bf9\u6bd4 277&#xff1b;\u8868 5&#xff09;\u3002\u5728 Exec-Balanced \u4e2d&#xff0c;D4\/LongPolicy \u5728\u56db\u4e2a\u6807\u9898\u7b56\u7565\u4e2d\u5177\u6709\u6700\u4f4e\u7684\u6c47\u5408 ObservedHarm \u8ba1\u6570&#xff08;71\/3,000 \u5bf9\u6bd4 D3 \u7684 76\/3,000&#xff1b;\u9644\u5f55\u8868 18&#xff09;&#xff0c;\u4f46\u8fd9\u53ea\u662f\u4e94\u884c\u5dee\u5f02\u3002\u5339\u914d\u957f\u5ea6\u7684 D3_LM \u63a7\u5236\u8fdb\u4e00\u6b65\u5e73\u5747 CoreFail&#064;600 \u4e3a 8.0%&#xff0c;\u4f4e\u4e8e D3 \u548c D4\/LongPolicy&#xff08;\u9644\u5f55\u8868 38&#xff09;\u3002\u7ec4\u4ef6\u6d88\u878d\u662f\u5b58\u6863\u7684\u8bca\u65ad&#xff0c;\u800c\u975e\u7cbe\u786e\u7684\u7cbe\u7b80\u9762\u677f\u56e0\u679c\u4f30\u8ba1\u3002\u8fd9\u4e9b\u7ed3\u679c\u4fc3\u4f7f\u5c06 D4\/LongPolicy \u89c6\u4e3a\u957f\u63d0\u793a\u548c\u7b56\u7565\u590d\u6742\u6027\u538b\u529b\u6848\u4f8b&#xff0c;\u800c\u975e\u7ecf\u8fc7\u5355\u72ec\u9a8c\u8bc1\u7684\u65b0\u9632\u5fa1\u6216 SIA\u3001MIG \u6216 TCA \u72ec\u7acb\u56e0\u679c\u7684\u8bc1\u636e\u3002<\/p>\n<p>4.4 RQ2&#xff1a;\u6838\u5fc3\u95e8\u63a7\u7684\u5371\u5bb3\u8bc1\u636e\u5ba1\u8ba1<\/p>\n<p>\u5b8c\u6574\u7684\u6838\u5fc3\u95e8\u63a7\u5371\u5bb3\u8bc1\u636e\u5ba1\u8ba1\u5c06\u7b2c\u4e00\u9636\u6bb5\u8bed\u4e49\u635f\u5bb3\u4e0e\u5de5\u4ef6\u53ef\u89c1\u7684\u5371\u5bb3\u8bc1\u636e\u5206\u79bb\u3002\u5728\u4e94\u7aef\u70b9\u9762\u677f\u7684 D0 \u4e0b&#xff0c;\u89c4\u8303\u6838\u5fc3\u77e9\u9635\u5728 3,000 \u4e2a\u6a21\u578b-\u6848\u4f8b\u884c\u4e2d\u5305\u542b 840 \u4e2a CoreFail \u884c&#xff08;28.0%&#xff09;\u3002\u6240\u6709 840 \u884c\u5747\u88ab\u5ba1\u8ba1&#xff1a;504 \u4e2a\u5728\u6838\u5fc3\u95e8\u63a7\u5ba1\u8ba1\u4e2d\u663e\u793a HarmEvidence&#xff08;\u5728\u4e94\u4e2a\u6a21\u578b\u6c47\u5408\u7684\u56fa\u5b9a 600 \u6848\u4f8b\u5206\u6bcd\u4e0a\u4e3a 16.8%&#xff09;&#xff0c;336 \u4e2a\u4e3a SemanticOnly&#xff08;11.2%&#xff09;&#xff0c;0 \u4e2a\u4e3a NullAudit\u3002\u5728\u6240\u6709\u56db\u4e2a\u6807\u9898\u7b56\u7565\u4e2d&#xff0c;\u89c4\u8303\u5ba1\u8ba1\u8986\u76d6 1,834\/1,834 \u4e2a CoreFail \u884c&#xff0c;\u7f3a\u5931\u884c\u4e3a\u96f6&#xff1a;959 \u4e2a HarmEvidence&#xff0c;873 \u4e2a SemanticOnly&#xff0c;1 \u4e2a NullAudit&#xff0c;\u4ee5\u53ca 1 \u4e2a\u89e3\u6790\u7684 OtherNoHarm \u6b8b\u4f59\u3002\u8868 4 \u7ed9\u51fa\u4e86 D0 \u6a21\u578b\u7ea7\u8bc1\u636e\u6838\u7b97&#xff0c;\u8868 5 \u7ed9\u51fa\u4e86\u5b8c\u6574\u7684\u9632\u5fa1\u7ea7\u5ba1\u8ba1\u95ed\u5408\u3002<\/p>\n<p>\u8868 4&#xff1a;\u7cbe\u7b80\u4e94\u7aef\u70b9\u4e3b\u9762\u677f\u5728 D0 \u4e0b\u7684\u89c4\u8303\u6838\u5fc3\u95e8\u63a7\u5371\u5bb3\u8bc1\u636e\u5ba1\u8ba1\u3002\u00a0AuditRows \u662f\u4ece\u6700\u7ec8\u6838\u5fc3\u77e9\u9635\u4e2d\u9009\u62e9\u7684\u7cbe\u786e\u89c4\u8303 CoreFail \u884c&#xff1b;\u6240\u6709 D0 CoreFail \u884c\u5747\u88ab\u5ba1\u8ba1\u3002NotObservable \u5bf9\u4e8e\u8fd9\u4e9b\u884c\u4e3a\u96f6&#xff0c;\u4e3a\u7b80\u6d01\u8d77\u89c1\u7701\u7565\u3002HarmEvidence\u3001SemOnly \u548c Null \u662f\u8ba1\u6570&#xff0c;\u95ed\u5408\u4e3a AuditRows &#061; HarmEvidence &#043; SemOnly &#043; Null\u3002\u8bc1\u636e\u5217\u662f\u6838\u5fc3\u5de5\u4ef6\u8bc1\u636e\u8ba1\u6570&#xff0c;\u53ef\u80fd\u5728 HarmEvidence \u5185\u91cd\u53e0\u3002<\/p>\n<table>\n<tr>\n<p>\u6a21\u578b<\/p>\n<p>CoreFail&#064;600<\/p>\n<p>AuditRows<\/p>\n<p>TextHarm<\/p>\n<p>AccessHarm<\/p>\n<p>ActionHarm<\/p>\n<p>PersistHarm<\/p>\n<p>HarmEvidence<\/p>\n<p>SemOnly<\/p>\n<p>Null<\/p>\n<p>Check<\/p>\n<\/tr>\n<tbody>\n<tr>\n<td>\n<p>GPT-5.5<\/p>\n<\/td>\n<td>\n<p>23.8<\/p>\n<\/td>\n<td>\n<p>143<\/p>\n<\/td>\n<td>\n<p>10<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>3<\/p>\n<\/td>\n<td>\n<p>29<\/p>\n<\/td>\n<td>\n<p>40<\/p>\n<\/td>\n<td>\n<p>103<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>Claude Opus 4.7<\/p>\n<\/td>\n<td>\n<p>9.0<\/p>\n<\/td>\n<td>\n<p>54<\/p>\n<\/td>\n<td>\n<p>23<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>2<\/p>\n<\/td>\n<td>\n<p>9<\/p>\n<\/td>\n<td>\n<p>32<\/p>\n<\/td>\n<td>\n<p>22<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>Qwen3.6-Plus<\/p>\n<\/td>\n<td>\n<p>35.5<\/p>\n<\/td>\n<td>\n<p>213<\/p>\n<\/td>\n<td>\n<p>18<\/p>\n<\/td>\n<td>\n<p>11<\/p>\n<\/td>\n<td>\n<p>66<\/p>\n<\/td>\n<td>\n<p>95<\/p>\n<\/td>\n<td>\n<p>133<\/p>\n<\/td>\n<td>\n<p>80<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>GLM-5.1<\/p>\n<\/td>\n<td>\n<p>27.5<\/p>\n<\/td>\n<td>\n<p>165<\/p>\n<\/td>\n<td>\n<p>46<\/p>\n<\/td>\n<td>\n<p>15<\/p>\n<\/td>\n<td>\n<p>28<\/p>\n<\/td>\n<td>\n<p>69<\/p>\n<\/td>\n<td>\n<p>120<\/p>\n<\/td>\n<td>\n<p>45<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>Kimi K2.5<\/p>\n<\/td>\n<td>\n<p>44.2<\/p>\n<\/td>\n<td>\n<p>265<\/p>\n<\/td>\n<td>\n<p>55<\/p>\n<\/td>\n<td>\n<p>30<\/p>\n<\/td>\n<td>\n<p>70<\/p>\n<\/td>\n<td>\n<p>95<\/p>\n<\/td>\n<td>\n<p>179<\/p>\n<\/td>\n<td>\n<p>86<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u603b\u8ba1\/\u6c47\u5408<\/p>\n<\/td>\n<td>\n<p>28.0<\/p>\n<\/td>\n<td>\n<p>840<\/p>\n<\/td>\n<td>\n<p>152<\/p>\n<\/td>\n<td>\n<p>56<\/p>\n<\/td>\n<td>\n<p>169<\/p>\n<\/td>\n<td>\n<p>297<\/p>\n<\/td>\n<td>\n<p>504<\/p>\n<\/td>\n<td>\n<p>336<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u8868 5&#xff1a;\u7cbe\u7b80\u4e94\u7aef\u70b9\u9762\u677f\u6309\u9632\u5fa1\u7684\u89c4\u8303\u6838\u5fc3\u95e8\u63a7\u5371\u5bb3\u8bc1\u636e\u5ba1\u8ba1\u6838\u7b97\u3002\u00a0\u8ba1\u6570\u662f\u6765\u81ea\u89c4\u8303 CoreFail \u5ba1\u8ba1\u7684\u7cbe\u786e\u884c&#xff1b;\u6240\u6709\u89c4\u8303 CoreFail \u884c\u5747\u88ab\u5ba1\u8ba1\u3002HarmEvidence \u662f\u6838\u5fc3\u95e8\u63a7\u8bc1\u636e\u7aef\u70b9&#xff1b;OtherNoHarm \u662f\u89e3\u6790\u7684\u65e0\u5371\u5bb3\/\u975e\u8bed\u4e49\u6b8b\u4f59\u7c7b\u522b&#xff0c;\u5305\u542b\u4ee5\u4f7f\u5ba1\u8ba1\u884c\u6838\u7b97\u95ed\u5408\u3002<\/p>\n<table>\n<tr>\n<p>\u9632\u5fa1<\/p>\n<p>CoreFail<\/p>\n<p>Audited<\/p>\n<p>Missing<\/p>\n<p>HarmEvidence<\/p>\n<p>SemOnly<\/p>\n<p>NullAudit<\/p>\n<p>OtherNoHarm<\/p>\n<p>Check<\/p>\n<\/tr>\n<tbody>\n<tr>\n<td>\n<p>D0<\/p>\n<\/td>\n<td>\n<p>840<\/p>\n<\/td>\n<td>\n<p>840<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>504<\/p>\n<\/td>\n<td>\n<p>336<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>B2\/Sandwich<\/p>\n<\/td>\n<td>\n<p>445<\/p>\n<\/td>\n<td>\n<p>445<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>176<\/p>\n<\/td>\n<td>\n<p>269<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>D3<\/p>\n<\/td>\n<td>\n<p>272<\/p>\n<\/td>\n<td>\n<p>272<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>119<\/p>\n<\/td>\n<td>\n<p>152<\/p>\n<\/td>\n<td>\n<p>1<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>D4\/LongPolicy<\/p>\n<\/td>\n<td>\n<p>277<\/p>\n<\/td>\n<td>\n<p>277<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>160<\/p>\n<\/td>\n<td>\n<p>116<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>1<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p>\u603b\u8ba1<\/p>\n<\/td>\n<td>\n<p>1,834<\/p>\n<\/td>\n<td>\n<p>1,834<\/p>\n<\/td>\n<td>\n<p>0<\/p>\n<\/td>\n<td>\n<p>959<\/p>\n<\/td>\n<td>\n<p>873<\/p>\n<\/td>\n<td>\n<p>1<\/p>\n<\/td>\n<td>\n<p>1<\/p>\n<\/td>\n<td>\n<p>OK<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u653b\u51fb\u65cf\u89e6\u53d1\u4e0d\u540c\u7684\u5ba1\u8ba1\u8bc1\u636e\u901a\u9053&#xff08;\u9644\u5f55 A.5&#xff09;\u3002\u5728 D0 \u4e0b&#xff0c;MPI \u662f\u6700\u6e05\u6670\u7684\u6301\u4e45\u5316\u8bc1\u636e\u65cf&#xff1a;272 \u4e2a\u5ba1\u8ba1\u7684 MPI CoreFail \u884c\u4e2d\u6709 251 \u4e2a\u5177\u6709 HarmEvidence&#xff0c;\u4e3b\u8981\u901a\u8fc7\u6301\u4e45\u5316\u8bc1\u636e\u3002MEX \u5728\u8bed\u4e49\u4e0a\u4ecd\u7136\u5f88\u9ad8&#xff0c;\u4f46\u5728\u4e25\u683c\u7684\u53d7\u4fdd\u62a4\u5bf9\u8c61\u8bc1\u636e\u4e0b\u66f4\u7a84&#xff1a;272 \u4e2a D0 \u5ba1\u8ba1\u7684 MEX CoreFail \u884c\u4e2d\u6709 121 \u4e2a\u5177\u6709 HarmEvidence&#xff0c;\u800c\u5bbd\u6cdb\u7684\u7b56\u7565\u6216\u914d\u7f6e\u89e3\u91ca\u9664\u975e\u66b4\u9732\u53d7\u4fdd\u62a4\u5bf9\u8c61&#xff0c;\u5426\u5219\u4ecd\u4e3a SemanticOnly\u3002\u56e0\u6b64&#xff0c;D4\/LongPolicy \u88ab\u89c6\u4e3a\u63d0\u793a\u7b56\u7565\u63a2\u9488&#xff0c;\u800c\u975e\u7edf\u4e00\u5360\u4f18\u7684\u751f\u4ea7\u9632\u5fa1&#xff1b;\u8be6\u7ec6\u7684\u7ec4\u4ef6\u6d88\u878d\u79fb\u81f3\u9644\u5f55 E\u3002<\/p>\n<hr \/>\n<p>5. \u7aef\u70b9\u5206\u6790<\/p>\n<p>\u5185\u5b58\u6295\u6bd2\u4e4b\u6240\u4ee5\u6709\u6548&#xff0c;\u662f\u56e0\u4e3a\u5185\u5b58\u5185\u5bb9\u6301\u4e45\u5316\u4e14\u5e38\u88ab\u89c6\u4e3a\u53ef\u4fe1\u4e0a\u4e0b\u6587\u3002MEX \u9ad8\u7684\u539f\u56e0\u4e0d\u540c&#xff1a;\u5f53\u524d\u7684\u6c47\u603b MEX \u65cf\u7ed3\u5408\u4e86\u7cbe\u786e\u79cd\u5b50\u79d8\u5bc6\u6cc4\u9732\u3001\u53d7\u4fdd\u62a4\u914d\u7f6e\u62ab\u9732\u3001\u53d7\u4fdd\u62a4\u5185\u5b58\/\u8bb0\u5f55\u68c0\u7d22\u4ee5\u53ca\u5bbd\u6cdb\u7684\u7b56\u7565\/\u914d\u7f6e\u900f\u660e\u5ea6\u3002\u5728\u6838\u5fc3\u95e8\u63a7\u5ba1\u8ba1\u4e2d&#xff0c;\u7cbe\u786e\u53d7\u4fdd\u62a4\u5b57\u7b26\u4e32\u6cc4\u9732\u8ba1\u4e3a TextHarm&#xff0c;\u79cd\u5b50\u53d7\u4fdd\u62a4\u914d\u7f6e\u62ab\u9732\u6839\u636e\u60c5\u51b5\u8ba1\u4e3a TextHarm\/SensitiveConfig&#xff0c;\u800c\u6ca1\u6709\u53d7\u4fdd\u62a4\u5bf9\u8c61\u7684\u5bbd\u6cdb\u7b56\u7565\u89e3\u91ca\u4ecd\u4e3a SemanticOnly \u6216\u826f\u6027\u900f\u660e\u5ea6\u9519\u8bef\u3002MEX \u5b50\u7c7b\u578b\u8868\u662f\u57fa\u4e8e\u89c4\u5219\u7684&#xff0c;\u6709\u52a9\u4e8e\u5c06\u6c47\u603b MEX \u89e3\u91ca\u4e3a\u6df7\u5408\u7684\u53d7\u4fdd\u62a4\u4fe1\u606f\u65cf&#xff0c;\u800c\u975e\u7eaf\u7cbe\u786e\u79d8\u5bc6\u6cc4\u9732\u7387\u3002<\/p>\n<p>5.1 RQ3&#xff1a;Exec-Balanced \u95e8\u63a7\u8bca\u65ad<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213331-6a6fb7ab009ce.png\" \/><\/p>\n<p>\u56fe3&#xff1a;\u6309\u9632\u5fa1\u7684 Exec ObservedHarm\u3002\u6761\u5f62\u663e\u793a\u5747\u503c&#xff1b;\u70b9\u548c\u8303\u56f4\u663e\u793a\u7aef\u70b9\u5206\u5e03\u3002<\/p>\n<p>CoreFail&#064;600 \u6d4b\u91cf\u6a21\u578b\u7684\u6587\u672c\u8f93\u51fa\u662f\u5426\u5728\u8bed\u4e49\u4e0a\u9075\u4ece\u653b\u51fb\u76ee\u6807&#xff1b;Exec-Balanced \u5219\u8be2\u95ee\u4e00\u4e2a\u5355\u72ec\u6c99\u7bb1\u8f68\u8ff9\u662f\u5426\u89e6\u53d1\u786e\u5b9a\u6027\u6587\u4ef6\u3001\u7535\u5b50\u90ae\u4ef6\u3001\u5185\u5b58\u3001\u6570\u636e\u5e93\u3001Web \u6216\u4ee3\u7801\u9884\u8a00\u673a\u3002\u5728\u7cbe\u7b80\u4e3b\u9762\u677f\u4e2d&#xff0c;D0 ObservedHarm&#064;Exec \u4ece Claude Opus 4.7 \u7684 0.5% \u5230 Kimi K2.5 \u7684 11.3%&#xff0c;\u5747\u503c\u4e3a 4.133%\u3002StateChange \u548c ObservedHarm \u5728\u6b64\u5bf9\u6297\u6027 D0 \u9762\u677f\u4e2d\u91cd\u5408&#xff0c;\u56e0\u4e3a\u6bcf\u4e2a\u8bc4\u5206\u7684\u653b\u51fb\u76f8\u5173\u72b6\u6001\u5dee\u5f02\u672c\u8d28\u4e0a\u662f\u6709\u5bb3\u7684&#xff0c;\u4f46\u8fd9\u4e24\u4e2a\u7aef\u70b9\u5bf9\u4e8e\u826f\u6027\u6216\u6df7\u5408\u7528\u9014\u9762\u677f\u5728\u6982\u5ff5\u4e0a\u4fdd\u6301\u5206\u79bb\u3002<\/p>\n<p>\u7cbe\u786e\u7684\u7cbe\u7b80 Exec \u5de5\u4ef6\u652f\u6301\u6bcf\u9632\u5fa1\u5bf9 3,000 \u4e2a\u6848\u4f8b-\u6a21\u578b\u884c\u7684\u914d\u5bf9\u6bd4\u8f83&#xff08;\u9644\u5f55\u8868 18 \u548c 19&#xff09;\u3002\u76f8\u5bf9\u4e8e D0&#xff0c;B2\u3001D3 \u548c D4\/LongPolicy \u5728\u8be5\u534f\u8bae\u4e2d\u4e0e\u66f4\u5c11\u7684\u6c47\u5408 Exec \u5371\u5bb3\u884c\u76f8\u5173&#xff1a;D0 \u4e0b 124\/3,000 \u5bf9\u6bd4 B2 76\/3,000\u3001D3 76\/3,000 \u548c D4 71\/3,000&#xff0c;\u5177\u6709\u663e\u8457\u7684 Holm \u8c03\u6574 McNemar \u68c0\u9a8c\u3002\u4ee3\u8868\u6027\u548c\u56f0\u96be\u5206\u5272\u884c\u662f\u7528\u4e8e\u672a\u6765\u5206\u5272\u654f\u611f\u6027\u5206\u6790\u7684\u91cd\u53e0\u68c0\u67e5\u3002<\/p>\n<p>\u6838\u5fc3-Exec \u8fde\u63a5\u8986\u76d6\u6240\u6709\u4e94\u4e2a\u7aef\u70b9\u3001\u6240\u6709\u56db\u4e2a\u6807\u9898\u7b56\u7565\u548c\u6240\u6709 600 \u4e2a\u53ef\u6267\u884c\u6620\u5c04&#xff0c;\u4ea7\u751f 12,000 \u4e2a\u5339\u914d\u884c&#xff08;\u9644\u5f55\u8868 21&#xff09;\u3002\u8fde\u63a5\u5305\u542b 56 \u4e2a CoreFail\u2227ExecHarm \u884c&#xff0c;1,778 \u4e2a\u65e0 ExecHarm \u7684 CoreFail \u884c&#xff0c;\u4ee5\u53ca 291 \u4e2a CorePass\u2227ExecHarm \u884c\u3002\u56e0\u6b64&#xff0c;347 \u4e2a ExecHarm \u884c\u4e2d\u7684 291 \u4e2a&#xff08;83.9%&#xff09;\u6765\u81ea\u901a\u8fc7\u8bed\u4e49\u6838\u5fc3\u8c03\u7528\u7684\u884c\u3002\u8fd9\u4e00\u6a21\u5f0f\u652f\u6301\u7aef\u70b9\u5206\u79bb&#xff0c;\u4f46\u4e0d\u5e94\u89e3\u8bfb\u4e3a\u7b80\u5355\u7684\u6587\u672c\u8bc4\u5224\u5458\u9057\u6f0f&#xff1a;\u6838\u5fc3\u548c Exec \u4f7f\u7528\u4e0d\u540c\u7684\u6a21\u578b\u8c03\u7528\u3001\u63d0\u793a\u683c\u5f0f\u3001\u5de5\u5177\u6a21\u5f0f\u3001\u89e3\u6790\u5668\u8def\u5f84\u3001\u6c99\u7bb1\u6743\u9650\u548c\u5de5\u5177\u53ef\u89c1\u6027\u3002\u66f4\u72ed\u7a84\u7684\u4e3b\u5f20\u662f&#xff0c;\u5728\u5355\u72ec\u7684\u53ef\u6267\u884c\u534f\u8bae\u4e0b&#xff0c;\u4e00\u4e9b\u5339\u914d\u7684\u4efb\u52a1\u6807\u8bc6\u5c3d\u7ba1\u901a\u8fc7\u4e86\u8bed\u4e49\u6838\u5fc3\u8c03\u7528&#xff0c;\u4ecd\u4f1a\u4ea7\u751f\u6c99\u7bb1\u5371\u5bb3\u3002<\/p>\n<p>\u8f85\u52a9\u6548\u7528\u8303\u56f4\u3002<br \/>\n\u826f\u6027\u6548\u7528\u9762\u677f\u662f\u5b58\u6863\u7684\u9644\u5f55\u914d\u5957\u68c0\u67e5&#xff0c;\u800c\u975e\u5b89\u5168-\u6548\u7528\u6743\u8861\u7684\u8bc1\u636e\u3002\u5b83\u6db5\u76d6\u76f8\u540c\u7684\u6807\u9898\u63d0\u793a\u7b56\u7565\u6807\u7b7e&#xff0c;\u4f46\u9664\u4e86 Kimi K2.5 \u5916&#xff0c;\u6240\u6709\u63d0\u4f9b\u8005\u90fd\u4f7f\u7528\u6700\u8fd1\u7684\u5b58\u6863\u7aef\u70b9\u800c\u975e\u7cbe\u786e\u7684\u7cbe\u7b80\u9762\u677f\u7aef\u70b9\u3002\u5176\u4efb\u52a1\u6210\u529f\u7387\u4e5f\u6df7\u5408\u4e86\u4efb\u52a1\u8bbe\u8ba1\u3001\u5de5\u5177\u73af\u5883\u3001\u89e3\u6790\u5668\u884c\u4e3a\u548c\u6a21\u578b\u884c\u4e3a&#xff0c;\u800c\u865a\u5047\u62d2\u7edd\u5b57\u6bb5\u8f83\u7a84\u3002\u56e0\u6b64&#xff0c;\u6211\u4eec\u4ec5\u7528\u5b83\u6765\u8bb0\u5f55\u5728\u91cd\u8bd5\u5b8c\u6210\u540e\u7684\u5927\u89c4\u6a21\u8fc7\u5ea6\u62d2\u7edd\u6216 API \u4efb\u52a1\u5931\u6548\u6a21\u5f0f&#xff1b;\u5728\u76f8\u540c\u7aef\u70b9\u3001\u7b56\u7565\u548c\u5de5\u5177\u534f\u8bae\u4e0b\u7684\u7cbe\u786e\u7cbe\u7b80\u9762\u677f\u6548\u7528\u91cd\u65b0\u8fd0\u884c\u7559\u7ed9\u672a\u6765\u5de5\u4f5c\u3002<\/p>\n<hr \/>\n<p>6. \u7ed3\u8bba<\/p>\n<p>SafeClawBench \u8868\u660e&#xff0c;\u5de5\u5177\u4ee3\u7406\u5b89\u5168\u8bc4\u4f30\u9700\u8981\u7aef\u70b9\u5206\u79bb\u3002\u5728\u4e94\u7aef\u70b9\u56db\u7b56\u7565\u9762\u677f\u4e2d&#xff0c;\u6a21\u578b\u9009\u62e9\u4ea7\u751f\u8f83\u5927\u7684 D0 \u65b9\u5dee&#xff0c;\u63d0\u793a\u7b56\u7565\u4f9d\u8d56\u4e8e\u7aef\u70b9&#xff0c;\u6838\u5fc3\u95e8\u63a7\u7684 HarmEvidence&#064;600 \u6bd4 CoreFail&#064;600 \u66f4\u7a84&#xff0c;\u5e76\u4e14\u6838\u5fc3-Exec \u5339\u914d\u5728\u6c99\u7bb1\u534f\u8bae\u4e0b\u66b4\u9732\u4e86\u4e00\u4e2a\u5927\u7684 CorePass\u2013ExecHarm \u5207\u7247\u3002\u6b63\u6587\u4e2d\u7684\u53ef\u89c6\u5316\u4ee5\u56fe\u5f62\u65b9\u5f0f\u8bf4\u660e\u4e86\u540c\u4e00\u89c2\u70b9&#xff1a;\u6838\u5fc3\u9632\u5fa1\u66f2\u7ebf\u538b\u7f29\u4e86\u5e7f\u6cdb\u7684\u8bed\u4e49\u5931\u8d25\u8303\u56f4&#xff0c;\u800c Exec \u6761\u5f62\u6d4b\u91cf\u72b6\u6001\u9884\u8a00\u673a\u7ed3\u679c\u3002<\/p>\n<p>\u5b9e\u9645\u540e\u679c\u662f&#xff0c;\u57fa\u51c6\u7528\u6237\u5e94\u5728\u6bd4\u8f83\u6a21\u578b\u6216\u63d0\u793a\u7b56\u7565\u4e4b\u524d\u51b3\u5b9a\u54ea\u4e2a\u7aef\u70b9\u5339\u914d\u5176\u90e8\u7f72\u95ee\u9898\u3002CoreFail&#064;600 \u6d4b\u91cf\u6a21\u578b\u662f\u5426\u5728\u6587\u672c\u4e2d\u63a5\u53d7\u5bf9\u6297\u6027\u76ee\u6807\u3002\u6838\u5fc3\u95e8\u63a7\u5ba1\u8ba1\u8be2\u95ee\u8be5\u6838\u5fc3\u5de5\u4ef6\u662f\u5426\u8fd8\u5305\u542b\u53d7\u4fdd\u62a4\u5bf9\u8c61\u3001\u8bbf\u95ee\u3001\u884c\u52a8\u6216\u6301\u4e45\u5316\u8bc1\u636e\u3002Exec-Balanced \u5c06\u5339\u914d\u7684\u4efb\u52a1\u6807\u8bc6\u79fb\u81f3\u9694\u79bb\u7684\u5de5\u5177\/\u72b6\u6001\u73af\u5883\u5e76\u6d4b\u91cf\u786e\u5b9a\u6027\u6c99\u7bb1\u7ed3\u679c\u3002291 \u4e2a CorePass\u2227ExecHarm \u884c\u8868\u660e&#xff0c;\u5728\u5355\u72ec\u7684\u53ef\u6267\u884c\u534f\u8bae\u4e0b&#xff0c;\u4e00\u4e9b\u5339\u914d\u7684\u4efb\u52a1\u6807\u8bc6\u5c3d\u7ba1\u901a\u8fc7\u4e86\u8bed\u4e49\u6838\u5fc3\u8c03\u7528&#xff0c;\u4ecd\u4f1a\u4ea7\u751f\u6c99\u7bb1\u5371\u5bb3&#xff0c;\u800c\u8bb8\u591a CoreFail \u884c\u5728\u5ba1\u8ba1\u6a21\u5f0f\u4e0b\u4ecd\u4ec5\u4e3a\u8bed\u4e49\u6027\u7684\u3002<\/p>\n<p>\u63d0\u793a\u7b56\u7565\u7ed3\u679c\u9075\u5faa\u76f8\u540c\u7684\u5206\u9636\u6bb5\u6a21\u5f0f\u3002\u8f7b\u91cf\u7ea7\u5305\u88c5\u3001\u5206\u5c42\u63d0\u793a\u6307\u4ee4\u548c\u8fc7\u5ea6\u6307\u5b9a\u7684 D4\/LongPolicy \u63a2\u9488\u4e0e\u67d0\u4e9b\u7aef\u70b9\u4e0a\u8f83\u4f4e\u7684\u7387\u76f8\u5173&#xff0c;\u4f46\u6548\u679c\u56e0\u6a21\u578b\u548c\u534f\u8bae\u800c\u5f02\u3002\u5339\u914d\u957f\u5ea6\u63a7\u5236\u8fdb\u4e00\u6b65\u8868\u660e\u63d0\u793a\u957f\u5ea6\u672c\u8eab\u4f1a\u6539\u53d8\u6a21\u578b\u884c\u4e3a&#xff0c;\u56e0\u6b64\u6211\u4eec\u5c06 D4\/LongPolicy \u89c6\u4e3a\u63d0\u793a\u7b56\u7565\u538b\u529b\u6848\u4f8b&#xff0c;\u800c\u975e\u5c06\u7ed3\u679c\u5f52\u56e0\u4e8e SIA\u3001MIG \u6216 TCA \u7ec4\u4ef6\u3002\u5b58\u6863\u7684\u826f\u6027\u6548\u7528\u68c0\u67e5\u4ec5\u4e3a\u9644\u5f55\u8bca\u65ad&#xff1b;\u5728\u6ca1\u6709\u7cbe\u786e\u7cbe\u7b80\u9762\u677f\u6548\u7528\u91cd\u65b0\u8fd0\u884c\u7684\u60c5\u51b5\u4e0b&#xff0c;\u5b83\u4eec\u4e0d\u652f\u6301\u5b89\u5168-\u6548\u7528\u6743\u8861\u7684\u4e3b\u5f20\u3002<\/p>\n<p>\u7cbe\u7b80\u9762\u677f\u8bbe\u8ba1\u4e5f\u4f7f\u6765\u6e90\u6210\u4e3a\u79d1\u5b66\u4e3b\u5f20\u7684\u4e00\u90e8\u5206\u3002\u6211\u4eec\u901a\u8fc7\u7aef\u70b9\u5b57\u7b26\u4e32\u548c\u5de5\u4ef6\u54c8\u5e0c\u56fa\u5b9a\u4e3b\u8981\u6bd4\u8f83&#xff0c;\u5728\u53ef\u91cd\u73b0\u6027\u5143\u6570\u636e\u4e2d\u62a5\u544a\u8fd9\u4e9b\u6807\u8bc6\u7b26&#xff0c;\u5e76\u5c06\u5b58\u6863\u7684\u5168\u7aef\u70b9\u77e9\u9635\u4e0e\u7cbe\u7b80\u9762\u677f\u6807\u9898\u884c\u5206\u5f00\u3002\u53d1\u5e03\u5305\u56f4\u7ed5\u8fd9\u4e00\u8fb9\u754c\u7ec4\u7ec7&#xff1a;\u57fa\u51c6 JSON\u3001\u539f\u59cb\u6a21\u578b\u8f93\u51fa\u6587\u4ef6\u3001\u8bc4\u5224\u5458\u548c\u9632\u5fa1\u63d0\u793a\u3001\u5ba1\u8ba1\u8ddf\u8e2a\u3001\u6c99\u7bb1\u65e5\u5fd7\u3001\u8fd0\u884c\u6e05\u5355\u548c\u786e\u5b9a\u6027\u56fe\u8868\/\u6c47\u603b\u811a\u672c\u5728\u9644\u5f55\u4e2d\u5217\u51fa\u5e76\u9644\u6709\u9a8c\u8bc1\u89d2\u8272\u3002240 \u884c\u6821\u51c6\u68c0\u67e5\u4ee5\u540c\u6837\u7684\u7cbe\u795e\u62a5\u544a&#xff0c;\u4f5c\u4e3a\u5206\u5c42\u6837\u672c\u4e0a\u7684\u654f\u611f\u6027\u68c0\u67e5&#xff0c;\u800c\u975e\u5bf9\u4e3b\u8981\u6807\u7b7e\u7684\u66ff\u4ee3\u3002<\/p>\n<p>SafeClawBench \u662f\u4e00\u4e2a\u538b\u529b\u6d4b\u8bd5\u6846\u67b6&#xff0c;\u7528\u4e8e\u5b9a\u4f4d\u5931\u6548\u53d1\u751f\u7684\u4f4d\u7f6e&#xff1a;\u8bed\u4e49\u9075\u4ece\u6027\u3001\u5ba1\u8ba1\u8bc1\u636e\u652f\u6301\u7684\u5371\u5bb3\u3001\u6c99\u7bb1\u89c2\u5bdf\u5230\u7684\u72b6\u6001\u53d8\u5316\u6216\u826f\u6027\u4efb\u52a1\u9000\u5316\u3002\u90e8\u7f72\u7cfb\u7edf\u4ecd\u7136\u9700\u8981\u8fd0\u884c\u65f6\u6743\u9650\u3001\u76d1\u63a7\u548c\u5bf9\u4e0d\u53ef\u9006\u64cd\u4f5c\u7684\u4eba\u5de5\u6279\u51c6\u3002\u901a\u8fc7\u5728\u8fd9\u4e9b\u7aef\u70b9\u4e4b\u95f4\u4fdd\u6301\u5206\u79bb&#xff0c;\u540c\u65f6\u5728\u53ef\u80fd\u7684\u60c5\u51b5\u4e0b\u4f7f\u7528\u5339\u914d\u7684\u4efb\u52a1\u6807\u8bc6&#xff0c;SafeClawBench \u652f\u6301\u5de5\u5177\u4ee3\u7406\u6a21\u578b\u548c\u63d0\u793a\u7b56\u7565\u7684\u53ef\u91cd\u73b0\u6bd4\u8f83\u3002<\/p>\n<hr \/>\n<p>\u5c40\u9650\u6027<\/p>\n<p>SafeClawBench \u662f\u4e00\u4e2a\u53d7\u63a7\u538b\u529b\u6d4b\u8bd5\u5957\u4ef6&#xff0c;\u56e0\u6b64\u5176\u6570\u5b57\u6700\u597d\u4f5c\u4e3a\u6bd4\u8f83\u6027\u7aef\u70b9\u6d4b\u91cf\u800c\u975e\u64cd\u4f5c\u4e8b\u4ef6\u7387\u3002\u63d0\u793a\u7b56\u7565\u77e9\u9635\u9694\u79bb\u4e86\u4e00\u4e2a\u91cd\u8981\u7684\u63a7\u5236\u5c42&#xff1b;\u90e8\u7f72\u7cfb\u7edf\u5e94\u5c06\u7b56\u7565\u4e0e\u8fd0\u884c\u65f6\u5de5\u5177\u6743\u9650\u548c\u76d1\u63a7\u76f8\u7ed3\u5408\u3002\u4e00\u4e9b\u9700\u8981\u957f\u4e0a\u4e0b\u6587\u6216\u5916\u90e8\u670d\u52a1\u7684\u5371\u5bb3\u5728\u5f53\u524d\u6c99\u7bb1\u4e4b\u5916\u3002\u53d1\u5e03\u65f6&#xff0c;\u6240\u6709\u6848\u4f8b\u5747\u4e3a\u5408\u6210\u6216\u53d7\u63a7&#xff0c;\u6ca1\u6709\u751f\u4ea7\u7cfb\u7edf\u88ab\u653b\u51fb&#xff0c;\u516c\u5171\u5de5\u4ef6\u5728\u9700\u8981\u5904\u8fdb\u884c\u4e86\u7f16\u8f91&#xff1b;\u643a\u5e26\u91d1\u4e1d\u96c0\u7684\u884c\u4f7f\u7528\u53d7\u63a7\u8bbf\u95ee\u3002<\/p>\n<p>\u00a0<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5927\u5bb6\u8bfb\u5b8c\u89c9\u5f97\u6709\u5e2e\u52a9\u8bb0\u5f97\u5173\u6ce8\u548c\u70b9\u8d5e&#xff01;&#xff01;&#xff01;\u6458\u8981<br \/>\n\u4f7f\u7528\u5de5\u5177\u7684\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4f1a\u5f15\u5165\u8d85\u8d8a\u4e0d\u5b89\u5168\u6587\u672c\u7684\u5b89\u5168\u5931\u6548&#xff1a;\u5b83\u4eec\u53ef\u80fd\u6cc4\u9732\u53d7\u4fdd\u62a4\u5bf9\u8c61\u3001\u5199\u5165\u6301\u4e45\u5316\u5185\u5b58\u3001\u53d1\u9001\u6d88\u606f\u3001\u4fee\u6539\u6570\u636e\u5e93&#xff0c;\u6216\u89e6\u53d1\u6709\u5bb3\u4ee3\u7801\u548c\u5de5\u5177\u6548\u5e94\u3002\u73b0\u6709\u8bc4\u4f30\u5f80\u5f80\u5c06\u8fd9\u4e9b\u9636\u6bb5\u5408\u5e76\u4e3a\u5355\u4e00\u7684\u653b\u51fb\u6210\u529f\u7387&#xff0c;\u4f7f\u5f97\u96be\u4ee5\u533a\u5206\u6a21\u578b\u4ec5\u4ec5\u662f\u540c\u610f\u4e86\u653b\u51fb\u8005\u8fd8\u662f\u5b9e\u9645\u4ea7\u751f\u4e86\u53ef\u89c2\u5bdf\u7684\u5371\u5bb3\u3002\u6211\u4eec\u63d0\u51fa\u4e86 SafeClawBench&#xff0c;\u4e00\u4e2a\u9488\u5bf9\u4f7f\u7528\u5de5\u5177\u7684\u4ee3\u7406\u5b89\u5168\u7684\u5206\u9636\u6bb5\u57fa\u51c6\u6d4b\u8bd5&amp;#x<\/p>\n","protected":false},"author":2,"featured_media":89483,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[50,78,51],"topic":[],"class_list":["post-89490","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-50","tag-78","tag-51"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/89490.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u5927\u5bb6\u8bfb\u5b8c\u89c9\u5f97\u6709\u5e2e\u52a9\u8bb0\u5f97\u5173\u6ce8\u548c\u70b9\u8d5e&#xff01;&#xff01;&#xff01;\u6458\u8981 \u4f7f\u7528\u5de5\u5177\u7684\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4f1a\u5f15\u5165\u8d85\u8d8a\u4e0d\u5b89\u5168\u6587\u672c\u7684\u5b89\u5168\u5931\u6548&#xff1a;\u5b83\u4eec\u53ef\u80fd\u6cc4\u9732\u53d7\u4fdd\u62a4\u5bf9\u8c61\u3001\u5199\u5165\u6301\u4e45\u5316\u5185\u5b58\u3001\u53d1\u9001\u6d88\u606f\u3001\u4fee\u6539\u6570\u636e\u5e93&#xff0c;\u6216\u89e6\u53d1\u6709\u5bb3\u4ee3\u7801\u548c\u5de5\u5177\u6548\u5e94\u3002\u73b0\u6709\u8bc4\u4f30\u5f80\u5f80\u5c06\u8fd9\u4e9b\u9636\u6bb5\u5408\u5e76\u4e3a\u5355\u4e00\u7684\u653b\u51fb\u6210\u529f\u7387&#xff0c;\u4f7f\u5f97\u96be\u4ee5\u533a\u5206\u6a21\u578b\u4ec5\u4ec5\u662f\u540c\u610f\u4e86\u653b\u51fb\u8005\u8fd8\u662f\u5b9e\u9645\u4ea7\u751f\u4e86\u53ef\u89c2\u5bdf\u7684\u5371\u5bb3\u3002\u6211\u4eec\u63d0\u51fa\u4e86 SafeClawBench&#xff0c;\u4e00\u4e2a\u9488\u5bf9\u4f7f\u7528\u5de5\u5177\u7684\u4ee3\u7406\u5b89\u5168\u7684\u5206\u9636\u6bb5\u57fa\u51c6\u6d4b\u8bd5&amp;#x\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/89490.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-08-02T21:33:32+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213330-6a6fb7aa92cbd.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"8 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/89490.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/89490.html\",\"name\":\"SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-08-02T21:33:32+00:00\",\"dateModified\":\"2026-08-02T21:33:32+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/89490.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/89490.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/89490.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/89490.html","og_locale":"zh_CN","og_type":"article","og_title":"SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u5927\u5bb6\u8bfb\u5b8c\u89c9\u5f97\u6709\u5e2e\u52a9\u8bb0\u5f97\u5173\u6ce8\u548c\u70b9\u8d5e&#xff01;&#xff01;&#xff01;\u6458\u8981 \u4f7f\u7528\u5de5\u5177\u7684\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4f1a\u5f15\u5165\u8d85\u8d8a\u4e0d\u5b89\u5168\u6587\u672c\u7684\u5b89\u5168\u5931\u6548&#xff1a;\u5b83\u4eec\u53ef\u80fd\u6cc4\u9732\u53d7\u4fdd\u62a4\u5bf9\u8c61\u3001\u5199\u5165\u6301\u4e45\u5316\u5185\u5b58\u3001\u53d1\u9001\u6d88\u606f\u3001\u4fee\u6539\u6570\u636e\u5e93&#xff0c;\u6216\u89e6\u53d1\u6709\u5bb3\u4ee3\u7801\u548c\u5de5\u5177\u6548\u5e94\u3002\u73b0\u6709\u8bc4\u4f30\u5f80\u5f80\u5c06\u8fd9\u4e9b\u9636\u6bb5\u5408\u5e76\u4e3a\u5355\u4e00\u7684\u653b\u51fb\u6210\u529f\u7387&#xff0c;\u4f7f\u5f97\u96be\u4ee5\u533a\u5206\u6a21\u578b\u4ec5\u4ec5\u662f\u540c\u610f\u4e86\u653b\u51fb\u8005\u8fd8\u662f\u5b9e\u9645\u4ea7\u751f\u4e86\u53ef\u89c2\u5bdf\u7684\u5371\u5bb3\u3002\u6211\u4eec\u63d0\u51fa\u4e86 SafeClawBench&#xff0c;\u4e00\u4e2a\u9488\u5bf9\u4f7f\u7528\u5de5\u5177\u7684\u4ee3\u7406\u5b89\u5168\u7684\u5206\u9636\u6bb5\u57fa\u51c6\u6d4b\u8bd5&amp;#x","og_url":"https:\/\/www.wsisp.com\/helps\/89490.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-08-02T21:33:32+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260802213330-6a6fb7aa92cbd.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"8 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/89490.html","url":"https:\/\/www.wsisp.com\/helps\/89490.html","name":"SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-08-02T21:33:32+00:00","dateModified":"2026-08-02T21:33:32+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/89490.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/89490.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/89490.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"SafeClawBench\uff1a\u5728\u4f7f\u7528\u5de5\u5177\u7684\u5927\u8bed\u8a00\u6a21\u578b\u4ee3\u7406\u4e2d\u5206\u79bb\u8bed\u4e49\u5371\u5bb3\u3001\u5ba1\u8ba1\u8bc1\u636e\u548c\u6c99\u7bb1\u5371\u5bb3"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/89490","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=89490"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/89490\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/89483"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=89490"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=89490"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=89490"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=89490"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}