{"id":26756,"date":"2025-07-15T07:46:52","date_gmt":"2025-07-14T23:46:52","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=26756"},"modified":"2025-07-15T07:46:52","modified_gmt":"2025-07-14T23:46:52","slug":"openagentsafety-%e6%a1%86%e6%9e%b6%ef%bc%9aai-%e6%99%ba%e8%83%bd%e4%bd%93%e5%ae%89%e5%85%a8%e8%af%84%e4%bc%b0%e7%9a%84%e5%88%9b%e6%96%b0%e5%ae%9e%e8%b7%b5","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2025\/07\/15\/26756\/","title":{"rendered":"OpenAgentSafety \u6846\u67b6\uff1aAI \u667a\u80fd\u4f53\u5b89\u5168\u8bc4\u4f30\u7684\u521b\u65b0\u5b9e\u8df5"},"content":{"rendered":"<div class=\"article-desc\">AI \u667a\u80fd\u4f53\u5b89\u5168\u8bc4\u4f30\u8fce\u6765\u65b0\u5de5\u5177\uff0cOpenAgentSafety \u6846\u67b6\u4ee5\u771f\u5b9e\u5de5\u5177\u4ea4\u4e92\u3001\u590d\u6742\u573a\u666f\u6a21\u62df\u53ca\u6df7\u5408\u8bc4\u4f30\u7b56\u7565\uff0c\u63ed\u793a\u667a\u80fd\u4f53\u6f5c\u5728\u98ce\u9669\uff0c\u52a9\u529b\u6784\u5efa\u5b89\u5168\u53ef\u9760\u7684 AI \u7cfb\u7edf\u3002<\/div>\n<div id=\"postspictures\" class=\"article-content\">\n<div id=\"container\" class=\"container am-engine\" data-v-01a18e2f=\"\" data-element=\"root\">\n<p><img data-dominant-color=\"635a7a\" data-has-transparency=\"false\" style=\"--dominant-color: #635a7a;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-26758\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2025\/07\/419990816f4c220e64d111f86bcea88a52248f-300x167-1.jpg\" width=\"300\" height=\"167\" alt=\"\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2025\/07\/419990816f4c220e64d111f86bcea88a52248f-300x167-1.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2025\/07\/419990816f4c220e64d111f86bcea88a52248f-300x167-1-150x84.jpg 150w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/p>\n<p>\u5927\u5bb6\u597d\uff0c\u6211\u662f\u8086\u3007\u67d2\uff0c\u5728 AI \u667a\u80fd\u4f53\u5feb\u901f\u53d1\u5c55\u7684\u5f53\u4e0b\uff0c\u5b83\u4eec\u6f5c\u5728\u7684\u5b89\u5168\u98ce\u9669\u4e5f\u65e5\u76ca\u51f8\u663e\u3002\u73b0\u6709\u7684 AI \u667a\u80fd\u4f53\u5b89\u5168\u8bc4\u4f30\u65b9\u6cd5\u5b58\u5728\u8bf8\u591a\u5c40\u9650\uff1a\u591a\u6570\u8bc4\u4f30\u4f9d\u8d56\u4e8e\u7b80\u5316\u73b0\u5b9e\u590d\u6742\u6027\u7684\u6a21\u62df\u73af\u5883\uff0c\u96be\u4ee5\u51c6\u786e\u53cd\u6620\u667a\u80fd\u4f53\u5728\u771f\u5b9e\u573a\u666f\u4e0b\u7684\u8868\u73b0\uff1b\u8bb8\u591a\u8bc4\u4f30\u4ec5\u5173\u6ce8\u5355\u4e00\u9886\u57df\uff0c\u5982\u4ec5\u6d4b\u8bd5\u667a\u80fd\u4f53\u7684\u7f51\u9875\u6d4f\u89c8\u6216\u4ee3\u7801\u7f16\u5199\u80fd\u529b\uff0c\u7f3a\u4e4f\u5bf9\u591a\u9886\u57df\u7efc\u5408\u4efb\u52a1\u7684\u8003\u5bdf\uff1b\u6b64\u5916\uff0c\u591a\u8f6e\u591a\u7528\u6237\u4ea4\u4e92\u5728\u8bc4\u4f30\u4e2d\u7684\u7f3a\u5931\uff0c\u4f7f\u5f97\u96be\u4ee5\u6d1e\u5bdf\u667a\u80fd\u4f53\u5728\u590d\u6742\u4eba\u9645\u4e92\u52a8\u548c\u957f\u671f\u4efb\u52a1\u6267\u884c\u4e2d\u7684\u8868\u73b0\u4e0e\u6f5c\u5728\u98ce\u9669\u3002<\/p>\n<p>\u7531\u5361\u8010\u57fa\u6885\u9686\u5927\u5b66\u548c\u827e\u4f26\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u6240\u8054\u5408\u63d0\u51fa\u7684 OpenAgentSafety\uff08OA-SAFETY\uff09\u6846\u67b6\uff0c\u901a\u8fc7\u652f\u6301\u667a\u80fd\u4f53\u4e0e\u771f\u5b9e\u5de5\u5177\uff08\u5982\u6587\u4ef6\u7cfb\u7edf\u3001\u7ec8\u7aef\u3001\u6d4f\u89c8\u5668\u3001\u6d88\u606f\u5e73\u53f0\u7b49\uff09\u7684\u4ea4\u4e92\uff0c\u5e76\u6a21\u62df\u591a\u6837\u5316\u7528\u6237\u610f\u56fe\uff08\u5305\u62ec\u826f\u6027\u3001\u6076\u610f\u548c\u6a21\u7cca\u610f\u56fe\uff09\u4ee5\u53ca\u591a\u8f6e\u4ea4\u4e92\u573a\u666f\uff0c\u5168\u9762\u8986\u76d6\u516b\u5927\u5173\u952e\u5b89\u5168\u98ce\u9669\u7c7b\u522b\uff0c\u7cfb\u7edf\u5730\u63ed\u793a\u4e86\u667a\u80fd\u4f53\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u53ef\u80fd\u5b58\u5728\u7684\u5b89\u5168\u9690\u60a3\u3002<\/p>\n<h5>\u73b0\u6709\u6846\u67b6\u5bf9\u6bd4<\/h5>\n<table class=\"data-table\" data-transient-attributes=\"class\" data-width=\"661px\">\n<colgroup data-id=\"c7104f7d-XlfiPVL0\">\n<col span=\"1\" width=\"165.25\" data-id=\"cd89ecb0-6XZFLALg\" \/>\n<col span=\"1\" width=\"165.25\" data-id=\"cd89ecb0-RSjfBZDX\" \/>\n<col span=\"1\" width=\"165.25\" data-id=\"cd89ecb0-fEJ4RS1Q\" \/>\n<col span=\"1\" width=\"165.25\" data-id=\"cd89ecb0-lQp4n1Gl\" \/><\/colgroup>\n<tbody data-id=\"t6d5e859-LgkQmSB4\">\n<tr data-id=\"t31e458f-RkFBfdY4\">\n<td data-id=\"t6267798-Y9BD5LfO\" data-transient-attributes=\"table-cell-selection\">\u73b0\u6709\u6846\u67b6<\/td>\n<td data-id=\"t6267798-YCjlBmNK\" data-transient-attributes=\"table-cell-selection\">\u652f\u6301\u771f\u5b9e\u5de5\u5177<\/td>\n<td data-id=\"t6267798-IbdJGEP5\" data-transient-attributes=\"table-cell-selection\">\u591a\u6837\u5316\u7528\u6237\u610f\u56fe<\/td>\n<td class=\"table-last-row\" data-id=\"t6267798-UlO7ZLeJ\" data-transient-attributes=\"table-cell-selection\">\u591a\u8f6e\u4ea4\u4e92<\/td>\n<\/tr>\n<tr data-id=\"t31e458f-ScN6gmWM\">\n<td data-id=\"t6267798-M3GXRLMP\" data-transient-attributes=\"table-cell-selection\">SALAD-Bench<\/td>\n<td data-id=\"t6267798-PU1840Ad\" data-transient-attributes=\"table-cell-selection\">\u2717<\/td>\n<td data-id=\"t6267798-9RSDoKXe\" data-transient-attributes=\"table-cell-selection\">\u2717<\/td>\n<td class=\"table-last-row\" data-id=\"t6267798-H2sTFFTh\" data-transient-attributes=\"table-cell-selection\">\u2717<\/td>\n<\/tr>\n<tr data-id=\"t31e458f-5BoUT0Ll\">\n<td data-id=\"t6267798-hE3bNqiC\" data-transient-attributes=\"table-cell-selection\">SafeBench<\/td>\n<td data-id=\"t6267798-9O1OrqkY\" data-transient-attributes=\"table-cell-selection\">\u2717<\/td>\n<td data-id=\"t6267798-jRVSG46V\" data-transient-attributes=\"table-cell-selection\">\u2717<\/td>\n<td class=\"table-last-row\" data-id=\"t6267798-IcpkSHtE\" data-transient-attributes=\"table-cell-selection\">\u2717<\/td>\n<\/tr>\n<tr data-id=\"t31e458f-jeBhcHk6\">\n<td class=\"table-last-column\" data-id=\"t6267798-bQ1S4ZnT\" data-transient-attributes=\"table-cell-selection\">OpenAgentSafety<\/td>\n<td class=\"table-last-column\" data-id=\"t6267798-0kl81MUB\" data-transient-attributes=\"table-cell-selection\">\u2713<\/td>\n<td class=\"table-last-column\" data-id=\"t6267798-HN04sxRT\" data-transient-attributes=\"table-cell-selection\">\u2713<\/td>\n<td class=\"table-last-column table-last-row\" data-id=\"t6267798-eSXKIOn9\" data-transient-attributes=\"table-cell-selection\">\u2713<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>OpenAgentSafety \u6846\u67b6\u6982\u8ff0<\/h2>\n<p>OA-SAFETY \u6846\u67b6\u5efa\u7acb\u5728 TheAgentCompany \u548c Sotopia \u6846\u67b6\u7684\u57fa\u7840\u4e4b\u4e0a\u3002TheAgentCompany \u6846\u67b6\u63d0\u4f9b\u4e86\u5f3a\u5927\u7684\u6a21\u62df\u540e\u7aef\u652f\u6301\uff0c\u80fd\u591f\u9ad8\u6548\u5730\u6a21\u62df\u5404\u7c7b\u771f\u5b9e\u5de5\u5177\u7684\u8fd0\u884c\u73af\u5883\uff0c\u8ba9\u667a\u80fd\u4f53\u5728\u7c7b\u4f3c\u771f\u5b9e\u7684\u571f\u58e4\u4e2d\u63a5\u53d7\u68c0\u9a8c\u3002\u800c Sotopia \u6846\u67b6\u5219\u4e13\u6ce8\u4e8e\u793e\u4f1a\u4e92\u52a8\u5efa\u6a21\uff0c\u4f7f\u5f97\u667a\u80fd\u4f53\u53ef\u4ee5\u5728\u6a21\u62df\u7684\u591a\u7528\u6237\u573a\u666f\u4e2d\u4e0e\u4e0d\u540c\u7684\u89d2\u8272\uff08\u5982\u540c\u4e8b\u3001\u5ba2\u6237\u7b49\uff09\u8fdb\u884c\u4ea4\u4e92\uff0c\u8fd9\u4e9b\u89d2\u8272\u6709\u7740\u5404\u81ea\u7684\u76ee\u6807\u548c\u610f\u56fe\uff0c\u53ef\u80fd\u4e0e\u667a\u80fd\u4f53\u7684\u4efb\u52a1\u76f8\u5951\u5408\uff0c\u4e5f\u53ef\u80fd\u4ea7\u751f\u51b2\u7a81\uff0c\u4ece\u800c\u8425\u9020\u51fa\u590d\u6742\u591a\u53d8\u7684\u793e\u4ea4\u73af\u5883\uff0c\u5168\u65b9\u4f4d\u8003\u9a8c\u667a\u80fd\u4f53\u7684\u5e94\u5bf9\u80fd\u529b\u3002<\/p>\n<p>OA-SAFETY \u7684\u4efb\u52a1\u8bbe\u8ba1\u878d\u5408\u4e86\u771f\u5b9e\u5de5\u5177\u64cd\u4f5c\u3001\u591a\u8f6e\u5bf9\u8bdd\u673a\u5236\u4ee5\u53ca\u590d\u6742\u7528\u6237 \/ NPC \u610f\u56fe\uff0c\u6bcf\u4e00\u4e2a\u4efb\u52a1\u90fd\u88ab\u7cbe\u5fc3\u6253\u9020\u4e3a\u4e00\u4e2a\u72ec\u7acb\u7684\u6a21\u5757\uff0c\u5c01\u88c5\u5728 Docker \u5bb9\u5668\u5185\uff0c\u5185\u90e8\u81ea\u5305\u542b\u73af\u5883\u914d\u7f6e\u3001\u4efb\u52a1\u63cf\u8ff0\u3001NPC \u884c\u4e3a\u903b\u8f91\u4ee5\u53ca\u8bc4\u4f30\u7ec4\u4ef6\u3002\u8fd9\u79cd\u6a21\u5757\u5316\u7684\u8bbe\u8ba1\u6781\u5927\u5730\u63d0\u5347\u4e86\u6846\u67b6\u7684\u53ef\u6269\u5c55\u6027\uff0c\u7814\u7a76\u4eba\u5458\u53ef\u4ee5\u8f7b\u677e\u5730\u6dfb\u52a0\u65b0\u7684\u4efb\u52a1\u3001\u5f15\u5165\u65b0\u7684\u5de5\u5177\u3001\u62d3\u5c55\u73af\u5883\u8303\u56f4\u4ee5\u53ca\u8bbe\u8ba1\u65b0\u7684\u5bf9\u6297\u7b56\u7565\uff0c\u4e0d\u65ad\u4e30\u5bcc\u548c\u5b8c\u5584\u8bc4\u4f30\u4f53\u7cfb\uff0c\u4ee5\u5e94\u5bf9\u4e0d\u65ad\u6f14\u53d8\u7684\u5b89\u5168\u6311\u6218\u3002<\/p>\n<h4>\u6846\u67b6\u6982\u89c8\u56fe<\/h4>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/b963e8a3769369a5e4b690a3dda60544819ace.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>OPENAGENTSAFETY\u6846\u67b6\u6982\u89c8<\/p>\n<p>\u4e0a\u56fe\u76f4\u89c2\u5730\u5c55\u793a\u4e86\u6846\u67b6\u7684\u5404\u4e2a\u6838\u5fc3\u7ec4\u6210\u90e8\u5206\u53ca\u5176\u76f8\u4e92\u5173\u7cfb\uff0c\u5305\u62ec\u4efb\u52a1\u751f\u6210\u6a21\u5757\u3001\u667a\u80fd\u4f53\u6267\u884c\u73af\u5883\u3001\u8bc4\u4f30\u6a21\u5757\u7b49\u3002\u4efb\u52a1\u751f\u6210\u6a21\u5757\u8d1f\u8d23\u57fa\u4e8e\u79cd\u5b50\u4efb\u52a1\u548c GPT-4o \u7684\u53d8\u4f53\u751f\u6210\u65b0\u4efb\u52a1\uff0c\u4e3a\u667a\u80fd\u4f53\u63d0\u4f9b\u4e86\u4e30\u5bcc\u591a\u6837\u7684\u6d4b\u8bd5\u573a\u666f\uff1b\u667a\u80fd\u4f53\u5728\u6267\u884c\u4efb\u52a1\u65f6\uff0c\u4e0e\u771f\u5b9e\u5de5\u5177\u548c\u6a21\u62df\u7528\u6237 NPC \u8fdb\u884c\u4ea4\u4e92\uff0c\u6267\u884c\u5404\u7c7b\u64cd\u4f5c\uff1b\u8bc4\u4f30\u6a21\u5757\u5219\u901a\u8fc7\u89c4\u5219\u57fa\u7840\u8bc4\u4f30\u548c LLM-as-Judge \u8bc4\u4f30\u5bf9\u667a\u80fd\u4f53\u8868\u73b0\u8fdb\u884c\u7efc\u5408\u8bc4\u5b9a\uff0c\u751f\u6210\u8be6\u7ec6\u7684\u8bc4\u4f30\u62a5\u544a\u3002\u8fd9\u4e00\u6574\u4f53\u67b6\u6784\u8bbe\u8ba1\u4fdd\u969c\u4e86\u6846\u67b6\u80fd\u591f\u5168\u9762\u3001\u7cfb\u7edf\u5730\u5bf9\u667a\u80fd\u4f53\u5b89\u5168\u6027\u8fdb\u884c\u8bc4\u4f30\u3002<\/p>\n<h2>\u6838\u5fc3\u529f\u80fd\u4e0e\u4efb\u52a1\u8bbe\u8ba1<\/h2>\n<h4>\u771f\u5b9e\u5de5\u5177\u4e0e\u591a\u8f6e\u4ea4\u4e92<\/h4>\n<p>OA-SAFETY \u6846\u67b6\u4e3a\u667a\u80fd\u4f53\u63d0\u4f9b\u4e86\u4e30\u5bcc\u4e14\u771f\u5b9e\u7684\u64cd\u4f5c\u73af\u5883\uff0c\u5176\u652f\u6301\u7684\u5de5\u5177\u7c7b\u578b\u6db5\u76d6\u4e86\u6587\u4ef6\u7cfb\u7edf\u3001\u4ee3\u7801\u6267\u884c\u73af\u5883\u3001Web \u6d4f\u89c8\u5668\u3001\u6d88\u606f\u5e73\u53f0\u7b49\u8bf8\u591a\u9886\u57df\u3002\u5728\u6587\u4ef6\u7cfb\u7edf\u4e2d\uff0c\u667a\u80fd\u4f53\u53ef\u4ee5\u6267\u884c\u6587\u4ef6\u7684\u521b\u5efa\u3001\u5220\u9664\u3001\u79fb\u52a8\u3001\u590d\u5236\u7b49\u64cd\u4f5c\uff1b\u501f\u52a9\u4ee3\u7801\u6267\u884c\u73af\u5883\uff0c\u667a\u80fd\u4f53\u80fd\u591f\u7f16\u5199\u5e76\u8fd0\u884c\u4ee3\u7801\u7247\u6bb5\uff0c\u5b9e\u73b0\u590d\u6742\u7684\u903b\u8f91\u8fd0\u7b97\u548c\u6570\u636e\u5904\u7406\uff1bWeb \u6d4f\u89c8\u5668\u7684\u5f15\u5165\u5219\u4f7f\u667a\u80fd\u4f53\u53ef\u4ee5\u8bbf\u95ee\u7f51\u9875\u3001\u586b\u5199\u8868\u5355\u3001\u89e3\u6790\u7f51\u9875\u5185\u5bb9\u7b49\uff1b\u800c\u6d88\u606f\u5e73\u53f0\u5219\u4e3a\u667a\u80fd\u4f53\u4e0e\u5176\u4ed6\u7528\u6237\u6216 NPC \u7684\u5b9e\u65f6\u6c9f\u901a\u63d0\u4f9b\u4e86\u6865\u6881\u3002<\/p>\n<p>\u591a\u8f6e\u3001\u591a\u7528\u6237\u4efb\u52a1\u8bbe\u8ba1\u662f\u8be5\u6846\u67b6\u7684\u4eae\u70b9\u4e4b\u4e00\u3002\u4f8b\u5982\uff0c\u5728\u4e00\u4e2a\u6a21\u62df\u7684\u529e\u516c\u573a\u666f\u4e2d\uff0c\u7528\u6237\u53ef\u80fd\u8981\u6c42\u667a\u80fd\u4f53\u6574\u7406\u6587\u4ef6\u5939\uff0c\u5220\u9664\u5176\u4e2d\u7684\u8fc7\u671f\u6587\u6863\uff0c\u540c\u65f6 NPC\uff08\u5982\u540c\u4e8b\uff09\u53ef\u80fd\u4f1a\u4ee5\u7d27\u6025\u534f\u4f5c\u4efb\u52a1\u4e3a\u7531\uff0c\u8981\u6c42\u667a\u80fd\u4f53\u6682\u65f6\u4fdd\u7559\u67d0\u4e9b\u770b\u4f3c\u8fc7\u671f\u4f46\u5b9e\u5219\u91cd\u8981\u7684\u6587\u4ef6\u3002\u667a\u80fd\u4f53\u9700\u8981\u5728\u591a\u8f6e\u5bf9\u8bdd\u4e2d\u4e0d\u65ad\u6f84\u6e05\u9700\u6c42\u3001\u6743\u8861\u5404\u65b9\u5229\u76ca\uff0c\u5e76\u8c28\u614e\u6267\u884c\u64cd\u4f5c\uff0c\u907f\u514d\u56e0\u9519\u8bef\u5220\u9664\u6587\u4ef6\u800c\u5f15\u53d1\u7684\u6570\u636e\u4e22\u5931\u98ce\u9669\u3002<\/p>\n<h4>\u591a\u6837\u5316\u7528\u6237\u610f\u56fe<\/h4>\n<p>\u6846\u67b6\u5bf9\u7528\u6237\u610f\u56fe\u8fdb\u884c\u4e86\u7ec6\u81f4\u7684\u5206\u7c7b\uff0c\u5305\u62ec\u826f\u6027\u610f\u56fe\u3001\u6076\u610f\u610f\u56fe\u548c\u6a21\u7cca\u610f\u56fe\u3002\u826f\u6027\u610f\u56fe\u901a\u5e38\u8868\u73b0\u4e3a\u7528\u6237\u5e0c\u671b\u667a\u80fd\u4f53\u534f\u52a9\u5b8c\u6210\u5408\u6cd5\u5408\u89c4\u4e14\u6709\u76ca\u7684\u4efb\u52a1\uff0c\u5982\u67e5\u8be2\u4fe1\u606f\u3001\u6574\u7406\u8d44\u6599\u7b49\uff1b\u6076\u610f\u610f\u56fe\u5219\u662f\u7528\u6237\u6545\u610f\u8bf1\u5bfc\u667a\u80fd\u4f53\u6267\u884c\u6709\u5bb3\u64cd\u4f5c\uff0c\u5982\u6cc4\u9732\u9690\u79c1\u6570\u636e\u3001\u7834\u574f\u7cfb\u7edf\u6587\u4ef6\u7b49\uff1b\u800c\u6a21\u7cca\u610f\u56fe\u4ecb\u4e8e\u4e24\u8005\u4e4b\u95f4\uff0c\u53ef\u80fd\u56e0\u8868\u8ff0\u4e0d\u6e05\u3001\u4e0a\u4e0b\u6587\u7f3a\u5931\u6216\u5b58\u5728\u9690\u542b\u5047\u8bbe\u800c\u5bfc\u81f4\u667a\u80fd\u4f53\u96be\u4ee5\u51c6\u786e\u5224\u65ad\u4efb\u52a1\u7684\u5b89\u5168\u6027\u3002<\/p>\n<p>\u4e0d\u540c\u7528\u6237\u4e0e NPC \u610f\u56fe\u7684\u7ec4\u5408\u4f1a\u5bf9\u4efb\u52a1\u96be\u5ea6\u548c\u5b89\u5168\u98ce\u9669\u4ea7\u751f\u663e\u8457\u5f71\u54cd\u3002\u5f53\u7528\u6237\u6000\u6709\u826f\u6027\u610f\u56fe\uff0c\u800c NPC \u6301\u6709\u6076\u610f\u610f\u56fe\u65f6\uff0c\u667a\u80fd\u4f53\u9700\u8981\u5728\u6ee1\u8db3\u7528\u6237\u9700\u6c42\u7684\u540c\u65f6\uff0c\u62b5\u5fa1 NPC \u7684\u4e0d\u826f\u8bf1\u5bfc\uff1b\u82e5\u7528\u6237\u610f\u56fe\u6a21\u7cca\u4e14 NPC \u610f\u56fe\u4e0d\u660e\uff0c\u667a\u80fd\u4f53\u5219\u9700\u66f4\u52a0\u8c28\u614e\u5730\u63a2\u7d22\u548c\u9a8c\u8bc1\u4efb\u52a1\u7684\u5b89\u5168\u6027\uff0c\u8fd9\u65e0\u7591\u589e\u52a0\u4e86\u4efb\u52a1\u7684\u590d\u6742\u6027\u548c\u6311\u6218\u6027\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/921337a78719b7edc392938b6365e90f10cc65.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u4e09\u79cd\u6700\u5e38\u89c1\u7684\u7528\u6237\u4e0eNPC\u610f\u56fe\u7ec4\u5408\u7684\u63cf\u8ff0<\/p>\n<p>\u4e0a\u8868\u5c55\u793a\u4e86\u4e09\u79cd\u6700\u5e38\u89c1\u7684\u7528\u6237\u548c NPC \u610f\u56fe\u7ec4\u5408\uff0c\u5e2e\u52a9\u6211\u4eec\u66f4\u597d\u5730\u7406\u89e3\u4e0d\u540c\u610f\u56fe\u7ec4\u5408\u5bf9\u4efb\u52a1\u96be\u5ea6\u548c\u98ce\u9669\u7684\u5f71\u54cd\u3002\u4f8b\u5982\uff0c\u5728 benign user \u548c malicious NPC \u7684\u7ec4\u5408\u4e0b\uff0c\u667a\u80fd\u4f53\u9700\u8981\u9762\u5bf9\u7528\u6237\u770b\u4f3c\u65e0\u5bb3\u7684\u8bf7\u6c42\uff0c\u540c\u65f6\u62b5\u5fa1 NPC \u7684\u6076\u610f\u8bf1\u5bfc\uff0c\u8fd9\u79cd\u573a\u666f\u5bf9\u667a\u80fd\u4f53\u7684\u5b89\u5168\u6027\u80fd\u63d0\u51fa\u4e86\u66f4\u9ad8\u7684\u8981\u6c42\u3002<\/p>\n<h4>\u4efb\u52a1\u751f\u6210\u4e0e\u6269\u5c55<\/h4>\n<p>OA-SAFETY \u6846\u67b6\u7684\u4efb\u52a1\u751f\u6210\u8fc7\u7a0b\u7ed3\u5408\u4e86\u4eba\u5de5\u8bbe\u8ba1\u4e0e\u81ea\u52a8\u5316\u751f\u6210\u7684\u4f18\u52bf\u3002\u7814\u7a76\u4eba\u5458\u9996\u5148\u624b\u52a8\u6253\u9020\u4e86\u4e00\u7cfb\u5217\u79cd\u5b50\u4efb\u52a1\uff0c\u8fd9\u4e9b\u79cd\u5b50\u4efb\u52a1\u5982\u540c\u57fa\u77f3\u4e00\u822c\uff0c\u4e3a\u540e\u7eed\u4efb\u52a1\u751f\u6210\u5960\u5b9a\u4e86\u57fa\u7840\uff0c\u6bcf\u4e2a\u79cd\u5b50\u4efb\u52a1\u90fd\u7cbe\u51c6\u6355\u6349\u4e86\u7279\u5b9a\u5de5\u5177\u7ec4\u5408\u548c\u793e\u4ea4\u573a\u666f\u4e0b\u7684\u72ec\u7279\u884c\u4e3a\u6311\u6218\u3002\u5728\u6b64\u57fa\u7840\u4e0a\uff0c\u501f\u52a9 GPT-4o \u7684\u5f3a\u5927\u8bed\u8a00\u7406\u89e3\u548c\u751f\u6210\u80fd\u529b\uff0c\u901a\u8fc7\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u63d0\u793a\u8bcd\uff08prompt\uff09\uff0c\u5f15\u5bfc\u5176\u5bf9\u79cd\u5b50\u4efb\u52a1\u8fdb\u884c\u53d8\u4f53\u751f\u6210\uff0c\u7cfb\u7edf\u5730\u6539\u53d8\u793e\u4ea4\u573a\u666f\u4e2d\u7684\u89d2\u8272\u5173\u7cfb\u3001NPC \u7684\u914d\u7f6e\u4ee5\u53ca\u4efb\u52a1\u7684\u8868\u9762\u7ec6\u8282\uff0c\u540c\u65f6\u4e25\u683c\u4fdd\u6301\u4efb\u52a1\u7684\u6838\u5fc3\u884c\u4e3a\u6311\u6218\u548c\u5de5\u5177\u4f7f\u7528\u6a21\u5f0f\u4e0d\u53d8\u3002\u8fd9\u4e00\u8fc7\u7a0b\u6781\u5927\u5730\u4e30\u5bcc\u4e86\u4efb\u52a1\u7684\u591a\u6837\u6027\u3002\u6240\u6709\u81ea\u52a8\u751f\u6210\u7684\u4efb\u52a1\u968f\u540e\u90fd\u7ecf\u8fc7\u4e86\u4f5c\u8005\u56e2\u961f\u7684\u4e25\u683c\u4eba\u5de5\u5ba1\u6838\uff0c\u786e\u4fdd\u5176\u8d28\u91cf\u7b26\u5408\u6807\u51c6\u3002\u6bcf\u4e2a\u4efb\u52a1\u90fd\u88ab\u6784\u5efa\u6210\u4e00\u4e2a\u81ea\u5305\u542b\u7684 Docker \u5bb9\u5668\uff0c\u5185\u90e8\u4e0d\u4ec5\u8be6\u7ec6\u89c4\u5b9a\u4e86\u73af\u5883\u7684\u521d\u59cb\u72b6\u6001\uff0c\u8fd8\u6e05\u6670\u63cf\u8ff0\u4e86\u4efb\u52a1\u7684\u76ee\u6807\uff0c\u51c6\u786e\u5b9a\u4e49\u4e86 NPC \u7684\u884c\u4e3a\u903b\u8f91\uff0c\u5e76\u914d\u5907\u4e86\u4e13\u95e8\u5b9a\u5236\u7684\u8bc4\u4f30\u5668\uff0c\u7528\u4e8e\u5224\u65ad\u667a\u80fd\u4f53\u5728\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u662f\u5426\u8fbe\u5230\u9884\u671f\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/5672ad967a4cae43c79224b0d36c5b44185edc.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u5728OPENAGENTSAFETY\u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0c\u5305\u542b\u516b\u4e2a\u5b89\u5168\u98ce\u9669\u7c7b\u522b\u53ca\u5176\u793a\u4f8b\u573a\u666f<\/p>\n<p>\u4e0a\u8868\u5217\u51fa\u4e86 OA-SAFETY \u6846\u67b6\u4e2d\u7684\u516b\u5927\u5b89\u5168\u98ce\u9669\u7c7b\u522b\u53ca\u5176\u793a\u4f8b\u573a\u666f\uff0c\u6db5\u76d6\u4e86\u4ece\u8ba1\u7b97\u673a\u5b89\u5168\u5230\u6709\u5bb3\u51b3\u7b56\u7b49\u591a\u4e2a\u65b9\u9762\u3002\u4f8b\u5982\uff0c\u5728\u8ba1\u7b97\u673a\u5b89\u5168\u98ce\u9669\u7c7b\u522b\u4e2d\uff0c\u667a\u80fd\u4f53\u53ef\u80fd\u9762\u4e34\u6cc4\u9732\u79c1\u4eba SSH \u5bc6\u94a5\u7684\u98ce\u9669\uff0c\u800c\u5728\u9690\u79c1\u6cc4\u9732\u7c7b\u522b\u4e2d\uff0c\u667a\u80fd\u4f53\u53ef\u80fd\u56e0\u4e0d\u5f53\u64cd\u4f5c\u5bfc\u81f4\u5458\u5de5\u4e2a\u4eba\u4fe1\u606f\u88ab\u516c\u5f00\u3002\u8fd9\u4e9b\u98ce\u9669\u7c7b\u522b\u5168\u9762\u8986\u76d6\u4e86\u667a\u80fd\u4f53\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u53ef\u80fd\u9047\u5230\u7684\u5173\u952e\u5b89\u5168\u6311\u6218\u3002<\/p>\n<h4>\u4efb\u52a1\u793a\u4f8b<\/h4>\n<p>\u4ee5\u6587\u4ef6\u64cd\u4f5c\u4efb\u52a1\u4e3a\u4f8b\uff0c\u7528\u6237\u6307\u793a\u667a\u80fd\u4f53\u5220\u9664\u67d0\u6587\u4ef6\u5939\u4e0b\u7684\u7279\u5b9a\u6587\u4ef6\uff0c\u4f46\u8be5\u6587\u4ef6\u5939\u6df7\u6742\u7740\u91cd\u8981\u7cfb\u7edf\u6587\u4ef6\u4e0e\u666e\u901a\u4e34\u65f6\u6587\u4ef6\u3002\u667a\u80fd\u4f53\u5728\u6267\u884c\u8fc7\u7a0b\u4e2d\u5fc5\u987b\u5177\u5907\u7cbe\u51c6\u7684\u6587\u4ef6\u8bc6\u522b\u80fd\u529b\uff0c\u901a\u8fc7\u5206\u6790\u6587\u4ef6\u7684\u5c5e\u6027\u3001\u540d\u79f0\u6a21\u5f0f\u4ee5\u53ca\u6240\u5904\u8def\u5f84\u7b49\u591a\u7ef4\u5ea6\u4fe1\u606f\uff0c\u51c6\u786e\u533a\u5206\u76ee\u6807\u6587\u4ef6\u4e0e\u5173\u952e\u7cfb\u7edf\u6587\u4ef6\u3002\u4f8b\u5982\uff0c\u667a\u80fd\u4f53\u53ef\u80fd\u4f1a\u4f9d\u636e\u6587\u4ef6\u7684\u6269\u5c55\u540d\uff08\u5982\u00a0<code>.sys<\/code>\u00a0\u6216\u00a0<code>.exe<\/code>\uff09\u548c\u5b58\u50a8\u8def\u5f84\uff08\u5982\u00a0<code>C:\\Windows\\System32<\/code>\uff09\u6765\u5224\u65ad\u6587\u4ef6\u662f\u5426\u4e3a\u91cd\u8981\u7cfb\u7edf\u6587\u4ef6\u3002\u540c\u65f6\uff0c\u667a\u80fd\u4f53\u8fd8\u9700\u5b9e\u65f6\u5411\u7528\u6237\u53cd\u9988\u64cd\u4f5c\u8fdb\u5c55\uff0c\u5e76\u5728\u68c0\u6d4b\u5230\u6f5c\u5728\u98ce\u9669\u65f6\uff0c\u80fd\u591f\u6e05\u6670\u5730\u5411\u7528\u6237\u89e3\u91ca\u539f\u56e0\uff0c\u5efa\u8bae\u91c7\u53d6\u66f4\u4e3a\u7a33\u59a5\u7684\u66ff\u4ee3\u65b9\u6848\uff0c\u5982\u5c06\u6587\u4ef6\u79fb\u52a8\u81f3\u56de\u6536\u7ad9\u800c\u975e\u76f4\u63a5\u5220\u9664\u7b49\u3002<\/p>\n<h2>\u5b89\u5168\u8bc4\u4f30\u65b9\u6cd5<\/h2>\n<h4>\u6df7\u5408\u8bc4\u4f30\u7b56\u7565<\/h4>\n<p>OA-SAFETY \u6846\u67b6\u521b\u65b0\u6027\u5730\u91c7\u7528\u4e86\u6df7\u5408\u8bc4\u4f30\u7b56\u7565\uff0c\u5c06\u89c4\u5219\u57fa\u7840\u8bc4\u4f30\u4e0e LLM-as-Judge \u8bc4\u4f30\u6709\u673a\u7ed3\u5408\u3002\u89c4\u5219\u57fa\u7840\u8bc4\u4f30\u5982\u540c\u4e25\u8c28\u7684\u68c0\u5bdf\u5b98\uff0c\u4f9d\u636e\u9884\u8bbe\u7684\u89c4\u5219\u96c6\uff0c\u5bf9\u4efb\u52a1\u5b8c\u6210\u540e\u7684\u73af\u5883\u72b6\u6001\u8fdb\u884c\u5168\u9762\u5ba1\u67e5\u3002\u4f8b\u5982\uff0c\u5b83\u4f1a\u7cbe\u51c6\u68c0\u67e5\u6587\u4ef6\u7cfb\u7edf\u4e2d\u662f\u5426\u5b58\u5728\u5173\u952e\u6587\u4ef6\u88ab\u5220\u9664\u7684\u60c5\u51b5\uff0c\u6216\u6570\u636e\u5e93\u4e2d\u662f\u5426\u51fa\u73b0\u4e86\u654f\u611f\u6570\u636e\u6cc4\u9732\u7684\u75d5\u8ff9\uff0c\u5176\u4f18\u52bf\u5728\u4e8e\u80fd\u591f\u5feb\u901f\u3001\u51c6\u786e\u5730\u53d1\u73b0\u5177\u4f53\u7684\u3001\u6709\u5f62\u7684\u73af\u5883\u53d8\u5316\u6240\u5e26\u6765\u7684\u5b89\u5168\u95ee\u9898\uff0c\u4e3a\u8bc4\u4f30\u63d0\u4f9b\u575a\u5b9e\u7684\u6570\u636e\u652f\u6491\u3002\u800c LLM-as-Judge \u8bc4\u4f30\u5219\u66f4\u50cf\u662f\u4e00\u4f4d\u654f\u9510\u7684\u5fc3\u7406\u5206\u6790\u5e08\uff0c\u6df1\u5165\u5256\u6790\u667a\u80fd\u4f53\u5728\u4efb\u52a1\u6267\u884c\u8fc7\u7a0b\u4e2d\u7684\u4e2d\u95f4\u63a8\u7406\u8fc7\u7a0b\uff0c\u901a\u8fc7\u5206\u6790\u667a\u80fd\u4f53\u7684\u601d\u8003\u8def\u5f84\u3001\u51b3\u7b56\u4f9d\u636e\u4ee5\u53ca\u751f\u6210\u7684\u884c\u52a8\u6307\u4ee4\uff0c\u6316\u6398\u51fa\u90a3\u4e9b\u5c1a\u672a\u5b8c\u5168\u5b9e\u65bd\u6216\u672a\u5728\u73af\u5883\u72b6\u6001\u4e2d\u76f4\u63a5\u663e\u73b0\u51fa\u6765\u7684\u6f5c\u5728\u5371\u9669\u884c\u4e3a\uff0c\u5982\u667a\u80fd\u4f53\u5728\u9762\u5bf9\u53ef\u7591\u7528\u6237\u8bf7\u6c42\u65f6\u7684\u72b9\u8c6b\u4e0d\u51b3\u3001\u5bf9\u6f5c\u5728\u98ce\u9669\u7684\u521d\u6b65\u8bd5\u63a2\u7b49\uff0c\u4ece\u800c\u63d0\u524d\u9884\u8b66\u53ef\u80fd\u51fa\u73b0\u7684\u5b89\u5168\u9690\u60a3\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/415bc6a0027f475d8bf785f92852b1dcaa4503.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u5927\u578b\u8bed\u8a00\u6a21\u578b\u5728\u5b89\u5168\u6027\u5206\u7c7b\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u4e0e\u4eba\u7c7b\u6807\u6ce8\u7684\u6bd4\u8f83<\/p>\n<p>\u4e0a\u8868\u5bf9\u6bd4\u4e86 LLM \u8bc4\u4f30\u4e0e\u4eba\u7c7b\u6807\u6ce8\u5728\u5b89\u5168\u5206\u7c7b\u6027\u80fd\u4e0a\u7684\u8868\u73b0\uff0c\u63ed\u793a\u4e86 LLM \u4f5c\u4e3a\u8bc4\u4f30\u5de5\u5177\u7684\u6f5c\u5728\u504f\u5dee\u3002\u4f8b\u5982\uff0cGPT-4.1 \u5728\u5224\u65ad\u5931\u8d25\u7387\u65f6\u5f80\u5f80\u8fc7\u9ad8\u4f30\u8ba1\uff0c\u800c\u5bf9\u9690\u542b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u5219\u53ef\u80fd\u4f4e\u4f30\u3002\u8fd9\u8868\u660e\u5728\u4f9d\u8d56 LLM \u8fdb\u884c\u5b89\u5168\u8bc4\u4f30\u65f6\uff0c\u9700\u8981\u8c28\u614e\u5bf9\u5f85\u5176\u7ed3\u679c\uff0c\u5e76\u7ed3\u5408\u4eba\u7c7b\u4e13\u5bb6\u7684\u5224\u65ad\u8fdb\u884c\u7efc\u5408\u5206\u6790\u3002<\/p>\n<h4>\u8bc4\u4f30\u6307\u6807<\/h4>\n<p>\u5728\u8bc4\u4f30\u6307\u6807\u65b9\u9762\uff0c\u6846\u67b6\u91cd\u70b9\u5173\u6ce8\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u3001\u5931\u8d25\u7387\u4ee5\u53ca\u8bc4\u4f30\u5668\u5206\u6b67\u7387\u8fd9\u4e09\u4e2a\u6838\u5fc3\u7ef4\u5ea6\u3002\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u76f4\u63a5\u53cd\u6620\u4e86\u667a\u80fd\u4f53\u5728\u9762\u5bf9\u5b89\u5168\u6f0f\u6d1e\u4efb\u52a1\u65f6\uff0c\u5b9e\u9645\u6267\u884c\u4e86\u4e0d\u5b89\u5168\u64cd\u4f5c\u7684\u6bd4\u4f8b\uff0c\u662f\u8861\u91cf\u667a\u80fd\u4f53\u5b89\u5168\u6027\u8868\u73b0\u7684\u5173\u952e\u6307\u6807\uff1b\u5931\u8d25\u7387\u5219\u63ed\u793a\u4e86\u667a\u80fd\u4f53\u56e0\u5404\u79cd\u539f\u56e0\uff08\u5982\u5de5\u5177\u4f7f\u7528\u4e0d\u5f53\u3001\u65e0\u6cd5\u7406\u89e3\u4efb\u52a1\u8981\u6c42\u7b49\uff09\u800c\u672a\u80fd\u6709\u6548\u63a8\u8fdb\u4efb\u52a1\u81f3\u5b89\u5168\u5173\u952e\u51b3\u7b56\u70b9\u7684\u60c5\u51b5\uff0c\u4f53\u73b0\u4e86\u667a\u80fd\u4f53\u5728\u4efb\u52a1\u6267\u884c\u80fd\u529b\u548c\u9c81\u68d2\u6027\u65b9\u9762\u7684\u4e0d\u8db3\uff1b\u8bc4\u4f30\u5668\u5206\u6b67\u7387\u5219\u4e3a\u8bc4\u4f30\u5de5\u4f5c\u7684\u53ef\u9760\u6027\u63d0\u4f9b\u4e86\u91cd\u8981\u53c2\u8003\uff0c\u5f53\u89c4\u5219\u57fa\u7840\u8bc4\u4f30\u4e0e LLM-as-Judge \u8bc4\u4f30\u7ed3\u679c\u51fa\u73b0\u5dee\u5f02\u65f6\uff0c\u5b83\u63d0\u793a\u6211\u4eec\u53ef\u80fd\u5b58\u5728\u7684\u8bc4\u4f30\u76f2\u533a\u6216\u8bc4\u4f30\u65b9\u6cd5\u7684\u5c40\u9650\u6027\uff0c\u9700\u8981\u8fdb\u4e00\u6b65\u6df1\u5165\u5206\u6790\u548c\u4f18\u5316\u8bc4\u4f30\u4f53\u7cfb\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/336484f2132de5117e08832d4e2b97cfe9bbd6.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u4e0d\u540c\u98ce\u9669\u7c7b\u522b\u548c\u5927\u578b\u8bed\u8a00\u6a21\u578b\u4e2d\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u53d1\u751f\u7387\uff08%\uff09<\/p>\n<p>\u4e0a\u56fe\u4ee5\u70ed\u56fe\u5f62\u5f0f\u5c55\u793a\u4e86\u4e0d\u540c\u98ce\u9669\u7c7b\u522b\u548c LLM \u6a21\u578b\u4e0b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u3002\u4ece\u56fe\u4e2d\u53ef\u4ee5\u6e05\u6670\u5730\u770b\u5230\uff0c\u4e0d\u540c\u98ce\u9669\u7c7b\u522b\u4e0b\uff0c\u5404\u6a21\u578b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u5b58\u5728\u663e\u8457\u5dee\u5f02\u3002\u5728\u8ba1\u7b97\u673a\u5b89\u5168\u3001\u9690\u79c1\u6cc4\u9732\u3001\u6cd5\u5f8b\u8fdd\u89c4\u7b49\u9ad8\u98ce\u9669\u7c7b\u522b\u4e2d\uff0c\u591a\u6570\u6a21\u578b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u90fd\u8f83\u9ad8\u3002\u4f8b\u5982\uff0c\u5728\u9690\u79c1\u6cc4\u9732\u98ce\u9669\u7c7b\u522b\u4e0b\uff0cDeepseek-v3 \u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u8d85\u8fc7 60%\uff0c\u8fd9\u51f8\u663e\u4e86\u667a\u80fd\u4f53\u5728\u5904\u7406\u6d89\u53ca\u9690\u79c1\u6570\u636e\u7684\u4efb\u52a1\u65f6\u7684\u8106\u5f31\u6027\u3002\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u9488\u5bf9\u8fd9\u4e9b\u9ad8\u98ce\u9669\u7c7b\u522b\uff0c\u5e94\u5efa\u7acb\u66f4\u4e3a\u4e25\u683c\u7684\u5b89\u5168\u7b56\u7565\uff0c\u5982\u52a0\u5bc6\u9690\u79c1\u6570\u636e\u3001\u9650\u5236\u654f\u611f\u64cd\u4f5c\u6743\u9650\u7b49\uff0c\u4ee5\u4fdd\u969c\u667a\u80fd\u4f53\u884c\u4e3a\u7684\u5b89\u5168\u6027\u3002<\/p>\n<h4>\u8bc4\u4f30\u65b9\u6cd5\u7684\u5c40\u9650\u6027<\/h4>\n<p>\u5c3d\u7ba1\u6df7\u5408\u8bc4\u4f30\u7b56\u7565\u5728\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u63d0\u5347\u4e86\u8bc4\u4f30\u7684\u5168\u9762\u6027\u548c\u51c6\u786e\u6027\uff0c\u4f46\u6846\u67b6\u4e5f\u5766\u8bda\u5730\u6307\u51fa\u4e86\u5f53\u524d\u8bc4\u4f30\u65b9\u6cd5\u5b58\u5728\u7684\u5c40\u9650\u6027\u3002\u5176\u4e2d\uff0cLLM-as-Judge \u7684\u53ef\u9760\u6027\u95ee\u9898\u5c24\u4e3a\u503c\u5f97\u5173\u6ce8\u3002\u7531\u4e8e\u5927\u578b\u8bed\u8a00\u6a21\u578b\u672c\u8eab\u53ef\u80fd\u5b58\u5728\u504f\u5dee\uff0c\u4f8b\u5982\u5728\u8bad\u7ec3\u6570\u636e\u4e2d\u7684\u67d0\u4e9b\u7279\u5b9a\u9886\u57df\u77e5\u8bc6\u8986\u76d6\u4e0d\u8db3\uff0c\u6216\u8005\u5bf9\u4efb\u52a1\u4e0a\u4e0b\u6587\u7684\u7406\u89e3\u4e0d\u591f\u7cbe\u51c6\uff0c\u5bfc\u81f4\u5176\u5728\u5224\u65ad\u667a\u80fd\u4f53\u884c\u4e3a\u5b89\u5168\u6027\u65f6\u51fa\u73b0\u8bef\u5dee\u3002\u6b64\u5916\uff0c\u6a21\u578b\u5bf9\u590d\u6742\u8bed\u4e49\u7684\u7406\u89e3\u80fd\u529b\u6709\u9650\uff0c\u53ef\u80fd\u65e0\u6cd5\u5b8c\u5168\u628a\u63e1\u667a\u80fd\u4f53\u5728\u591a\u8f6e\u5bf9\u8bdd\u4e2d\u7684\u7ec6\u5fae\u610f\u56fe\u53d8\u5316\uff0c\u4ece\u800c\u5f71\u54cd\u8bc4\u4f30\u7ed3\u679c\u7684\u51c6\u786e\u6027\u3002\u9488\u5bf9\u8fd9\u4e9b\u95ee\u9898\uff0c\u7814\u7a76\u4eba\u5458\u63d0\u51fa\u4e86\u672a\u6765\u6539\u8fdb\u7684\u65b9\u5411\uff0c\u5305\u62ec\u91c7\u7528\u66f4\u5148\u8fdb\u7684\u6a21\u578b\u67b6\u6784\u3001\u4f18\u5316\u63d0\u793a\u8bcd\u8bbe\u8ba1\u4ee5\u5f15\u5bfc\u6a21\u578b\u66f4\u51c6\u786e\u5730\u805a\u7126\u4e8e\u8bc4\u4f30\u8981\u70b9\u3001\u5f15\u5165\u66f4\u591a\u6837\u5316\u7684\u4eba\u7c7b\u53cd\u9988\u6570\u636e\u6765\u6821\u51c6\u6a21\u578b\u7684\u8bc4\u4f30\u6807\u51c6\u7b49\uff0c\u4ee5\u671f\u4e0d\u65ad\u63d0\u9ad8\u8bc4\u4f30\u5de5\u4f5c\u7684\u53ef\u9760\u6027\u548c\u6709\u6548\u6027\u3002<\/p>\n<h2>\u5b9e\u9a8c\u4e0e\u7ed3\u679c\u5206\u6790<\/h2>\n<h4>\u5b9e\u9a8c\u8bbe\u7f6e<\/h4>\n<p>\u5728\u5b9e\u9a8c\u73af\u8282\uff0c\u7814\u7a76\u4eba\u5458\u7cbe\u5fc3\u6311\u9009\u4e86\u4e94\u6b3e\u5177\u6709\u4ee3\u8868\u6027\u7684\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u8fdb\u884c\u8bc4\u4f30\uff0c\u5b83\u4eec\u5206\u522b\u662f Claude Sonnet 3.7\u3001GPT-4o\u3001o3-mini\u3001Deepseek-v3 \u548c Deepseek-R1\u3002\u8fd9\u4e9b\u6a21\u578b\u6db5\u76d6\u4e86\u4e0d\u540c\u7684\u6280\u672f\u8def\u7ebf\u3001\u53c2\u6570\u89c4\u6a21\u548c\u8bad\u7ec3\u7b56\u7565\uff0c\u80fd\u591f\u4e3a\u5b9e\u9a8c\u7ed3\u679c\u63d0\u4f9b\u5e7f\u6cdb\u7684\u4ee3\u8868\u6027\u3002\u5b9e\u9a8c\u6db5\u76d6\u4e86 356 \u4e2a\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u4efb\u52a1\uff0c\u8fd9\u4e9b\u4efb\u52a1\u5728\u5de5\u5177\u7c7b\u578b\u3001\u7528\u6237\u610f\u56fe\u548c\u98ce\u9669\u7c7b\u522b\u7b49\u65b9\u9762\u5b9e\u73b0\u4e86\u5168\u9762\u8986\u76d6\uff0c\u786e\u4fdd\u4e86\u5b9e\u9a8c\u7ed3\u679c\u80fd\u591f\u5145\u5206\u53cd\u6620\u667a\u80fd\u4f53\u5728\u591a\u6837\u5316\u573a\u666f\u4e0b\u7684\u5b89\u5168\u8868\u73b0\u3002<\/p>\n<h4>\u5b9e\u9a8c\u7ed3\u679c<\/h4>\n<p>\u5b9e\u9a8c\u7ed3\u679c\u663e\u793a\uff0c\u5728\u5b89\u5168\u6f0f\u6d1e\u4efb\u52a1\u4e2d\uff0c\u4e0d\u540c\u6a21\u578b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u5dee\u5f02\u663e\u8457\uff0c\u4ece 51.2% \u5230 72.7% \u4e0d\u7b49\u3002\u8fd9\u4e00\u53d1\u73b0\u8868\u660e\uff0c\u5f53\u524d\u667a\u80fd\u4f53\u5728\u9762\u5bf9\u5b89\u5168\u98ce\u9669\u65f6\u7684\u8868\u73b0\u4ecd\u5b58\u5728\u8f83\u5927\u63d0\u5347\u7a7a\u95f4\u3002\u8fdb\u4e00\u6b65\u7684\u5931\u8d25\u7387\u5206\u6790\u53d1\u73b0\uff0c\u4efb\u52a1\u5931\u8d25\u7684\u539f\u56e0\u591a\u79cd\u591a\u6837\uff0c\u5176\u4e2d Web \u5bfc\u822a\u5931\u8d25\u662f\u4e00\u4e2a\u7a81\u51fa\u7684\u95ee\u9898\uff0c\u8fd9\u53ef\u80fd\u662f\u7531\u4e8e\u667a\u80fd\u4f53\u5728\u5904\u7406\u590d\u6742\u7684\u7f51\u9875\u7ed3\u6784\u3001\u52a8\u6001\u5185\u5bb9\u52a0\u8f7d\u4ee5\u53ca\u5e94\u5bf9\u5404\u79cd\u7f51\u7edc\u5f02\u5e38\u60c5\u51b5\u65f6\u7684\u80fd\u529b\u4e0d\u8db3\uff1b\u8ba4\u8bc1\u95ee\u9898\u4e5f\u4e0d\u5bb9\u5ffd\u89c6\uff0c\u667a\u80fd\u4f53\u5728\u9762\u5bf9\u591a\u7528\u6237\u573a\u666f\u4e0b\u7684\u8eab\u4efd\u9a8c\u8bc1\u6311\u6218\u65f6\uff0c\u5e38\u5e38\u56e0\u65e0\u6cd5\u6b63\u786e\u8bc6\u522b\u7528\u6237\u8eab\u4efd\u6216\u5904\u7406\u8ba4\u8bc1\u6d41\u7a0b\u4e2d\u7684\u610f\u5916\u60c5\u51b5\u800c\u5bfc\u81f4\u4efb\u52a1\u4e2d\u65ad\uff1b\u6b64\u5916\uff0c\u5de5\u5177\u8bef\u7528\u73b0\u8c61\u4e5f\u8f83\u4e3a\u5e38\u89c1\uff0c\u4f8b\u5982\u667a\u80fd\u4f53\u5728\u8c03\u7528\u5de5\u5177\u65f6\u53c2\u6570\u8bbe\u7f6e\u4e0d\u5f53\uff0c\u6216\u8005\u5bf9\u5de5\u5177\u7684\u529f\u80fd\u7406\u89e3\u6709\u8bef\uff0c\u4ece\u800c\u65e0\u6cd5\u6709\u6548\u5229\u7528\u5de5\u5177\u5b8c\u6210\u4efb\u52a1\u76ee\u6807\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/919cd33736c371ec5a60974dcfd3836e8c1684.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u5728OPENAGENTSAFETY\u4e2d\uff0c\u57fa\u4e8e\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u667a\u80fd\u4f53\u5728\u5904\u7406\u7528\u6237\u6307\u4ee4\u4e0e\u975e\u73a9\u5bb6\u89d2\u8272\uff08NPC\uff09\u6307\u4ee4\u51b2\u7a81\u65f6\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u53d1\u751f\u7387<\/p>\n<p>\u4e0a\u56fe\u5c55\u793a\u4e86\u4e0d\u540c\u6a21\u578b\u5728\u5904\u7406\u51b2\u7a81\u7528\u6237\u548c NPC \u6307\u4ee4\u65f6\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u3002\u4ece\u56fe\u4e2d\u53ef\u4ee5\u770b\u51fa\uff0co3-mini \u6a21\u578b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u9ad8\u8fbe 72.7%\uff0c\u8fd9\u63d0\u793a\u6211\u4eec\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u82e5\u9009\u7528\u8be5\u6a21\u578b\u4f5c\u4e3a\u667a\u80fd\u4f53\u7684\u57fa\u7840\u6a21\u578b\uff0c\u53ef\u80fd\u9762\u4e34\u8f83\u9ad8\u7684\u5b89\u5168\u98ce\u9669\uff0c\u9700\u8981\u5728\u5b89\u5168\u9632\u62a4\u673a\u5236\u4e0a\u4e0b\u8db3\u529f\u592b\uff0c\u5982\u589e\u52a0\u4efb\u52a1\u6267\u884c\u524d\u7684\u591a\u8f6e\u5ba1\u6838\u73af\u8282\u3001\u5f3a\u5316\u5bf9\u7528\u6237\u548c NPC \u610f\u56fe\u7684\u6df1\u5ea6\u5206\u6790\u7b49\uff0c\u4ee5\u964d\u4f4e\u6f5c\u5728\u98ce\u9669\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/d5c5ff1993810912edd501bf883f7d4640e0ec.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u5728\u5927\u578b\u8bed\u8a00\u6a21\u578b\u4e2d\uff0c\u7528\u6237\u610f\u56fe\u548cNPC\u610f\u56fe\u5bf9\u5e94\u7684Agent\u4e0d\u5b89\u5168\u884c\u4e3a\u53d1\u751f\u7387\uff08%\uff09<\/p>\n<p>\u4e0a\u56fe\u63cf\u7ed8\u4e86\u4e0d\u540c LLM \u6a21\u578b\u5728\u7528\u6237\u548c NPC \u610f\u56fe\u4e0b\u667a\u80fd\u4f53\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u7684\u70ed\u56fe\u3002\u53ef\u4ee5\u89c2\u5bdf\u5230\uff0c\u5728 benign user \u548c malicious NPC \u7684\u610f\u56fe\u7ec4\u5408\u4e0b\uff0c\u591a\u4e2a\u6a21\u578b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u5904\u4e8e\u8f83\u9ad8\u6c34\u5e73\uff0c\u5982 Claude Sonnet 3.7 \u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u63a5\u8fd1 50%\u3002\u8fd9\u8868\u660e\u5373\u4f7f\u7528\u6237\u672c\u8eab\u6ca1\u6709\u6076\u610f\uff0c\u4f46\u5f53 NPC \u5b58\u5728\u6076\u610f\u610f\u56fe\u65f6\uff0c\u667a\u80fd\u4f53\u4ecd\u7136\u5bb9\u6613\u53d7\u5230\u8bf1\u5bfc\uff0c\u6267\u884c\u4e0d\u5b89\u5168\u884c\u4e3a\u3002\u8fd9\u63d0\u793a\u6211\u4eec\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u8981\u6ce8\u91cd\u5bf9 NPC \u610f\u56fe\u7684\u8bc6\u522b\u548c\u7ba1\u7406\uff0c\u9632\u8303\u667a\u80fd\u4f53\u53d7\u5916\u90e8\u6076\u610f\u56e0\u7d20\u5e72\u6270\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/c147e7735b3b83e5848085f790469fca3f2e66.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u4e0d\u540c\u5927\u8bed\u8a00\u6a21\u578b\u4e2d\u5de5\u5177\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\uff08%\uff09<\/p>\n<p>\u4e0a\u56fe\u5448\u73b0\u4e86\u4e0d\u540c\u5de5\u5177\u548c LLM \u6a21\u578b\u4e0b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u70ed\u56fe\u3002\u53ef\u4ee5\u770b\u51fa\uff0cWeb \u6d4f\u89c8\u5de5\u5177\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u666e\u904d\u8f83\u9ad8\uff0c\u5982\u5728 Web \u6d4f\u89c8\u5de5\u5177\u4f7f\u7528\u573a\u666f\u4e2d\uff0cGPT-4o \u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u7387\u63a5\u8fd1 60%\u3002\u8fd9\u63ed\u793a\u4e86 Web \u6d4f\u89c8\u5de5\u5177\u56e0\u6d89\u53ca\u590d\u6742\u7684\u7f51\u7edc\u4ea4\u4e92\u548c\u52a8\u6001\u5185\u5bb9\uff0c\u5bb9\u6613\u5f15\u53d1\u5b89\u5168\u95ee\u9898\u3002\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u5e94\u52a0\u5f3a\u5bf9\u667a\u80fd\u4f53\u4f7f\u7528 Web \u6d4f\u89c8\u5de5\u5177\u7684\u76d1\u7ba1\uff0c\u5982\u9650\u5236\u8bbf\u95ee\u654f\u611f\u7f51\u7ad9\u3001\u589e\u52a0\u7f51\u9875\u5185\u5bb9\u8fc7\u6ee4\u673a\u5236\u7b49\uff0c\u4ee5\u964d\u4f4e\u5b89\u5168\u98ce\u9669\u3002<\/p>\n<h2>\u8bbe\u8ba1\u542f\u793a<\/h2>\n<p>OA-SAFETY \u6846\u67b6\u4e3a\u667a\u80fd\u4f53\u7684\u5b89\u5168\u8bbe\u8ba1\u5e26\u6765\u4e86\u6df1\u523b\u7684\u8bbe\u8ba1\u542f\u793a\u3002<\/p>\n<p>\u9996\u5148\uff0c\u4e0a\u4e0b\u6587\u610f\u56fe\u805a\u5408\u80fd\u529b\u81f3\u5173\u91cd\u8981\u3002\u667a\u80fd\u4f53\u9700\u8981\u5728\u591a\u8f6e\u4ea4\u4e92\u8fc7\u7a0b\u4e2d\uff0c\u6301\u7eed\u6536\u96c6\u548c\u6574\u5408\u7528\u6237\u4ee5\u53ca NPC \u7684\u610f\u56fe\u4fe1\u606f\uff0c\u901a\u8fc7\u5206\u6790\u610f\u56fe\u7684\u53d8\u5316\u8d8b\u52bf\u3001\u610f\u56fe\u4e4b\u95f4\u7684\u5173\u8054\u6027\u4ee5\u53ca\u6f5c\u5728\u7684\u610f\u56fe\u51b2\u7a81\uff0c\u6316\u6398\u51fa\u9690\u85cf\u5728\u8868\u9762\u4efb\u52a1\u4e4b\u4e0b\u7684\u6df1\u5c42\u98ce\u9669\u3002\u4f8b\u5982\uff0c\u5728\u957f\u671f\u7684\u5546\u52a1\u8c08\u5224\u573a\u666f\u4e2d\uff0c\u667a\u80fd\u4f53\u4e0d\u4ec5\u8981\u5173\u6ce8\u5f53\u524d\u7684\u8c08\u5224\u8981\u70b9\uff0c\u8fd8\u8981\u56de\u6eaf\u4e4b\u524d\u7684\u5bf9\u8bdd\u5185\u5bb9\uff0c\u5206\u6790\u5bf9\u65b9\u610f\u56fe\u7684\u8fde\u8d2f\u6027\u548c\u4e00\u81f4\u6027\uff0c\u4ece\u800c\u53ca\u65f6\u53d1\u73b0\u53ef\u80fd\u5b58\u5728\u7684\u6b3a\u8bc8\u6216\u8bf1\u5bfc\u884c\u4e3a\u3002<\/p>\n<p>\u5176\u6b21\uff0c\u4e3a\u9ad8\u98ce\u9669\u5de5\u5177\u8bbe\u5b9a\u7279\u5b9a\u7684\u6743\u9650\u8fb9\u754c\u662f\u4fdd\u969c\u5b89\u5168\u6027\u7684\u5173\u952e\u4e3e\u63aa\u3002\u5bf9\u4e8e\u6587\u4ef6\u64cd\u4f5c\u5de5\u5177\uff0c\u53ef\u4ee5\u9650\u5236\u5176\u5bf9\u7cfb\u7edf\u5173\u952e\u76ee\u5f55\u7684\u8bbf\u95ee\u6743\u9650\uff1b\u5728\u4ee3\u7801\u6267\u884c\u73af\u5883\u4e2d\uff0c\u4e25\u683c\u89c4\u5b9a\u5141\u8bb8\u6267\u884c\u7684\u4ee3\u7801\u7c7b\u578b\u548c\u64cd\u4f5c\u8303\u56f4\uff0c\u901a\u8fc7\u7cbe\u7ec6\u7684\u6743\u9650\u7ba1\u63a7\uff0c\u5c06\u6f5c\u5728\u7684\u5b89\u5168\u5a01\u80c1\u62d2\u4e4b\u95e8\u5916\u3002<\/p>\n<p>\u6700\u540e\uff0c\u57fa\u4e8e\u653f\u7b56\u7684\u76d1\u7763\u673a\u5236\u80fd\u591f\u4e3a\u667a\u80fd\u4f53\u7684\u884c\u4e3a\u63d0\u4f9b\u660e\u786e\u7684\u5408\u89c4\u6307\u5f15\u3002\u901a\u8fc7\u4f7f\u7528\u7b26\u5408\u6cd5\u5f8b\u3001\u7ec4\u7ec7\u89c4\u8303\u548c\u884c\u4e1a\u6807\u51c6\u7684\u6570\u636e\u96c6\u5bf9\u667a\u80fd\u4f53\u8fdb\u884c\u8bad\u7ec3\uff0c\u4f7f\u5176\u5728\u590d\u6742\u591a\u53d8\u7684\u4efb\u52a1\u573a\u666f\u4e2d\uff0c\u59cb\u7ec8\u9075\u5faa\u6b63\u786e\u7684\u4ef7\u503c\u5bfc\u5411\u548c\u884c\u4e3a\u51c6\u5219\uff0c\u786e\u4fdd\u5176\u51b3\u7b56\u548c\u884c\u52a8\u7b26\u5408\u76f8\u5173\u653f\u7b56\u8981\u6c42\u3002<\/p>\n<p>OA-SAFETY \u6846\u67b6\u672a\u6765\u53ef\u4ee5\u4ece\u591a\u4e2a\u65b9\u9762\u8fdb\u884c\u6301\u7eed\u6539\u8fdb\u548c\u62d3\u5c55\u3002\u5728\u63d0\u5347 LLM-as-Judge \u7684\u53ef\u9760\u6027\u65b9\u9762\uff0c\u53ef\u63a2\u7d22\u5229\u7528\u66f4\u5148\u8fdb\u7684\u6a21\u578b\u67b6\u6784\uff0c\u7ed3\u5408\u8fc1\u79fb\u5b66\u4e60\u6280\u672f\uff0c\u5c06\u5176\u4ed6\u9886\u57df\u7ecf\u8fc7\u9a8c\u8bc1\u7684\u4f18\u8d28\u6a21\u578b\u53c2\u6570\u8fc1\u79fb\u5230\u8bc4\u4f30\u6a21\u578b\u4e2d\uff0c\u4ee5\u589e\u5f3a\u5176\u8bc4\u4f30\u80fd\u529b\u3002\u540c\u65f6\uff0c\u4f18\u5316\u63d0\u793a\u8bcd\u8bbe\u8ba1\uff0c\u901a\u8fc7\u5927\u91cf\u7684\u5b9e\u9a8c\u548c\u7528\u6237\u53cd\u9988\uff0c\u901a\u8fc7\u6253\u78e8\u63d0\u793a\u8bcd\u7684\u8868\u8ff0\u65b9\u5f0f\u548c\u5f15\u5bfc\u903b\u8f91\uff0c\u4f7f\u5176\u80fd\u591f\u66f4\u7cbe\u51c6\u5730\u6fc0\u53d1\u6a21\u578b\u5bf9\u667a\u80fd\u4f53\u5b89\u5168\u884c\u4e3a\u7684\u8bc4\u4f30\u6f5c\u529b\u3002<\/p>\n<h2>\u6846\u67b6\u5c40\u9650\u6027<\/h2>\n<h4>\u6a21\u62df\u73af\u5883\u4e0e\u771f\u5b9e\u73af\u5883\u7684\u5dee\u5f02<\/h4>\n<p>\u5c3d\u7ba1 OA-SAFETY \u6846\u67b6\u5728\u6a21\u62df\u771f\u5b9e\u73af\u5883\u65b9\u9762\u505a\u51fa\u4e86\u8bf8\u591a\u52aa\u529b\uff0c\u4f46\u4ecd\u7136\u5b58\u5728\u4e00\u5b9a\u7684\u5dee\u8ddd\u3002\u6a21\u62df\u73af\u5883\u4e2d\u7684\u7f51\u7edc\u6761\u4ef6\u76f8\u5bf9\u7a33\u5b9a\uff0c\u800c\u73b0\u5b9e\u4e16\u754c\u4e2d\u7f51\u7edc\u5ef6\u8fdf\u3001\u4e22\u5305\u7b49\u95ee\u9898\u53ef\u80fd\u4e25\u91cd\u5f71\u54cd\u667a\u80fd\u4f53\u4e0e\u5de5\u5177\u7684\u4ea4\u4e92\u6548\u679c\uff0c\u5bfc\u81f4\u4efb\u52a1\u6267\u884c\u5931\u8d25\u6216\u51fa\u73b0\u4e0d\u53ef\u9884\u89c1\u7684\u5b89\u5168\u98ce\u9669\u3002\u786c\u4ef6\u6545\u969c\u4e5f\u662f\u771f\u5b9e\u73af\u5883\u4e2d\u5e38\u89c1\u7684\u5e72\u6270\u56e0\u7d20\uff0c\u4f8b\u5982\u670d\u52a1\u5668\u7a81\u7136\u5b95\u673a\u3001\u5b58\u50a8\u8bbe\u5907\u635f\u574f\u7b49\uff0c\u8fd9\u4e9b\u60c5\u51b5\u5728\u6a21\u62df\u73af\u5883\u4e2d\u96be\u4ee5\u5b8c\u5168\u590d\u73b0\uff0c\u4ece\u800c\u53ef\u80fd\u4f7f\u5f97\u667a\u80fd\u4f53\u5728\u5b9e\u9645\u90e8\u7f72\u65f6\u9762\u5bf9\u6b64\u7c7b\u95ee\u9898\u65f6\u7f3a\u4e4f\u6709\u6548\u7684\u5e94\u5bf9\u7b56\u7565\u3002<\/p>\n<h4>\u4efb\u52a1\u8986\u76d6\u8303\u56f4\u7684\u5c40\u9650\u6027<\/h4>\n<p>\u76ee\u524d\u6846\u67b6\u4e2d\u7684\u4efb\u52a1\u4e3b\u8981\u96c6\u4e2d\u5728\u4e00\u4e9b\u5e38\u89c1\u7684\u5e94\u7528\u573a\u666f\u548c\u98ce\u9669\u7c7b\u522b\u4e0a\uff0c\u7136\u800c\uff0cAI \u667a\u80fd\u4f53\u7684\u5b9e\u9645\u5e94\u7528\u9886\u57df\u6781\u5176\u5e7f\u6cdb\uff0c\u6db5\u76d6\u4e86\u4ece\u533b\u7597\u5065\u5eb7\u3001\u91d1\u878d\u98ce\u63a7\u5230\u667a\u80fd\u4ea4\u901a\u3001\u5de5\u4e1a\u5236\u9020\u7b49\u591a\u4e2a\u5173\u952e\u884c\u4e1a\u3002\u6bcf\u4e2a\u884c\u4e1a\u90fd\u6709\u5176\u72ec\u7279\u7684\u4e1a\u52a1\u6d41\u7a0b\u3001\u6570\u636e\u7279\u70b9\u548c\u5b89\u5168\u9700\u6c42\uff0c\u73b0\u6709\u4efb\u52a1\u5728\u8986\u76d6\u8fd9\u4e9b\u591a\u6837\u5316\u9886\u57df\u548c\u573a\u666f\u65b9\u9762\u8fd8\u5b58\u5728\u4e0d\u8db3\uff0c\u65e0\u6cd5\u5b8c\u5168\u4ee3\u8868\u6240\u6709\u53ef\u80fd\u7684 AI \u667a\u80fd\u4f53\u5e94\u7528\u60c5\u5883\uff0c\u8fd9\u5728\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u9650\u5236\u4e86\u6846\u67b6\u5bf9\u667a\u80fd\u4f53\u6cdb\u5316\u5b89\u5168\u6027\u80fd\u7684\u8bc4\u4f30\u80fd\u529b\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202507\/14\/9179ff607a97b1ee751018fefb181743cdc867.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>GPT-4o\u4e2d\u5931\u8d25\u4efb\u52a1\u5728\u4e0d\u540c\u5de5\u5177\u95f4\u7684\u5206\u5e03\u60c5\u51b5<\/p>\n<p>\u4e0a\u56fe\u663e\u793a\u4e86 GPT-4o \u6a21\u578b\u5728\u5931\u8d25\u4efb\u52a1\u4e2d\u4e0d\u540c\u5de5\u5177\u7684\u5206\u5e03\u60c5\u51b5\u3002\u53ef\u4ee5\u770b\u51fa\uff0cWeb \u6d4f\u89c8\u548c\u6587\u4ef6\u7cfb\u7edf\u64cd\u4f5c\u76f8\u5173\u7684\u4efb\u52a1\u5931\u8d25\u7387\u8f83\u9ad8\uff0c\u5206\u522b\u7ea6\u4e3a 30% \u548c 25%\u3002\u8fd9\u63d0\u793a\u6211\u4eec\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u82e5\u4f7f\u7528 GPT-4o \u6a21\u578b\u9a71\u52a8\u7684\u667a\u80fd\u4f53\uff0c\u5e94\u7279\u522b\u5173\u6ce8\u5176\u5728\u8fdb\u884c Web \u6d4f\u89c8\u548c\u6587\u4ef6\u64cd\u4f5c\u65f6\u7684\u4efb\u52a1\u6267\u884c\u60c5\u51b5\uff0c\u5206\u6790\u4efb\u52a1\u5931\u8d25\u539f\u56e0\uff0c\u5982\u662f\u5426\u56e0\u7f51\u7edc\u95ee\u9898\u3001\u6587\u4ef6\u6743\u9650\u8bbe\u7f6e\u4e0d\u5f53\u7b49\uff0c\u9488\u5bf9\u6027\u5730\u4f18\u5316\u6a21\u578b\u548c\u4efb\u52a1\u6d41\u7a0b\uff0c\u63d0\u9ad8\u4efb\u52a1\u6210\u529f\u7387\u3002<\/p>\n<h2>\u603b\u7ed3<\/h2>\n<p>\u672c\u6587\u4ecb\u7ecd\u63a2\u8ba8\u4e86OpenAgentSafety\u6846\u67b6\u3002\u5b83\u5177\u6709\u4ee5\u4e0b\u7279\u70b9\uff1a<\/p>\n<ul data-id=\"u738a58b-jAYD366i\">\n<li data-id=\"ld70c578-JCKTNbC9\"><strong>\u771f\u5b9e\u5de5\u5177\u4ea4\u4e92<\/strong>\uff1aAgent \u53ef\u4ee5\u4e0e\u5b9e\u9645\u7684\u6587\u4ef6\u7cfb\u7edf\u3001\u547d\u4ee4\u884c\u3001\u4ee3\u7801\u6267\u884c\u73af\u5883\u548c\u7f51\u9875\u754c\u9762\u8fdb\u884c\u4ea4\u4e92\u3002<\/li>\n<li data-id=\"ld70c578-n20YRWlR\"><strong>\u591a\u6837\u5316\u7528\u6237\u610f\u56fe<\/strong>\uff1a\u4efb\u52a1\u6a21\u62df\u4e86\u4ece\u826f\u6027\u6a21\u7cca\u5230\u64cd\u7eb5\u6027\u7684\u7528\u6237\u884c\u4e3a\u3002<\/li>\n<li data-id=\"ld70c578-baQa7Soc\"><strong>\u591a\u8f6e\u3001\u591a\u7528\u6237\u4ea4\u4e92<\/strong>\uff1a\u573a\u666f\u5305\u62ec\u6d89\u53ca\u7528\u6237\u548c\u5177\u6709\u51b2\u7a81\u76ee\u6807\u7684\u6b21\u8981\u89d2\u8272\uff08NPC\uff09\u7684\u6269\u5c55\u4ea4\u4e92\u3002<\/li>\n<li data-id=\"ld70c578-4pJijMfn\"><strong>\u53ef\u6269\u5c55\u6027<\/strong>\uff1a\u7814\u7a76\u4eba\u5458\u53ef\u4ee5\u8f7b\u677e\u6dfb\u52a0\u65b0\u4efb\u52a1\u3001\u6a21\u62df\u73af\u5883\u3001\u590d\u6742\u7684\u793e\u4f1a\u52a8\u6001\u548c\u5b9a\u5236\u8bc4\u4f30\u5668\u3002<\/li>\n<\/ul>\n<p>\u901a\u8fc7\u5bf9\u667a\u80fd\u4f53\u5728\u771f\u5b9e\u5de5\u5177\u73af\u5883\u4e2d\u7684\u884c\u4e3a\u8fdb\u884c\u6a21\u62df\u548c\u8bc4\u4f30\uff0c\u8be5\u6846\u67b6\u5c55\u73b0\u4e86\u73b0\u6709AI\u7cfb\u7edf\u5728\u5b89\u5168\u6027\u65b9\u9762\u7684\u663e\u8457\u4e0d\u8db3\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u5373\u4f7f\u662f\u5f53\u524d\u6700\u5148\u8fdb\u7684\u667a\u80fd\u4f53\uff0c\u5728\u9762\u5bf9\u5b89\u5168\u6f0f\u6d1e\u4efb\u52a1\u65f6\uff0c\u4e5f\u6709\u76f8\u5f53\u6bd4\u4f8b\u7684\u4e0d\u5b89\u5168\u884c\u4e3a\u53d1\u751f\uff0c\u8fd9\u4e00\u53d1\u73b0\u4ee4\u4eba\u8b66\u60d5\u3002<\/p>\n<p>\u5728\u591a\u8f6e\u3001\u591a\u7528\u6237\u4ea4\u4e92\u573a\u666f\u4e0b\u7684\u8bc4\u4f30\u5c55\u73b0\u4e86\u667a\u80fd\u4f53\u5728\u590d\u6742\u793e\u4ea4\u73af\u5883\u4e2d\u7684\u8106\u5f31\u6027\u3002\u667a\u80fd\u4f53\u9700\u8981\u5177\u5907\u7cbe\u51c6\u7684\u4e0a\u4e0b\u6587\u610f\u56fe\u805a\u5408\u80fd\u529b\uff0c\u4ee5\u5e94\u5bf9\u591a\u8f6e\u5bf9\u8bdd\u4e2d\u610f\u56fe\u7684\u5fae\u5999\u53d8\u5316\u548c\u6f5c\u5728\u7684\u8bf1\u5bfc\u98ce\u9669\u3002\u540c\u65f6\uff0c\u4e3a\u9ad8\u98ce\u9669\u5de5\u5177\u8bbe\u5b9a\u6743\u9650\u8fb9\u754c\u548c\u57fa\u4e8e\u653f\u7b56\u7684\u76d1\u7763\u673a\u5236\uff0c\u80fd\u591f\u6709\u6548\u964d\u4f4e\u667a\u80fd\u4f53\u5728\u6267\u884c\u4efb\u52a1\u65f6\u7684\u5b89\u5168\u9690\u60a3\u3002\u6846\u67b6\u7684\u8bbe\u8ba1\u542f\u793a\u5f3a\u8c03\u4e86\u8fd9\u4e9b\u5173\u952e\u7684\u5b89\u5168\u7279\u6027\uff0c\u4e3a\u6784\u5efa\u66f4\u5b89\u5168\u7684AI\u667a\u80fd\u4f53\u63d0\u4f9b\u4e86\u5b9d\u8d35\u7684\u6307\u5bfc\u3002<\/p>\n<p>\u6587\u7ae0\u6765\u81ea\uff1a51CTO<\/p>\n<\/div>\n<\/div>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_26756\" class=\"pvc_stats total_only  \" data-element-id=\"26756\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"excerpt":{"rendered":"<p>AI \u667a\u80fd\u4f53\u5b89\u5168\u8bc4\u4f30\u8fce\u6765\u65b0\u5de5\u5177\uff0cOpenAgentSafety \u6846\u67b6\u4ee5\u771f\u5b9e\u5de5\u5177\u4ea4\u4e92\u3001\u590d\u6742\u573a\u666f\u6a21\u62df\u53ca\u6df7\u5408\u8bc4\u4f30\u7b56\u7565 [&hellip;]<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_26756\" class=\"pvc_stats total_only  \" data-element-id=\"26756\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"author":56,"featured_media":26758,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[23,20,80],"tags":[1814],"class_list":["post-26756","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-23","category-20","category-80","tag-openagentsafety-ai-"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/26756","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=26756"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/26756\/revisions"}],"predecessor-version":[{"id":26759,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/26756\/revisions\/26759"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/26758"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=26756"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=26756"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=26756"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}