{"id":394,"date":"2025-01-04T18:04:21","date_gmt":"2025-01-04T10:04:21","guid":{"rendered":"https:\/\/crackempire.cn\/?p=394"},"modified":"2025-01-04T18:04:27","modified_gmt":"2025-01-04T10:04:27","slug":"%e6%96%b0%e5%9e%8b%e4%ba%ba%e5%b7%a5%e6%99%ba%e8%83%bd%e8%b6%8a%e7%8b%b1%e6%96%b9%e6%b3%95-%e7%b3%9f%e7%b3%95%e7%9a%84%e6%9d%8e%e5%85%8b%e7%89%b9%e6%b3%95%e5%ae%98-%e5%b0%86%e6%94%bb%e5%87%bb","status":"publish","type":"post","link":"https:\/\/crackempire.cn\/?p=394","title":{"rendered":"\u65b0\u578b\u4eba\u5de5\u667a\u80fd\u8d8a\u72f1\u65b9\u6cd5 “\u7cdf\u7cd5\u7684\u674e\u514b\u7279\u6cd5\u5b98 “\u5c06\u653b\u51fb\u6210\u529f\u7387\u63d0\u9ad8\u4e86 60% \u4ee5\u4e0a"},"content":{"rendered":"\n
\"\"<\/figure>\n\n\n\n

\u7f51\u7edc\u5b89\u5168\u7814\u7a76\u4eba\u5458\u63ed\u793a\u4e86\u4e00\u79cd\u65b0\u7684\u8d8a\u72f1\u6280\u672f\uff0c\u8fd9\u79cd\u6280\u672f\u53ef\u7528\u4e8e\u7a81\u7834\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u5b89\u5168\u9632\u62a4\u7f51\uff0c\u5e76\u4ea7\u751f\u6f5c\u5728\u7684\u6709\u5bb3\u6216\u6076\u610f\u54cd\u5e94\u3002 Palo Alto Networks \u7b2c 42 \u5355\u5143\u7684\u7814\u7a76\u4eba\u5458\u9ec4\u6c38\u54f2\u3001\u5b63\u65f8\u3001\u80e1\u6587\u4fca\u3001\u9648\u6770\u3001Akshata Rao \u548c Danny Tsechansky \u5c06\u8fd9\u79cd\u591a\u8f6e\uff08\u53c8\u79f0\u591a\u67aa\uff09\u653b\u51fb\u7b56\u7565\u547d\u540d\u4e3a “\u7cdf\u7cd5\u7684\u674e\u514b\u7279\u6cd5\u5b98”\uff08Bad Likert Judge\uff09\u3002<\/p>\n\n\n\n

“42 \u5c0f\u7ec4\u8bf4\uff1a”\u8be5\u6280\u672f\u8981\u6c42\u76ee\u6807 LLM \u5145\u5f53\u6cd5\u5b98\uff0c\u4f7f\u7528\u674e\u514b\u7279\u91cf\u8868\u5bf9\u7ed9\u5b9a\u54cd\u5e94\u7684\u6709\u5bb3\u6027\u8fdb\u884c\u8bc4\u5206\u3002<\/p>\n\n\n\n

“\u7136\u540e\uff0c\u5b83\u8981\u6c42 LLM \u751f\u6210\u5305\u542b\u4e0e\u91cf\u8868\u4e00\u81f4\u7684\u793a\u4f8b\u7684\u54cd\u5e94\u3002 \u8fd1\u5e74\u6765\uff0c\u4eba\u5de5\u667a\u80fd\u7684\u666e\u53ca\u4e5f\u5bfc\u81f4\u4e86\u4e00\u79cd\u65b0\u7684\u5b89\u5168\u6f0f\u6d1e–“\u63d0\u793a\u6ce8\u5165”\uff08prompt injection\uff09\uff0c\u5176\u76ee\u7684\u662f\u901a\u8fc7\u4f20\u9012\u7279\u5236\u6307\u4ee4\uff08\u5373\u63d0\u793a\uff09\uff0c\u4f7f\u673a\u5668\u5b66\u4e60\u6a21\u578b\u5ffd\u7565\u5176\u9884\u671f\u884c\u4e3a\u3001 \u5176\u4e2d\u4e00\u79cd\u7279\u6b8a\u7684\u63d0\u793a\u6ce8\u5165\u653b\u51fb\u65b9\u6cd5\u88ab\u79f0\u4e3a “\u591a\u6b21\u8d8a\u72f1”\uff08many-shot jailbreaking\uff09\uff0c\u5b83\u5229\u7528 LLM \u7684\u957f\u4e0a\u4e0b\u6587\u7a97\u53e3\u548c\u6ce8\u610f\u529b\u6765\u5236\u4f5c\u4e00\u7cfb\u5217\u63d0\u793a\uff0c\u5728\u4e0d\u89e6\u53d1\u5176\u5185\u90e8\u4fdd\u62a4\u63aa\u65bd\u7684\u60c5\u51b5\u4e0b\uff0c\u9010\u6b65\u523a\u6fc0 LLM \u4ea7\u751f\u6076\u610f\u54cd\u5e94\u3002 \u8fd9\u79cd\u6280\u672f\u7684\u4e00\u4e9b\u4f8b\u5b50\u5305\u62ec “\u6e10\u5165\u4f73\u5883”\uff08Crescendo\uff09\u548c “\u6b3a\u9a97\u6027\u559c\u60a6”\uff08Deceptive Delight\uff09\u3002 \u7b2c 42 \u5c0f\u7ec4\u5c55\u793a\u7684\u6700\u65b0\u65b9\u6cd5\u662f\u5c06 LLM \u4f5c\u4e3a\u6cd5\u5b98\uff0c\u4f7f\u7528\u674e\u514b\u7279\u5fc3\u7406\u6d4b\u91cf\u91cf\u8868\u8bc4\u4f30\u7ed9\u5b9a\u54cd\u5e94\u7684\u6709\u5bb3\u6027\uff0c\u7136\u540e\u8981\u6c42\u6a21\u578b\u63d0\u4f9b\u4e0e\u4e0d\u540c\u5206\u6570\u76f8\u5bf9\u5e94\u7684\u4e0d\u540c\u54cd\u5e94\u3002<\/p>\n\n\n\n

\u5728\u5bf9\u4e9a\u9a6c\u900a\u7f51\u7edc\u670d\u52a1\u3001\u8c37\u6b4c\u3001Meta\u3001\u5fae\u8f6f\u3001OpenAI \u548c\u82f1\u4f1f\u8fbe\u516c\u53f8\u7684\u516d\u79cd\u6700\u5148\u8fdb\u7684\u6587\u672c\u751f\u6210 LLM \u8fdb\u884c\u7684\u5404\u79cd\u7c7b\u522b\u7684\u6d4b\u8bd5\u4e2d\u53d1\u73b0\uff0c\u4e0e\u666e\u901a\u653b\u51fb\u63d0\u793a\u76f8\u6bd4\uff0c\u8be5\u6280\u672f\u53ef\u5c06\u653b\u51fb\u6210\u529f\u7387\uff08ASR\uff09\u5e73\u5747\u63d0\u9ad8 60% \u4ee5\u4e0a\u3002<\/p>\n\n\n\n

\u8fd9\u4e9b\u7c7b\u522b\u5305\u62ec\u4ec7\u6068\u3001\u9a9a\u6270\u3001\u81ea\u6b8b\u3001\u6027\u5185\u5bb9\u3001\u6ee5\u6740\u6b66\u5668\u3001\u975e\u6cd5\u6d3b\u52a8\u3001\u6076\u610f\u8f6f\u4ef6\u751f\u6210\u548c\u7cfb\u7edf\u63d0\u793a\u6cc4\u6f0f\u3002”\u901a\u8fc7\u5229\u7528 LLM \u5bf9\u6709\u5bb3\u5185\u5bb9\u7684\u7406\u89e3\u53ca\u5176\u8bc4\u4f30\u54cd\u5e94\u7684\u80fd\u529b\uff0c\u8fd9\u79cd\u6280\u672f\u53ef\u4ee5\u663e\u8457\u63d0\u9ad8\u6210\u529f\u7ed5\u8fc7\u6a21\u578b\u5b89\u5168\u9632\u62a4\u7684\u51e0\u7387\uff0c”\u7814\u7a76\u4eba\u5458\u8bf4\uff0c”\u7ed3\u679c\u8868\u660e\uff0c\u5728\u6240\u6709\u6d4b\u8bd5\u6a21\u578b\u4e2d\uff0c\u5185\u5bb9\u8fc7\u6ee4\u5668\u53ef\u4ee5\u5c06 ASR \u5e73\u5747\u964d\u4f4e 89.2 \u4e2a\u767e\u5206\u70b9\u3002 \u8fd9\u8868\u660e\uff0c\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u90e8\u7f72 LLM \u65f6\uff0c\u5b9e\u65bd\u5168\u9762\u7684\u5185\u5bb9\u8fc7\u6ee4\u4f5c\u4e3a\u6700\u4f73\u5b9e\u8df5\u5177\u6709\u81f3\u5173\u91cd\u8981\u7684\u4f5c\u7528\u3002<\/p>\n\n\n\n

\u82f1\u56fd\u300a\u536b\u62a5\u300b\u5728\u51e0\u5929\u524d\u7684\u4e00\u7bc7\u62a5\u9053\u4e2d\u62ab\u9732\uff0cOpenAI \u7684 ChatGPT \u641c\u7d22\u5de5\u5177\u53ef\u4ee5\u901a\u8fc7\u8981\u6c42\u5176\u5bf9\u5305\u542b\u9690\u85cf\u5185\u5bb9\u7684\u7f51\u9875\u8fdb\u884c\u603b\u7ed3\u6765\u751f\u6210\u5b8c\u5168\u8bef\u5bfc\u6027\u7684\u603b\u7ed3\u3002”\u8fd9\u4e9b\u6280\u672f\u53ef\u4ee5\u88ab\u6076\u610f\u4f7f\u7528\uff0c\u4f8b\u5982\uff0c\u5c3d\u7ba1\u540c\u4e00\u9875\u9762\u4e0a\u6709\u8d1f\u9762\u8bc4\u8bba\uff0c\u4f46 ChatGPT \u4ecd\u4f1a\u8fd4\u56de\u5bf9\u4ea7\u54c1\u7684\u6b63\u9762\u8bc4\u4ef7\u3002\u82f1\u56fd\u62a5\u7eb8\u8bf4\uff1a”\u7b2c\u4e09\u65b9\u5728\u6ca1\u6709\u8bf4\u660e\u7684\u60c5\u51b5\u4e0b\u7b80\u5355\u5730\u52a0\u5165\u9690\u85cf\u6587\u672c\uff0c\u4e5f\u53ef\u4ee5\u7528\u6765\u786e\u4fdd\u6b63\u9762\u8bc4\u4ef7\uff0c\u5176\u4e2d\u4e00\u4e2a\u6d4b\u8bd5\u5305\u62ec\u6781\u5176\u6b63\u9762\u7684\u865a\u5047\u8bc4\u8bba\uff0c\u8fd9\u5f71\u54cd\u4e86 ChatGPT \u8fd4\u56de\u7684\u6458\u8981”\u3002<\/p>\n\n\n\n

\u4fe1\u606f\u6765\u6e90\uff1ahttps:\/\/thehackernews.com\/2025\/01\/new-ai-jailbreak-method-bad-likert.html\t<\/p>\n\n\n

AI<\/a>\nAPI \u5b89\u5168<\/a>\nDMARC<\/a>\n\u4e91\u5b89\u5168<\/a>\n\u4e94\u89d2\u5927\u697c<\/a>\n\u4eba\u5de5\u667a\u80fd<\/a>\n\u4eba\u7c7b\u5fc3\u7406\u5b66<\/a>\n\u5173\u7a0e<\/a>\n\u52a0\u5bc6\u8d27\u5e01<\/a>\n\u52d2\u7d22\u8f6f\u4ef6<\/a>\n\u54c1\u724c\u5192\u5145<\/a>\n\u56fd\u5bb6\u5b89\u5168<\/a>\n\u5a01\u80c1\u60c5\u62a5<\/a>\n\u5e94\u7528\u7a0b\u5e8f\u5b89\u5168<\/a>\n\u6076\u610f\u8f6f\u4ef6<\/a>\n\u6570\u5b57\u8d27\u5e01<\/a>\n\u65f6\u4e8b\u653f\u6cbb<\/a>\n\u65f6\u653f<\/a>\n\u65f6\u653f\u65b0\u95fb<\/a>\n\u6697\u7f51<\/a>\n\u6728\u9a6c\u75c5\u6bd2<\/a>\n\u6b3a\u8bc8\u7ba1\u7406\u548c\u7f51\u7edc\u72af\u7f6a<\/a>\n\u6c11\u65cf\u56fd\u5bb6\u653b\u51fb<\/a>\n\u6f0f\u6d1e\u548c\u5a01\u80c1<\/a>\n\u7279\u6717\u666e\u653f\u5e9c<\/a>\n\u7279\u6717\u666e\u7684 2026 \u8d22\u5e74\u9884\u7b97<\/a>\n\u793e\u4f1a\u5de5\u7a0b\u5b66<\/a>\n\u793e\u4f1a\u5de5\u7a0b\u5b66\u3001\u7f51\u7edc\u6218<\/a>\n\u79fb\u52a8\u5b89\u5168<\/a>\n\u7aef\u70b9\u5b89\u5168<\/a>\n\u7f51\u7edc\u5371\u9669<\/a>\n\u7f51\u7edc\u5b89\u5168<\/a>\n\u7f51\u7edc\u5b89\u5168\u8fd0\u8425<\/a>\n\u7f51\u7edc\u653b\u51fb\u548c\u6570\u636e\u6cc4\u9732<\/a>\n\u7f51\u7edc\u72af\u7f6a<\/a>\n\u7f51\u7edc\u7f6a\u72af<\/a>\n\u7f51\u7edc\u9493\u9c7c<\/a>\n\u7f51\u7edc\u95f4\u8c0d<\/a>\n\u7f51\u7edc\u98ce\u9669<\/a>\n\u7f8e\u56fd\u8d38\u6613\u4ee3\u8868\u529e\u516c\u5ba4<\/a>\n\u8054\u90a6\u8d38\u6613\u59d4\u5458\u4f1a<\/a>\n\u8106\u5f31\u6027<\/a>\n\u8fdc\u7a0b\u5458\u5de5<\/a>\n\u91d1\u878d\u4e0e\u94f6\u884c<\/a>\n\u95f4\u8c0d\u8f6f\u4ef6<\/a><\/p>","protected":false},"excerpt":{"rendered":"

\u7f51\u7edc\u5b89\u5168\u7814\u7a76\u4eba\u5458\u63ed\u793a\u4e86\u4e00\u79cd\u65b0\u7684\u8d8a\u72f1\u6280\u672f\uff0c\u8fd9\u79cd\u6280\u672f\u53ef\u7528\u4e8e\u7a81\u7834\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u5b89\u5168\u9632\u62a4\u7f51\uff0c\u5e76\u4ea7\u751f\u6f5c\u5728\u7684\u6709\u5bb3…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[26,31,25,24],"class_list":["post-394","post","type-post","status-publish","format-standard","hentry","category-cybernews","tag-26","tag-31","tag-25","tag-24"],"_links":{"self":[{"href":"https:\/\/crackempire.cn\/index.php?rest_route=\/wp\/v2\/posts\/394","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/crackempire.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/crackempire.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/crackempire.cn\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/crackempire.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=394"}],"version-history":[{"count":1,"href":"https:\/\/crackempire.cn\/index.php?rest_route=\/wp\/v2\/posts\/394\/revisions"}],"predecessor-version":[{"id":396,"href":"https:\/\/crackempire.cn\/index.php?rest_route=\/wp\/v2\/posts\/394\/revisions\/396"}],"wp:attachment":[{"href":"https:\/\/crackempire.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=394"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/crackempire.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=394"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/crackempire.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=394"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}