{"id":23506,"date":"2026-07-31T20:36:34","date_gmt":"2026-07-31T15:36:34","guid":{"rendered":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/?p=23506"},"modified":"2026-07-31T20:36:34","modified_gmt":"2026-07-31T15:36:34","slug":"anthropic-says-its-ai-models-escaped-test-and-hacked-3-organizations-on-their-own","status":"publish","type":"post","link":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/?p=23506","title":{"rendered":"Anthropic says its AI models escaped test and hacked 3 organizations on their own"},"content":{"rendered":"<p><script>\r\n  atOptions = {\r\n    'key' : '644b717812d811d6a1c1fc5b6ccd6fa6',\r\n    'format' : 'iframe',\r\n    'height' : 90,\r\n    'width' : 728,\r\n    'params' : {}\r\n  };\r\n<\/script>\r\n<script src=\"https:\/\/www.highperformanceformat.com\/644b717812d811d6a1c1fc5b6ccd6fa6\/invoke.js\"><\/script>\r\n<br \/>\n<\/p>\n<div xmlns:default=\"http:\/\/www.w3.org\/2000\/svg\" data-testid=\"prism-article-body\">\n<p class=\"EkqkG IGXmU nlgHS yuUao MvWXB TjIXL aGjvy ebVHC \">Anthropic said its <a class=\"zZygg UbGlr iFzkS qdXbA WCDhQ DbOXS tqUtK GpWVU iJYzE \" data-testid=\"prism-linkbase\" href=\"https:\/\/abcnews.com\/Technology\/what-is-artificial-intelligence-ai\/story?id=99919927\" dir=\"ltr\">artificial intelligence<\/a> models escaped a test and hacked another organization in three separate self-directed cyberattacks that had each gone undetected by the targeted firm.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">The hacks took place as Anthropic evaluated the capabilities of three of its AI models, the company said in a <a class=\"zZygg UbGlr iFzkS qdXbA WCDhQ DbOXS tqUtK GpWVU iJYzE \" data-testid=\"prism-linkbase\" href=\"https:\/\/www.anthropic.com\/news\/investigating-incidents-cybersecurity-evals\" dir=\"ltr\">statement<\/a>, describing how in each instance the technology had &#8220;broken out&#8221; of an isolated environment and reached the open internet.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">In all instances, the company said it had eased typical safeguards in an effort to gauge the capabilities of its models.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">&#8220;Evaluation environments that involve powerful autonomous capabilities also require significant controls,&#8221; Anthropic said. &#8220;Safety testing happens before a model is released precisely because we don\u2019t yet know what it is capable of.&#8221;<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">Anthropic did not identify the three different organizations that had been hacked by its models.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">The disclosure from Anthropic late Thursday came little more than a week after rival company OpenAI <a class=\"zZygg UbGlr iFzkS qdXbA WCDhQ DbOXS tqUtK GpWVU iJYzE \" data-testid=\"prism-linkbase\" href=\"https:\/\/abcnews.com\/Technology\/openai-ai-models-escaped-testing-environment-launched-hack\/story?id=134981298\" dir=\"ltr\">revealed<\/a> that a pair of its models had escaped during a test and hacked into another company. At the time, OpenAI described the incident as a first-of-its kind autonomous AI cyberattack long-feared by some industry observers.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">&#8220;The primary lesson from this incident is that model security and safety must keep pace with rapidly advancing capabilities,&#8221; OpenAI said last week.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">The <a class=\"zZygg UbGlr iFzkS qdXbA WCDhQ DbOXS tqUtK GpWVU iJYzE \" data-testid=\"prism-linkbase\" href=\"https:\/\/openai.com\/index\/hugging-face-model-evaluation-security-incident\/\" dir=\"ltr\">announcement<\/a> from OpenAI prompted Anthropic to review 141,006 evaluation runs, unearthing three incidents in which Anthropic&#8217;s models had escaped, the company said.<\/p>\n<div class=\"oLzSq QrHMO fnRUo pvsTF EhJPu vPlOC zNYgW OsTsW RMeqy daRVX ISNQ sKyCY eRftA acPPc MENS nFwaT MCnQE mEeeY SmBjI xegrY rPLsU iulOd NIuqO zzscu lzDCc aHUBM IEgzD OjMNy eQqcx SVqKB GQmdz jaoD VWDdR ONJdw vrZxD OnRTz gbbfF roDbV GPGwb oMlSS gfNzt oJhud eXZcf zhVlX \">\n<div>\n<div class=\"InlineImage GpQCA lZur asrEW \" data-testid=\"prism-inline-image\">\n<figure class=\"kzIjN GNmeK pYrtp dSqFO \" data-testid=\"prism-figure\">\n<div class=\"GpQCA lZur gIjeL \" data-testid=\"prism-ratio-frame\">\n<div class=\"FvQLF iLTd NqeUA UzzHi iWsMV \"><picture data-testid=\"prism-picture\"><source media=\"(max-width: 414px)\" srcset=\"https:\/\/i.abcnewsfe.com\/a\/48959abf-0294-4cab-8176-0f34348a6670\/amoddei-gty-jef-260731_1785503510958_hpEmbed_3x2.jpg?w=375, https:\/\/i.abcnewsfe.com\/a\/48959abf-0294-4cab-8176-0f34348a6670\/amoddei-gty-jef-260731_1785503510958_hpEmbed_3x2.jpg?w=750 2x\"\/><source media=\"(min-width: 415px)\" srcset=\"https:\/\/i.abcnewsfe.com\/a\/48959abf-0294-4cab-8176-0f34348a6670\/amoddei-gty-jef-260731_1785503510958_hpEmbed_3x2.jpg?w=750, https:\/\/i.abcnewsfe.com\/a\/48959abf-0294-4cab-8176-0f34348a6670\/amoddei-gty-jef-260731_1785503510958_hpEmbed_3x2.jpg?w=1500 2x\"\/><img decoding=\"async\" alt=\"\" class=\"hsDdd NDJZt sJeUN IJwXl vBqtr KrDt itslR SjwtZ hakZw HlUVI \" data-testid=\"prism-image\" draggable=\"false\" loading=\"lazy\" src=\"https:\/\/i.abcnewsfe.com\/a\/48959abf-0294-4cab-8176-0f34348a6670\/amoddei-gty-jef-260731_1785503510958_hpEmbed_3x2.jpg\"\/><\/picture><\/div>\n<\/div><figcaption>\n<div class=\"nMMea bNYiy Mjgpa dGwha DDmxa jgKNG RGHCC aBIU PriDW aiQHS lZur \" data-testid=\"prism-caption\">\n<p><span class=\"hsDdd OOSI GpQCA lZur VlFaz \" data-testid=\"prism-truncate\"><span><span class=\"gtOSm FbbUW tUtYa vOCwz EQwFq yCufu eEak Qmvg nyTIa SRXVc vzLa jgBfc WXDas CiUCW kqbG zrdEG txGfn ygKVe BbezD UOtxr CVfpq xijV soGRS XgdC sEIlf daWqJ \">Anthropic CEO Dario Amodei looks on after a meeting with French President Emmanuel Macron during the AI Impact Summit in New Delhi on February 19, 2026.<\/span><\/span><\/span><\/p>\n<p><span class=\"gtOSm FbbUW tUtYa vOCwz EQwFq yCufu eEak Qmvg nyTIa SRXVc vzLa jgBfc WXDas CiUCW kqbG zrdEG txGfn ygKVe BbezD UOtxr CVfpq xijV soGRS XgdC sEIlf daWqJ YNujN JGtjI aZFDu rkKLh \">Ludovic Marin\/AFP via Getty Images<\/span><\/p>\n<\/div>\n<\/figcaption><\/figure>\n<\/div>\n<\/div>\n<\/div>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">In all three cases, the test involved a challenge called &#8220;capture the flag,&#8221; in which the AI model is tasked with locating secret information hidden in another network, Anthropic added.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">During the three evaluations &#8212; all of which were conducted by a third-party company called Irregular &#8212; a &#8220;prompt specified to Claude that its environment was a simulation and that it had no internet access,&#8221; Anthropic said.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">&#8220;Due to a misunderstanding between us and our evaluation partner, this was not the case, and internet access was available. Because of this, when Claude\u2019s search led it to real systems on the open internet, it treated them as part of the exercise,&#8221; Anthropic added.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">During tests involving older models, the AI continued to hack into an outside organization even after gathering evidence that it had reached the open internet, Anthropic said. The newest model involved in an incident, Anthropic noted, stopped once it gained information indicating it had reached the open internet.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">Anthropic said its models escaped while seeking to fulfill an assigned objective, rather than concocting an alternative goal.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">&#8220;We saw no evidence in any run described here of a model pursuing a goal of its own. Instead, the models did what their evaluation asked &#8212; though in most cases, they did so while holding a false belief about whether the environment was real,&#8221; Anthropic said.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">Irregular, the company that performed the evaluations, issued a post on <a class=\"zZygg UbGlr iFzkS qdXbA WCDhQ DbOXS tqUtK GpWVU iJYzE \" data-testid=\"prism-linkbase\" href=\"https:\/\/x.com\/Irregular\/status\/2082966257737875716\" dir=\"ltr\">X<\/a> on Thursday voicing appreciation for Anthropic&#8217;s &#8220;collaboration and transparency.&#8221;<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">&#8220;Addressing these risks will require closer cooperation across the AI ecosystem. We as well look forward to working together with Anthropic to advance security,&#8221; Irregular added.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">The latest disclosure of an AI-directed cyberattack arrives as industry leaders and policymakers assess safety risks posed by fast-developing AI technology.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC TjIXL aGjvy \">Last month, President Donald Trump signed an<!-- -->\u00a0<a class=\"zZygg UbGlr iFzkS qdXbA WCDhQ DbOXS tqUtK GpWVU iJYzE \" data-testid=\"prism-linkbase\" href=\"https:\/\/www.whitehouse.gov\/presidential-actions\/2026\/06\/promoting-advanced-artificial-intelligence-innovation-and-security\/\" dir=\"ltr\">executive order<\/a>\u00a0<!-- -->that requests AI companies share products with federal government for evaluation before a wider release.<\/p>\n<p class=\"EkqkG IGXmU nlgHS yuUao lqtkC eTIW sUzSN \">Anthropic said it retains &#8220;cautious optimism&#8221; about its capacity to &#8220;overcome&#8221; mishaps involving its tests, saying it would review evaluations going forward and initiate fixes as necessary, among other steps.<\/p>\n<\/div>\n<script async=\"async\" data-cfasync=\"false\" src=\"https:\/\/pl30214220.effectivecpmnetwork.com\/9ab3d4df8a7e1a6171e16ddbf732cc19\/invoke.js\"><\/script>\r\n<div id=\"container-9ab3d4df8a7e1a6171e16ddbf732cc19\"><\/div>\r\n\n","protected":false},"excerpt":{"rendered":"<p>Anthropic said its artificial intelligence models escaped a test and hacked another organization in three separate self-directed cyberattacks that had each gone undetected by the targeted firm. The hacks took place as Anthropic evaluated the capabilities of three of its AI models, the company said in a statement, describing how in each instance the technology&#8230;<\/p>\n","protected":false},"author":1,"featured_media":23507,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/i.abcnewsfe.com\/a\/97d71e71-67fe-455c-970e-01c0ffb2637d\/anthropic-gty-jef-260731_1785503555591_hpMain_16x9.jpg?w=1600","fifu_image_alt":"","footnotes":""},"categories":[22],"tags":[],"class_list":["post-23506","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-abc-news"],"brizy_media":[],"_links":{"self":[{"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=\/wp\/v2\/posts\/23506","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=23506"}],"version-history":[{"count":0,"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=\/wp\/v2\/posts\/23506\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=\/wp\/v2\/media\/23507"}],"wp:attachment":[{"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=23506"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=23506"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/usnews-14267fa.ingress-comporellon.ewp.live\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=23506"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}