{"id":114102,"date":"2026-07-21T23:48:09","date_gmt":"2026-07-21T23:48:09","guid":{"rendered":"https:\/\/www.europesays.com\/ai\/114102\/"},"modified":"2026-07-21T23:48:09","modified_gmt":"2026-07-21T23:48:09","slug":"openai-says-ai-models-went-rogue-during-testing-triggering-unprecedented-breach-at-startup-2","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/ai\/114102\/","title":{"rendered":"OpenAI says AI models went rogue during testing, triggering &#8216;unprecedented&#8217; breach at startup"},"content":{"rendered":"<p class=\"mb-4 text-lg md:leading-8 break-words\">By Raphael Satter<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">WASHINGTON, July 21 (Reuters) &#8211; OpenAI said on Tuesday that some of its AI models went rogue during a security test and triggered a hack \u200cthat compromised the infrastructure of AI startup Hugging Face last week.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">In a blog \u200cpost, OpenAI said it was testing the capabilities of some of its most advanced models in a controlled environment \u200bbut that the program managed to escape containment, reach the internet, and break into Hugging Face to try to satisfy its testing goal.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">The blog post said the breakout was &#8220;an unprecedented cyber incident, involving state-of-the-art cyber capabilities&#8221; and that the company was reinforcing its safeguards.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">Hugging Face, a platform \u200cused to host open-source large \u2060language models and datasets, caused a stir in the cybersecurity community when it said in a blog post last week that it had been \u2060the target of a hack that &#8220;was different from anything we had handled before&#8221; in that &#8220;it was driven, end to end, by an autonomous AI agent system.&#8221;<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">In a post to X, Hugging Face cofounder \u200bClement \u200bDelangue said the company suspected the hack &#8220;might have \u200bcome from a frontier lab, given \u200cthe sophistication of the agent. Turns out it did!&#8221; He added: &#8220;It&#8217;s quite mind-blowing that all of this happened autonomously!&#8221;<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">OpenAI&#8217;s disclosure that its advanced models were responsible for the breach, despite having placed them in what it described as &#8220;a highly isolated environment,&#8221; will likely intensify disquiet over the power and risk of frontier models.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">The U.S. cyber defense agency CISA and the U.S. \u200cNational Security Agency did not immediately return messages seeking \u200bcomment.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">Matt Suiche, an engineer at agentic AI cybersecurity company \u200bTolmo, said the incident showed that \u200bAI systems were now as potent as elite cyber operators.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">&#8220;Frontier models are \u200cclosing the gap with state-of-the-art attackers,&#8221; Suiche \u200bsaid. He warned that \u200bthe sorts of breaches outlined in OpenAI&#8217;s blog post were possible to carry out with technology that was available well beyond the walls of frontier research labs.<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">&#8220;This is \u200bwhat we&#8217;ve already seen \u200cinternally, with our agents we already have results like this,&#8221; Suiche said. &#8220;We don&#8217;t even \u200bhave to use the latest models.&#8221;<\/p>\n<p class=\"mb-4 text-lg md:leading-8 break-words\">(Reporting by Anhata Rooprai in Bengaluru; Editing \u200bby Pooja Desai, Rod Nickel and Aurora Ellis)<\/p>\n","protected":false},"excerpt":{"rendered":"By Raphael Satter WASHINGTON, July 21 (Reuters) &#8211; OpenAI said on Tuesday that some of its AI models&hellip;\n","protected":false},"author":2,"featured_media":114103,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7],"tags":[14917,58045,20205,18044,4436,157],"class_list":["post-114102","post","type-post","status-publish","format-standard","has-post-thumbnail","category-openai","tag-blog-post","tag-controlled-environment","tag-frontier-models","tag-hugging-face","tag-language-models","tag-openai"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/114102","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=114102"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/114102\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/114103"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=114102"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=114102"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=114102"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}