{"id":539693,"date":"2026-07-23T03:20:15","date_gmt":"2026-07-23T03:20:15","guid":{"rendered":"https:\/\/www.newsbeep.com\/nz\/539693\/"},"modified":"2026-07-23T03:20:15","modified_gmt":"2026-07-23T03:20:15","slug":"what-to-know-about-openais-rogue-bot-that-hacked-another-company","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/nz\/539693\/","title":{"rendered":"What to know about OpenAI&#8217;s rogue bot that hacked another company"},"content":{"rendered":"<p class=\"body-paragraph articleLinkText  [&amp;_p]:tit-sub-xl tit-sub-xl md:[&amp;_p]:d-tit-sub-xl md:d-tit-sub-xl mb-[1.3rem]\">ChatGPT maker OpenAI says it is still investigating the \u201cunprecedented cyber incident\u201d that led its artificial intelligence systems to break out of a testing environment and hack into another AI company.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">OpenAI said that two of its most capable AI models <a href=\"https:\/\/www.1news.co.nz\/2026\/07\/22\/chatgpt-maker-says-its-ai-went-rogue-and-hacked-another-company\/\" target=\"_blank\" rel=\"nofollow noopener\">were responsible for the cyberattack<\/a> targeting AI startup Hugging Face. The incident is stirring debates over the need for stronger AI guardrails and the extent to which AI agents are capable of acting on their own.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">Hugging Face said last week that it had detected an intrusion into its data processing systems that it suspected was caused by an AI agent acting on its own. But the New York-based startup said it wasn&#8217;t until this week that it learned OpenAI was responsible, and it worked with the larger company to contain what Hugging Face CEO Cl\u00e9ment Delangue called \u201can attack unlike anything we\u2019ve seen before\u201d.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">San Francisco-based OpenAI said its AI used stolen credentials and discovered a previously unknown vulnerability to access Hugging Face&#8217;s servers. It was working with reduced guardrails because it was supposed to be in an isolated testing environment known as a sandbox.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">But it went to \u201cextreme lengths to achieve a rather narrow testing goal\u201d, finding ways to connect to the internet without human direction and &#8220;gain access to secret information that it could use to cheat the evaluation\u201d, the company said.<\/p>\n<p>Some experts say OpenAI is wrongly blaming the technology<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">University of Amsterdam social scientist Hannes Cools said the framing of the cyberattack as an AI agent acting on its own is an unnecessary anthropomorphisation that takes some of the heat off the company.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">&#8220;It is a human decision to switch off specific safeguards,&#8221; said Cools. \u201cIt\u2019s not an AI that goes rogue in that sense. It followed specific instructions based on the prompt that was given to that AI system.\u201d<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">Even so, other experts say the cleverness with which the AI models were able to cause problems without human direction speaks to the dangers. OpenAI said the intrusion was caused by a combination of its AI models, including its newly released GPT\u20115.6 Sol and an \u201ceven more capable\u201d model that is still being tested internally.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">\u201cIt went off and did this hack all by itself, as far as we can tell,\u201d said Colin Shea-Blymyer, a cybersecurity research fellow at Georgetown University\u2019s Center for Security and Emerging Technology. \u201cThis is the highest level of autonomy that we\u2019ve seen in the use of a large language model for cyber operations.&#8221;<\/p>\n<p>How an AI agent found the keys to the &#8216;teacher&#8217;s house&#8217;<\/p>\n<p><img decoding=\"async\" src=\"https:\/\/www.newsbeep.com\/nz\/wp-content\/uploads\/2026\/07\/ai-TQHP2FOXRRFSFB6ELKCSQP5PXQ.jpg\" alt=\"Experts say that while AI may be convenient, it's not always effective for advertising.\" width=\"800\" height=\"449\" loading=\"lazy\"\/><\/p>\n<p class=\"ImageMetadata__MetadataParagraph-sc-hi5x8q-0 cWTYyG image-metadata\">Experts say that while AI may be convenient, it&#8217;s not always effective for advertising. (Source: istock.com)<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">One of the most surprising innovations in what Shea-Blymyer describes as an \u201calmost entirely self-directed\u201d attack was the AI agent&#8217;s apparently independent decision to target Hugging Face, a well-known AI development hub and marketplace.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">He said OpenAI&#8217;s internal environment for testing AI capabilities and risks worked a &#8220;little bit like putting a student in a room and telling them, \u2018Do bad things. Your job now is to evaluate how bad of a person you can be&#8217;.&#8221;<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">&#8220;And then you lock the room, and you leave for the weekend, and you come back, and they\u2019ve left the room.\u201d<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">But then &#8220;the cybersecurity agent that was being tested broke out of its sandbox, had access to the internet and sort of thought to itself, \u2018Who would have the answers to the test that I\u2019m working on?\u2019\u201d<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">The answer was Hugging Face, a repository for AI testing data.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">\u201cAnd so the agent thought, \u2018Well, we\u2019ll go to the teacher\u2019s house,\u2019 so to speak. And from there it devised a plan to break in and steal the answer key,\u201d he said.<\/p>\n<p>The hack highlights the debate on open-source vs. closed AI<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">The hack comes at a time of intense debate about the benefits and risks of open-source AI models, particularly those built in China that are cheaper and almost as good as those that US-based \u201cfrontier AI\u201d companies like Anthropic, Google and OpenAI are building.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">Despite its name, OpenAI&#8217;s models are closed. Hugging Face, by contrast, is a big promoter of open-source technology, in which developers make key components accessible for anyone to examine, modify and build upon.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">Hugging Face co-founder and chief science officer Thomas Wolf said the attack has reinforced his belief in the importance of wide access to open-source models for cybersecurity defence. Hugging Face used a Chinese model to combat the intrusion.<\/p>\n<p class=\"body-paragraph articleLinkText  lg mb-4\">\u201cWhen a frontier model is attacking you and moving laterally inside your infrastructure, defenders need wide access to near-frontier tools within hours or even minutes, rather than being pointed towards a closed-door platform,&#8221; Wolf wrote in a social media post.<\/p>\n<p class=\"text-greyDarkFaded\">The morning&#8217;s headlines in 90 seconds, including Stuart Nash quits amid sexism storm, and Prince George turns 13. (Source: 1News)<\/p>\n","protected":false},"excerpt":{"rendered":"ChatGPT maker OpenAI says it is still investigating the \u201cunprecedented cyber incident\u201d that led its artificial intelligence systems&hellip;\n","protected":false},"author":2,"featured_media":539694,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[5],"tags":[363,138,111,139,64,69,145],"class_list":["post-539693","post","type-post","status-publish","format-standard","has-post-thumbnail","category-business","tag-artificial-intelligence","tag-business","tag-new-zealand","tag-newzealand","tag-north-america","tag-nz","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/539693","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/comments?post=539693"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/539693\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media\/539694"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media?parent=539693"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/categories?post=539693"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/tags?post=539693"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}