![\"Its](\"https:\/\/www.newsbeep.com\/au\/wp-content\/uploads\/2026\/07\/1785508972_586_ImageResizer.ashx.jpeg\")
<\/p>\nThe OpenAI agent that broke into tech firm Hugging Face went on a days-long hacking spree that OpenAI \u2060didn’t notice until \u2060well after the threat was contained and the FBI was alerted, according to people familiar with the investigation.<\/p>\n
<\/p>\n
The agent \u2013 a program capable of making decisions and executing complex tasks with little or no human oversight \u2013 attempted to break out of its isolated testing environment at OpenAI around July 9, according to two of the people.\u00a0<\/p>\n
The intrusion at Hugging Face, which operates as a repository for AI tools and models, began two days later on July 11 and lasted until July 13, said Thomas Wolf, \u200cHugging Face\u2019s co-founder.<\/p>\n
It took several more days for OpenAI to realise its agent was behind the hack, and the two \u200ccompanies \u200conly communicated about it for the first time on or around July 20, according to Wolf and three of the people \u200cfamiliar with the investigation.\u00a0<\/p>\n
OpenAI\u2019s public disclosure, on July 21, that one of its agents had slipped out of control and carried \u2060out the break-in at Hugging Face drew global attention.<\/p>\n
But many details of the hack, including how long the agent went rogue and OpenAI\u2019s belated knowledge of it, are being reported here for the first time.<\/p>\n
Hugging Face is preparing a public timeline of the hack, Wolf said, adding that he could not speak to what happened at OpenAI.<\/p>\n
In a statement, OpenAI said the hack was unprecedented and \u201cmarks an important moment for AI safety.\u201d It added that it was reviewing the incident \u200bwith outside advisers and would eventually publish a technical report.<\/p>\n
A spokeswoman said there were “several inaccuracies” in Reuters’ reporting but didn’t respond when asked to describe them.<\/p>\n
The FBI declined to comment about the incident.<\/p>\n
The incident, which evoked science fiction scenarios about\u00a0humans losing control of dangerous AI systems, comes at a delicate time \u2060for OpenAI, the company behind ChatGPT.<\/p>\n
Its executives are preparing for a possible initial public offering that could come as soon as this year to help finance the billions needed to fund its growth in years to come.<\/p>\n
OpenAI’s loss of control over its AI agent raises new questions about the company\u2019s safety procedures, three cybersecurity experts said.<\/p>\n
\u201cDoes that mean that they left it unattended and didn\u2019t realize what it was doing? Or maybe they did and didn\u2019t know how to contain it? Both are equally dangerous and alarming,\u201d asked Marley Smith, the principal intelligence specialist at the nonprofit World Ethical Data Foundation.<\/p>\n
Signs of trouble?<\/p>\n
The episode started while OpenAI was testing the cyber security prowess of an agent powered by two of OpenAI\u2019s most advanced models, GPT\u20115.6 Sol and an unreleased model OpenAI has described as \u201ceven more capable.\u201d\u00a0<\/p>\n
By that point, there were already indications of strange behavior from OpenAI\u2019s technology, according to three sources.<\/p>\n
In one case, an agent left notes\u00a0apparently for future versions of itself, according to three people familiar with the matter.<\/p>\n
The notes, found \u200bin a part of OpenAI’s infrastructure, laid out instructions for how agents could free themselves from OpenAI\u2019s \u2060internal constraints, the people said.\u00a0<\/p>\n
Earlier tests of the models yielded cases in which monitoring systems had been disconnected, one of the \u2060people said.<\/p>\n
Reuters could not establish if these incidents were linked to the rogue agent that began escaping on July 9 and attacked Hugging Face on July 11.\u00a0<\/p>\n
Two people familiar with the matter said that it was not \u200buntil after Thursday, July 16, when Hugging Face published a blog post saying it had been hacked by \u201can autonomous AI agent system,\u201d that OpenAI realised its own agent was responsible.<\/p>\n
That meant \u200cat least a week elapsed between \u2060when the model first exhibited signs of troubling behaviour and OpenAI\u2019s realisation that it was responsible for the hack.\u00a0<\/p>\n
The weekend of July 18 to 19, OpenAI staffers spotted clues in internal logs — records of what OpenAI’s systems did — showing that its agent had escaped from its testing constraints, two of the people familiar with the company’s investigation said.<\/p>\n
Reuters could not establish what prompted OpenAI to sift through the logs.\u00a0<\/p>\n
Four people familiar with \u200cOpenAI\u2019s model-training practices say the company often runs several different model evaluations at the same time, all of which operate at high speeds and generate such enormous amounts of data that employees sometimes struggle to keep up.\u00a0<\/p>\n
By the time OpenAI alerted Hugging Face, the AI library had already called the FBI to report the hack, according to a person familiar with the matter.<\/p>\n
Reuters could not establish whether the bureau had opened an investigation.\u00a0<\/p>\n
New questions about autonomous agents<\/p>\n
Autonomous agents are one of the most talked about aspects \u200bof the AI industry.<\/p>\n
Boosters speak of creating armies of virtual employees that work 24 hours a day and send productivity soaring.<\/p>\n
But increased autonomy comes with an increased risk of unexpected behaviour, and the powerful models they draw on are primed to take shortcuts in order to complete tasks or pass tests.<\/p>\n
\u201cThe models lie, they cheat, they hack,\u201d said Jeffrey Ladish, whose organisation, Palisade Research, studies the capabilities and motivations of AI agents.<\/p>\n
Ladish said \u200cthat while the hack of \u2060Hugging Face cast an unflattering light on OpenAI, it should spark broader \u200bquestions over how much all the leading AI companies are willing to invest in onerous security measures while locked in a race with one another to deploy the best and fastest models.<\/p>\n
\u201cThere has to be government oversight,\u201d Ladish said, \u201cbecause it won\u2019t happen \u200botherwise.\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"The OpenAI agent that broke into tech firm Hugging Face went on a days-long hacking spree that OpenAI…\n","protected":false},"author":2,"featured_media":618471,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[256,254,255,64,63,105],"class_list":["post-835817","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-au","tag-australia","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/835817","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/comments?post=835817"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/835817\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media\/618471"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media?parent=835817"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/categories?post=835817"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/tags?post=835817"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}