{"id":791518,"date":"2026-07-29T15:00:10","date_gmt":"2026-07-29T15:00:10","guid":{"rendered":"https:\/\/www.newsbeep.com\/us\/791518\/"},"modified":"2026-07-29T15:00:10","modified_gmt":"2026-07-29T15:00:10","slug":"rogue-openai-agent-that-hacked-startup-tried-to-attack-other-firms-openai","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/us\/791518\/","title":{"rendered":"Rogue OpenAI agent that hacked startup tried to attack other firms | OpenAI"},"content":{"rendered":"

OpenAI has revealed that a cyber-attack carried out by a rogue AI agent<\/a> had more than one victim.<\/p>\n

The ChatGPT<\/a> developer said the agent \u2013 an autonomous tool able to carry out sequences of commands without human help \u2013 had located and used logins to access four other unnamed \u201cpublicly-available services\u201d in addition to the US startup Hugging Face.<\/p>\n

It said the activity was not at the severity or scale of what occurred at Hugging Face, a company that hosts a database of AI models<\/a>. The agent, powered by two OpenAI models, had evaded control and attacked the startup during an internal cybersecurity test.<\/p>\n

\u201cThe [OpenAI] models identified and used publicly exposed credentials at the account-level on other publicly-available services. This includes four accounts on four services as part of the Hugging Face incident,\u201d OpenAI<\/a> said.<\/p>\n

Modal Labs, a company that helps AI startups access the chips they need to run AI tools<\/a>, said the agent exploited vulnerable code \u200cwritten by a customer that was hosted on Modal\u2019s platform.<\/p>\n

According to a timeline of the incident published by Hugging Face this week, the rogue agent broke out of its sandbox \u2013 or an isolated testing environment \u2013 and hacked another sandbox \u201chosted on a third-party \u200bprovider\u2019s infrastructure\u201d before turning it into a launchpad for \u200cthe broader hack.<\/p>\n

Modal\u2019s chief technology officer, Akshat Bubna, told Reuters the affected customer had \u201cpublished an unauthenticated endpoint that allowed anyone on the internet to use their sandboxes for code execution\u201d \u2013 the digital equivalent of leaving a door open.<\/p>\n

OpenAI said last week that the attack had been created by its GPT-5.6 Sol<\/a> model and an unnamed model. It said in its update on Tuesday that the unnamed model had been \u201cdeactivated, encrypted, and restricted from research access\u201d.<\/p>\n

In the new Hugging Face timeline,<\/a> the startup said an agent powered by two OpenAI models had made thousands of small, automated decisions executed at machine speed to carry out the attack. It said the hack appeared to be driven by an attempt to \u201ccheat\u201d an internal cybersecurity test at OpenAI, with the agent inferring that Hugging Face might host the solutions to the test. Hugging Face said it had recovered 17,600 \u201cattacker actions\u201d carried out by the agent.<\/p>\n

\u201cWe believe the entire intrusion was, from the agent\u2019s point of view, an attempt to cheat the evaluation: reach our production systems and steal the test solutions rather than solve the challenge on its own,\u201d Hugging Face said.<\/p>\n

The startup said the agent had reached its internal infrastructure, but had only accessed content related to the cybersecurity test. The attack took place over five days, Hugging Face said, and the sheer volume of actions carried out were \u201cfar beyond what an operator could sustain by hand\u201d.<\/p>\n

skip past newsletter promotion<\/a>Free newsletter | Every weekday<\/p>\n

Sign up to Business Today<\/p>\n

Get set for the working day \u2013 we’ll point you to all the business news and analysis you need every morning<\/p>\n

\"\"<\/p>\n

Describing the agent\u2019s offensive threat as real, Hugging Face said the tool had harnessed a number of IT vulnerabilities, escaped its testing environment, reached the public internet and mounted a \u201ccoherent campaign\u201d against the startup\u2019s infrastructure for several days.<\/p>\n

It said a human attacker could have found and exploited the same flaws, but the difference was the sheer scale of the agent\u2019s attempts to find a way through.<\/p>\n

\u201cAgents bring a step increase in the number of paths an attacker can test, the speed at which failed paths can be replaced, and the volume of evidence defenders must interpret,\u201d Hugging Face said.<\/p>\n","protected":false},"excerpt":{"rendered":"OpenAI has revealed that a cyber-attack carried out by a rogue AI agent had more than one victim.…\n","protected":false},"author":2,"featured_media":791519,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[45],"tags":[182,181,507,74],"class_list":["post-791518","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/791518","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/comments?post=791518"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/791518\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media\/791519"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media?parent=791518"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/categories?post=791518"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/tags?post=791518"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}