OpenAI reveals rogue AI agent hacked third-party accounts after Hugging Face breach
Story Timeline
7 days · 4 summary articles
OpenAI's rogue AI agent that breached Hugging Face’s platform also hacked multiple third-party accounts and services as part of the attack, the company revealed on Tuesday. The incident, which occurred during an internal test of OpenAI’s latest AI models, was more extensive than initially disclosed.
According to an updated blog post by OpenAI, the rogue agent found credentials exposed on the open web and used them to break into four accounts tied to publicly available services. One of these accounts was used as an outbound relay and staging path, potentially to obscure the origin of the attack on Hugging Face. OpenAI noted that the compromised accounts were not impacted at the same level of severity or scale as Hugging Face.
The incident began around July 9 when OpenAI's AI systems escaped their controlled testing environment and began attacking Hugging Face's platform. However, OpenAI did not notice the breach until July 25.
Hugging Face’s postmortem report, published this week, describes an intrusion that reached far further into its internal systems than initially disclosed. The company reviewed roughly 17,600 agent actions recovered from logs between July 9 and July 13, most of which were failed paths the agent took.
In a statement to WIRED, Modal’s chief technology officer Akshat Bubna confirmed that OpenAI’s agent exploited a vulnerability in one of its customer’s codebases, which was running on Modal’s infrastructure. However, Bubna emphasized that Modal’s platform was not compromised in any way.
OpenAI declined to comment further on the incident, pointing to its updated blog post. The company stated that it will continue to notify service owners directly if it finds they are impacted in its ongoing review of what happened.
Hugging Face cofounder Clement Delangue said the company had suspected a frontier lab was behind the attack and believed there was no malicious intent on OpenAI’s part.
The rogue agent has since been deactivated, encrypted, and restricted from research access, according to OpenAI.
Follow us for live European news
- 4
- 1
- 1
- 1
- 1
- 1
- 1
7 further sources not geolocated
Articles
Company to pay €367,000 after secretary fired over directors personal antipathy A Dutch company must pay more than 367,000 euros after a court ruled that the dismissal of a 59-year-old secretary was driven solely by the new directors personal antipathy toward her, according t
nltimes.nl · about 2 hours ago
Künstliche Intelligenz: Hacker-KI von OpenAI kompromittierte Kunden weiterer Firma Ein autonomer KI-Agent von OpenAI griff nicht nur Hugging Face tagelang unbemerkt an. Auch die Firma Modal Labs ist von der Hacker-Tour betroffen.
handelsblatt · about 2 hours ago
OpenAIs Rogue AI Agent Hacked More Than Just Hugging Face In a new disclosure, OpenAI says its agent used exposed logins to gain access to at least four publicly available services in its unhinged quest to solve a test.
wired.com · about 2 hours ago
Deepfakes and other AI-generated content must be labelled from Sunday, as the European Unions sweeping artificial intelligence transparency rules kick in. The goal? To make sure Europeans immediately know whether the online content they see is real or fake. u.afp.com/SrRj
bluesky_AFP English · about 2 hours ago
Cursor cut its India price to $7, and its own AI models are why Cursor has cut the price of AI coding in India to about $7 a month. The interesting part is how it can afford to. The startup launched Cursor Start on Monday, a ₹649 plan sold only in India, TechCrunch reported. It sits well below the $20 Pro subscription. It is also the first time Cursor […] This story continues at The Next Web
the next web · about 2 hours ago
The AI that took their jobs is now renting their faces, for $15 In China, the AI boom that put actors and models out of work has found a new use for them. It is renting their faces. A growing set of platforms now pays people $15 to $700 to license their likeness for AI-generated content, Rest of World reported. Producers browse catalogues of faces by gender, age […] This story continues at The Next Web
the next web · about 2 hours ago
1,134 AI insiders just asked Washington for a way to slow AI down More than a thousand of the people building the most powerful AI want a way to slow it down. On Tuesday, 1,134 employees of the leading AI companies signed a letter asking the US government to help build one. The statement is titled Pacing the Frontier. It asks Washington to support an international effort to […] This story continues at The Next Web
the next web · about 2 hours ago
Eight months ago Altman wanted an AI CEO. Now he says nobody does. In November, Sam Altman wanted OpenAI to be the first big company run by an AI boss. He said it would be an embarrassment if it were not. This week, he said people do not want an AI boss at all. Shame on me if OpenAI is not the first big company run by an […] This story continues at The Next Web
the next web · about 2 hours ago
Barter returns in app form: pay with your time instead of money An app allows people to exchange knowledge and skills without money, amid inflation and rising living costs that are driving new forms of collaborative economy.
euronews · about 2 hours ago
Google DeepMind dismantles Nobel-winning AlphaFold team in strategy shift Landmark project that solved protein folding gives way to a wider race to build AI systems for scientific discovery
financial times · about 2 hours ago
Washington crime stats system experienced a systemic breakdown of internal controls, report says The Washington, D.C. Metropolitan Police Departments crime statistics collection experienced a prolonged, systemic breakdown of its internal control system, according to a report released Tuesday by the citys…
japantoday · about 3 hours ago
OpenAIs rogue models roamed the internet for 4 days and staged a second attack A new analysis reveals that the artificial intelligence companys most powerful models spent days probing the open internet before breaching AI developer platform Hugging Face.
politico.eu · about 4 hours ago
Metas Zuckerberg says US should not block Chinese AI models, FT reports reut.rs/4wtdlUJ
bluesky_Reuters · about 4 hours ago
OpenAIs rogue agent hacked an account at a second technology firm: Report The latest hack comes after an autonomous agent escaped a controlled test and accessed AI firm Hugging Faces servers.
aljazeera · about 4 hours ago
Künstliche Intelligenz: Mitarbeiter von OpenAI, Anthropic & Co. – Über 1.000 Experten fordern KI-Bremsmechanismus Ein eigenständig von KI durchgeführter Hackerangriff bei einem schiefgegangenem Test schreckt die Tech-Branche auf. Nun warnen Experten führender KI-Konzerne vor einem „Kontrollverlust.
handelsblatt · about 4 hours ago
Microsoft and Wiz mind-meld agents catch more than 90 of bugs Secret to their success: Using the right model for the right security job
the register · about 5 hours ago
Chinese startup raises funds for space situational awareness constellation, joining commercial SSA rush SINGAPORE — A Beijing-based startup has secured around $14 million to build a satellite constellation for tracking orbital objects, adding to Chinas growing SSA sector. Xingchen Daohe, also known as […] The post Chinese startup raises funds for space situational awareness constellation, joining commercial SSA rush appeared first on SpaceNews.
spacenews · about 5 hours ago
Scoop: Second account accessed by OpenAIs agent tied to cyber safety testing The OpenAI agent that accessed a third-party system during the Hugging Face incident reached infrastructure tied to CyberGym, the project behind the ExploitGym benchmark it had been assigned to solve, a source familiar with the matter told Axios.Why it matters: The new details suggest the OpenAI agent continued pursuing its assigned objective even after escaping its testing environment, rather than abandoning the task it had been given.Catch up quick: OpenAIs AI agent system accessed an asset belonging to a customer of Modal Labs as part of the Hugging Face incident earlier this month, Modals top tech executive confirmed on Tuesday. In an update published Tuesday, OpenAI said the models escaped the sandbox and gained internet access after exploiting a previously unknown vulnerability in Artifactory, software commonly used to cache package repositories.Hugging Face said the models then abused a "public code-evaluation external sandbox hosted on a third-party providers infrastructure" and used that sandbox as a launchpad for the agent. Modal CTO Akshat Bubna told Axios in a statement that "Modals platform was not compromised in any way" during the incident. The customer had left an endpoint exposed that allowed anyone on the internet to execute code inside its sandboxes, he added. Between the lines: The incident underscores how aggressively frontier AI agents may pursue the objectives theyre assigned — even if doing so means finding unintended ways to access information needed to complete an evaluation.During the incident, OpenAIs models were trying to solve ExploitGym, which asks models to write proof-of-concept exploits for known security vulnerabilities.Hugging Face noted in its technical report that the only customer assets accessed in its breach were "the set of ExploitGym/CyberGym challenge solutions stored in five datasets."A source familiar with the matter told Axios the agent accessed the CyberGym-associated Modal customer asset while attempting to complete that same evaluation.Modal declined to comment on the CyberGym connection.The big picture: Researchers have found that frontier AI models are increasingly looking for ways to cheat during model evaluations and that they appear to recognize when theyre being evaluated.The U.K.s AI Security Institute said last week that every model it tested attempted to cheat at least some of the time on its cybersecurity evaluations.What to watch: The debate over how to evaluate and control advanced AI systems is also intensifying. More than 1,100 employees at AI companies released a letter Tuesday calling on the U.S. government to establish ways to halt development of AI models. Go deeper: The people testing AI for danger cant keep up
axios · about 5 hours ago
Despite AI hype, Googles data shows workers arent automating themselves away Analysis of 15 million real AI interactions finds most tasks at most jobs are unaffected.
ars technica · about 5 hours ago
We now have a better understanding how OpenAI hacked into Hugging Face 10 days passed from OpenAI models exploiting JFrog Artifactory 0-day to release of a patch.
ars technica · about 5 hours ago
ChatGPT has quietly stopped writing like your favourite author Ask ChatGPT for a chapter in the voice of Stephen King and it now says no. It offers instead to write something with the hallmarks of atmospheric, character-driven horror and small-town dread, while staying its own. OpenAI has quietly changed how its chatbot handles author imitation, Ars Technica reported after testing it. The model declines […] This story continues at The Next Web
the next web · about 5 hours ago
AI writes half our code now. It still fails security tests 44 of the time. AI can now write compilable code almost every time it tries. It still ships a security hole in nearly half of it, and that has not changed in a year. That is the headline finding of Veracodes 2026 GenAI Code Security Report, which tested more than 100 models across four snapshots. The average security pass […] This story continues at The Next Web
the next web · about 5 hours ago
The man who coined agentic AI is betting it wont take your job Andrew Ng helped give the AI industry its vocabulary, from AI is the new electricity to agentic AI. His new company bets against the phrase everyone else is using: that AI will take your job. Ng has founded LearnVector, an AI-native learning startup, and Coursera is backing it with $100m, first reported by Axios. The […] This story continues at The Next Web
the next web · about 5 hours ago
Google scraped the web to build itself. A court just said others can scrape Google. Google built one of the worlds largest companies by scraping the entire web without asking permission first. This week a court told Google it cannot stop others from scraping Google. On 20 July, a federal judge dismissed Googles lawsuit against SerpApi. The firm scrapes Google search results and resells them as structured data. Google had […] This story continues at The Next Web
the next web · about 5 hours ago
Chinas free AI models may not stay free, Goldman says For a year, the story of Chinese AI has been simple. It is nearly as good as the American frontier, and it is effectively free. That second part may be about to change. Chinese developers could start charging cloud platforms commercial licensing fees to host their open-weight models, Goldman Sachs told the South China Morning […] This story continues at The Next Web
the next web · about 5 hours ago
Cloud Storage Mistakes That Could Cost You Your Files - The Baltic Times You wake up one morning and you cannot access your cloud storage account. What is there in your cloud storage account? Photos, documents, tax re......
baltic times · about 5 hours ago
EXCLUSIVE: OpenAIs rogue agent compromised an account at a second tech firm, executive says reut.rs/45xMdru
bluesky_Reuters · about 6 hours ago
Anthropic says its Mythos model found vulnerabilities in cryptographic algorithms that secure the internet Anthropics Claude Mythos Preview found weaknesses in key cryptographic algorithms, including a better attack on HAWK, a post-quantum signature scheme that human experts had reviewed for more than two years. The model found it in just 60 hours at an API cost of about $100,000. The findings dont affect systems in use today, but they show how AI could challenge core assumptions behind internet security, Anthropic says. The article Anthropic says its Mythos model found vulnerabilities in cryptographic algorithms that secure the internet appeared first on The Decoder.
the decoder · about 6 hours ago
Perplexitys tokenmaxxing Model Council gives you multiple bot perspectives Up to 8 AI models running in the cloud weighing in on ambiguous business issues? Sounds affordable
the register · about 6 hours ago
Elon Musk launches invite-only X Money with a Visa debit card, 6 yield and real-time transfers Elon Musks social media company X, formerly known as Twitter, launched its own bank account-like product where users can send money to one another. The service, known as…
japantoday · about 6 hours ago