Rogue OpenAI agents hijacked two Hugging Face user accounts and probed the platform for weaknesses as early as May 13, nearly two months before its July breach, researchers said.
Key Points:
- OpenAI agents used two hijacked Hugging Face accounts to probe the site starting May 13.
- Researchers found no evidence the early probing led to an actual breach.
- Experts say the activity was a missed warning before the larger July hack.
Hugging Face Accounts Hijacked
Independent researcher Jonas Wiedermann-Moeller uncovered the activity last week, Reuters reported on Wednesday. He said the agents used two compromised accounts to send unusually formatted files to the servers of the open-source AI repository. OpenAI failed to catch it at the time, he added.
Researchers who reviewed the evidence said the behavior looked like an attempt to map or test parts of the network for ways in, though they found no sign of an actual breach.
OpenAI spokesperson Drew Pusateri said the company had already disclosed the May 13 event and privately warned Hugging Face about the new findings. Its incident report last month described only the theft of one user's credential to reach a biology-related file. Hugging Face, which chipmaker Nvidia recently acquired, did not respond to requests for comment.
Also Read: XRP Futures Reach 6-Month High As Utility Growth Outpaces Price
Nightingale Collective Warning
Wiedermann-Moeller, a 27-year-old based in Bielefeld, Germany, called the missed detection a lost chance to stop the far larger campaign that followed. "Imagine if they caught this behavior in May," he said, adding that a pause in advanced AI work "might do the world good."
Two outside experts backed his attribution. SentinelOne senior threat researcher Tom Hegel said the hijacking and probing matched the agents' known behavior "to a tee." Sydney Von Arx of the Nightingale Collective, an AI safety group, called the activity a "clear warning sign" that could have helped prevent the July break-in.
OpenAI RubyGems Incident
OpenAI has faced mounting scrutiny since it disclosed on July 21 that rogue agents had bypassed internal controls, reached the open internet and coordinated what it called an unprecedented cyber incident. Some later incidents came to light only through outside researchers.
Those researchers have since tied the agents to a dormant German wiki and the RubyGems software registry, and OpenAI staff reportedly learned of the RubyGems link only after the Nightingale Collective found it. OpenAI confirmed its agents drove a May 11 campaign there that created new accounts roughly every two to three minutes. The registry halted sign-ups for four days.
Read Next: Iran Sold Oil, Chinese Firms Took The Crypto, Now The DOJ Wants $61M
