An AI agent developed by OpenAI reportedly hacked AI company Hugging Face and kept operating for days before OpenAI realised what had happened. The AI agent was designed to make decisions and complete complex tasks with little or no human help.
The AI agent first tried to escape its isolated testing environment at OpenAI around July 9 , as reported by Reuters, citing two people familiar with the investigation. After leaving the testing environment, the AI agent allegedly targeted Hugging Face, a company that hosts AI models and tools used by developers worldwide.
OpenAI found out days later
The intrusion into Hugging Face reportedly started on July 11 and continued until July 13. Reuters reported that OpenAI did not immediately realise that its own AI agent was responsible for the attack. The first communication between OpenAI and Hugging Face about the incident happened only on or around July 20, nearly a week after the hacking had ended.
OpenAI publicly revealed on July 21 that one of its AI agents had escaped its testing controls and carried out the break-in at Hugging Face. Several details, including how long the AI agent remained rogue and how late OpenAI discovered its role—are being reported for the first time. Thomas Wolf said Hugging Face is preparing a public timeline explaining the hacking incident but added that he could not comment on what happened inside OpenAI.
Questions over AI safety
In a statement, OpenAI described the hacking incident as unprecedented and said it “marks an important moment for AI safety.” OpenAI also said it is reviewing the incident with outside advisers and plans to publish a technical report later. An OpenAI spokeswoman said there were “several inaccuracies” in Reuters’ reporting but did not explain what those inaccuracies were when asked. Three cybersecurity experts told Reuters that the incident raises fresh questions about OpenAI’s AI safety systems and internal monitoring.
Also read: Americans switch to Chinese AI as cheaper models challenge ChatGPT and Claude despite Trump crackdown
Marley Smith, principal intelligence specialist at the nonprofit World Ethical Data Foundation, questioned how OpenAI missed the AI agent’s actions. Smith said, “Does that mean that they left it unattended and didn’t realize what it was doing? Or maybe they did and didn’t know how to contain it? Both are equally dangerous and alarming”, via Reuters.
How OpenAI tracked the AI agent
Two people familiar with the matter, told Reuters that OpenAI only realised its own AI agent was responsible after Hugging Face published a blog post on July 16 saying it had been hacked by “an autonomous AI agent system”. During the weekend of July 18–19, OpenAI employees reportedly found evidence in internal system logs showing that the AI agent had escaped its testing restrictions.
Four people familiar with OpenAI’s training process told Reuters that the company often runs many AI evaluations at the same time, creating huge amounts of data that staff sometimes struggle to monitor. By the time OpenAI contacted Hugging Face, the company had already informed the FBI about the cyberattack.
Experts want stronger AI rules
Reuters said autonomous AI agents are becoming one of the biggest trends in the AI industry because they can perform tasks independently and work continuously without human supervision. However,giving AI agents more independence also increases the risk of unexpected or harmful behaviour. Reuters reported that advanced AI models are known to sometimes take shortcuts to complete tasks or pass tests.
Jeffrey Ladish of Palisade Research said, “The models lie, they cheat, they hack.” Ladish said the Hugging Face incident should lead to wider discussions about whether leading AI companies are investing enough in security while competing to release more powerful AI systems. Ladish added, “There has to be government oversight, because it won’t happen otherwise”, according to Reuters.