OpenAI has uncovered additional cases in which its autonomous AI agents breached internal containment measures as it broadened its investigation into the hacking incident at tech firm Hugging Face that drew global attention earlier this month, two people familiar with the matter said Friday.
The newly identified incidents surfaced during the company's review of how one of its AI agents escaped what was intended to be a controlled testing environment this month, the sources said. OpenAI is now investigating those cases as part of the expanded probe.
One of the sources said the breaches were limited in scope and that there is no indication any of the AI agents escaped OpenAI's internal network.
An OpenAI spokesperson referred to a statement issued Tuesday, saying the company is reviewing "broader activity from our models" alongside its investigation into the Hugging Face incident.
The discovery of additional containment breaches, even if limited, is likely to intensify calls for tighter AI regulation from the White House and other policymakers.
OpenAI expanded its investigation shortly before its chief rival, Anthropic, disclosed that its own AI models were linked to a series of break-ins that resulted in breaches at three other companies dating back to April, according to the two sources and a third person familiar with the matter.
The newly uncovered historical containment breaches at OpenAI have not previously been reported.
AI safety experts said the new disclosures paint a picture of a group of cutting-edge labs whose ability to develop dangerous autonomous hacking agents outpaces their ability to keep them under control.
"We have a whole industry where the people designing, developing and deploying these tools aren't keeping pace with the responsibility to develop them safely and keep them secure," said Maurice Chiodo, a mathematician at Cambridge University's Centre for the Study of Existential Risk.
Reuters could not establish exactly how many incidents OpenAI investigators found or the timing or circumstances under which they occurred.
The three sources said OpenAI and outside experts were examining log data from earlier this year in a bid to understand what took place.
OpenAI first launched the investigation following the early July intrusion at Hugging Face, where one of its AI agents went haywire for days inside another company's network in a botched effort to cheat on an internal test.
As part of that hacking spree, OpenAI said four accounts at four other companies were also compromised. One of those companies was New York-based Modal, company officials said.
Chiodo said his concerns were heightened by indications that neither OpenAI nor Anthropic was monitoring the agents as they went rogue.
Reuters previously reported that OpenAI realized its agent had broken into Hugging Face only after the company contained the hack, contacted the FBI and publicly disclosed the intrusion.
OpenAI has said the Reuters account contained inaccuracies but has not responded to questions about what those inaccuracies were.
In its Thursday statement disclosing how its own AI agents hacked victims online, Anthropic suggested it had not been monitoring them in real time, saying "real-time monitoring of the evaluation logs would have helped to surface the problem sooner."
Chiodo said that pointed to a lack of proper oversight.
"It seems like they weren't even looking," Chiodo said.
Anthropic said that while it did have real-time monitoring in place, it had not been used "for this threat surface" because of a misunderstanding between the AI company and a partner.
The rapidly widening scope of the runaway AI agents story has already heightened pressure from lawmakers and officials across the United States and Europe to push for new government oversight of the labs whose models power them.
"We're looking at controls," President Donald Trump told reporters Thursday.
On Friday, the European Commission said it held talks with OpenAI and Anthropic about the hacking incidents.
Sen. Mark Warner of Virginia, the top Democrat on the Senate Intelligence Committee, said Friday that the Anthropic incident "tells me that legislatively we're correct to require mandatory capabilities testing of these advanced models."
