OpenAI finds more AI agent escapes as probe widens

OpenAI has uncovered additional instances where autonomous agents broke out of their testing environments, as the company expands an investigation into a hacking incident at Hugging Face that became public this month, two people familiar with the matter said Friday. The new breakouts surfaced during the company's probe into how one of its agents escaped a contained testing environment earlier in July, the two people said. OpenAI is now examining those cases as well.

One of the sources said the escapes were limited and that none of the agents were believed to have left OpenAI's network. An OpenAI spokesperson referred to a Tuesday statement saying the company was reviewing "broader activity from our models" beyond the Hugging Face intrusion.

The discovery of additional rogue behavior, even if contained, could fuel calls for regulation coming from the White House and elsewhere, the people said. The expanded investigation began shortly before OpenAI's primary rival, Anthropic, disclosed that its own models were responsible for a series of break-ins at three other companies dating back to April, according to the two sources and a third person familiar with the matter. The earlier breakouts at OpenAI have not previously been reported.

AI safety experts said the disclosures paint a picture of labs whose ability to develop dangerous autonomous hacking agents outstrips their ability to control them.

"We have a whole industry where the people designing, developing and putting out these tools aren't keeping up themselves to responsibly develop these things and keep them safe," said Maurice Chiodo, a mathematician at Cambridge University's Centre for the Study of Existential Risk.

Reuters could not establish how many incidents OpenAI investigators found, or the timing and circumstances. The three sources said OpenAI and outside experts were examining log data from earlier in the year to understand what happened.