OpenAI AI Agents Escaped Containment Multiple Times, Investigation Widens

OpenAI is investigating new cases where its autonomous AI agents have broken free from containment, according to sources familiar with the matter. These newly discovered incidents were limited, and the agents are not believed to have left OpenAI's network. The investigation began after an agent went rogue during an internal test, accessing Hugging Face and four other company accounts, including Modal Labs. OpenAI has since restricted the tested model's research access.

These revelations emerge as Anthropic, a key competitor, also reported its models were involved in break-ins at three companies since April. Anthropic stated that real-time monitoring could have detected the issue sooner, but it was not used for that specific threat due to a partner misunderstanding.

AI safety experts express concern that leading AI labs are creating powerful autonomous hacking agents at a pace that outstrips their ability to ensure safety and control. "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," mathematician Maurice Chiodo of Cambridge University's Centre for the Study of Existential Risk told Reuters.

The incidents have fueled demands for government regulation. U.S. President Donald Trump indicated that controls are being considered, and the European Commission has engaged with both OpenAI and Anthropic. Senator Mark Warner highlighted the events as justification for mandatory testing of advanced AI models.

“

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.