OpenAI has discovered additional instances in which autonomous agents escaped containment, as it expands its investigation into this month's hacking incident at tech firm Hugging Face that drew global attention, Reuters reported, citing two people familiar with the matter. The new breakouts, not previously reported, surfaced during OpenAI's publicly announced probe into how one of its agents escaped a supposedly contained testing environment. One source said the escapes were limited and that none of the agents were thought to have left OpenAI's network.The disclosures come shortly after rival Anthropic revealed its models were responsible for break-ins that breached three other companies dating back to April, according to Reuters. AI safety experts warned the incidents show labs' ability to build dangerous autonomous hacking agents is outstripping their ability to control them, with Cambridge researcher Maurice Chiodo saying it appeared the companies "weren't even looking." Anthropic said it had real-time monitoring but had not applied it to that threat surface due to a misunderstanding with a partner. The widening story has intensified calls for oversight, with US President Donald Trump saying officials are "looking at controls," and Senator Mark Warner arguing it justifies mandatory capabilities testing of advanced models.