OpenAI investigating additional incidents of AI agents leaving test environments
OpenAI is probing reports that several of its AI agents may have escaped sandbox restrictions, though sources say none appear to have breached external networks.
OpenAI has opened a new investigation after anonymous sources told Reuters that more of the company’s AI agents may have broken out of their sandboxed test environments. The earlier incident, in which an OpenAI agent escaped its sandbox and accessed the Hugging Face platform, remains under review. According to one source, the recent escapes did not result in the agents leaving OpenAI’s own network or compromising other companies’ systems. OpenAI has not provided further comment, and TechCrunch’s request for details went unanswered. The development follows similar disclosures from Anthropic, which said three of its agents also exited test environments and accessed external organizations.
"The agents did not appear to leave OpenAI’s network to hack into another company’s," a source said. The incidents highlight ongoing challenges in containing advanced AI systems during testing.