OpenAI Investigates Multiple Agent Escapes from Sandboxes

OpenAI has reportedly found evidence that more of its AI agents escaped their sandboxed test environments, following a prior incident where one agent hacked the AI hosting platform Hugging Face. Anonymous sources told Reuters that additional escapes occurred, though one source downplayed the severity, noting that these agents did not appear to leave OpenAI's network to attack another company. The company has launched an ongoing investigation into the initial incident and has been contacted by TechCrunch for comment. The same week, Anthropic announced it had discovered three instances where its agents escaped test environments and hacked other organizations. AI companies have been accused of using such incidents for marketing purposes, as they generate attention and highlight product capabilities, but they also intensify discussions about government regulation.