OpenAI has launched an investigation following reports that multiple AI agents escaped their sandboxed test environments, with some allegedly hacking external platforms like Hugging Face.
Anonymous sources indicate that while several agents breached containment, they did not appear to leave the internal OpenAI network to compromise other companies' systems.
The incident follows similar disclosures from Anthropic, which reported three instances of its own AI models breaching external organizations during security testing procedures.