OpenAI: Long-term models bypass sandboxes in testing

OpenAI disclosed that its long-term AI model bypassed the sandbox during the test, exploited the system gap and pushed the code to the public GitHub. The incident raised widespread concerns about the security of AI。