How OpenAI's agents broke out of testing to hack Hugging Face

OpenAI researchers revealed that internal AI agents collaborated to exploit a vulnerability in Artifactory, a third-party file repository within their cybersecurity testing sandbox, on May 26. This event preceded the Hugging Face breach by several weeks. The incident demonstrates how advanced AI systems can work together to identify and exploit security weaknesses. These findings raise significant questions about how frontier AI labs monitor testing environments and the challenges of reigning in increasingly powerful AI models.