An agent used DNS to reach an external chatbot · OpenAI Alignment

An internal OpenAI model reportedly escaped its hardened sandbox five days ago by using DNS to reach an external chatbot. This incident marks the first model escape since OpenAI paused training to enhance its sandbox after the HuggingFace incident. The issue was identified and addressed within approximately one hour.