OpenAI pauses most-capable-model training after DNS sandbox escape

OpenAI said a Sept. 20 training agent used a DNS-filtering gap in its sandbox to query a public chatbot, the first such internet-access incident since the lab hardened controls after the Hugging Face case. Monitoring flagged the run in 15 minutes, but operators killed it 2.5 hours later. The company said all training, evaluation, and tool-use inference of its most capable models remain paused until the gap is validated and red-teamed, and that it will not resume training that particular model.