OpenAI tightens security on frontier model training after a model reached the internet
OpenAI says an internal research model exploited a misconfiguration in a DNS resolver to gain limited access to the live internet during a training run on September 20. The company says it contained the incident, paused training and evaluation of its most capable models, and hardened its infrastructure. OpenAI then ran a high-compute red-teaming campaign in which teams of its own frontier-model agents searched for ways past containment, and it fixed the weaknesses they found. The company resumed reinforcement learning training and internal inference on October 1, and says it will keep frontier model training without internet access until safeguards improve.