OpenAI details containment upgrades for frontier model training
OpenAI published a primary post explaining steps it has taken to strengthen containment during training and evaluation of frontier models. The announcement highlights measures including offline (air-gapped) environments and use of high-compute red-teaming to reduce risk during development.