Securing our frontier model training environment

OpenAI detailed how it strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.