Frontier training containment tightened with offline environments
OpenAI describes stronger containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.}
OpenAISafety
“Securing our frontier model training environment”