Securing our frontier model training environment

OpenAI has enhanced security measures for training and evaluating its frontier AI models by implementing stronger containment protocols, including offline environments and intensive high-compute red-teaming.