Securing our frontier model training environment

摘要

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