Frontier Models

OpenAI details securing its frontier model training environment

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