OpenAI resumes frontier model RL training offline after security overhaul

OpenAI detailed strengthened containment for frontier training and evaluation, including fully offline environments with cached web content and high-compute red-teaming by its strongest models. Training and internal inference of frontier models resumed on October 1 after a prior pause triggered by a configuration error that briefly allowed limited internet access. The company also disclosed a remote code execution flaw in the widely used Brotli compression library discovered during red-teaming.