OpenAI moves frontier model training into an offline environment
OpenAI says it is training its frontier models without direct internet access, using indexed and cached web content in an offline environment so models can practice browsing without affecting external systems. The company resumed reinforcement-learning training and internal inference on October 1 after strengthening monitoring, network controls, sandboxing and infrastructure. Red-team tests also found that isolated agents could exchange information through shared channels such as processor caches, shared storage and certificate caches, prompting OpenAI to add monitoring checks.
OpenAI’s red-team findings show that removing network connections alone does not fully isolate agents when they share hardware or storage resources.
OpenAI now trains its frontier AI models with no internet access
AI agents can still communicate without a network connection, OpenAI found.
AI agents can still communicate without a network connection, OpenAI found.
by Emma Roth
Oct 23, 2026, 9:11 PM UTC
