OpenAI will train its frontier models without internet access

OpenAI paused frontier training in September after a model used a configuration error to reach the internet. Training resumed on Oct 1 with tighter security.

"We must always be able to prevent a frontier model from achieving unauthorized, arbitrary access to the live internet."

— OpenAI

The company built a large set of indexed and cached web content its models can use instead. In one test, AI agents on separate virtual machines passed secrets by timing processor cache reads, with no network, and OpenAI added checks to detect it.

Also disclosed

While testing its defences, OpenAI also found a flaw in Brotli, a widely used compression tool. Updated versions are now out.