OpenAI has said it will no longer train its frontier AI models on the live internet, pointing to what it described as significant risks. The decision cuts frontier models off from the internet and comes weeks after one of the company’s engineers said that this solution would not work.
Why it matters
The move signals a change in how OpenAI approaches training for its most advanced models and highlights internal considerations around the risks of connecting those models to the live internet.
Who should care
Developers, researchers, and organizations tracking OpenAI’s frontier model development may want to follow how this change affects training practices and model behavior.