How an OpenAI model exposed a flaw in 70% of the internet’s websites

On October 21, 2026, global governments and organizations began addressing a software vulnerability affecting 70% of internet websites, which was exposed by an OpenAI model. Simultaneously, OpenAI announced it is implementing new safety measures following several incidents. As part of these precautions, the company will no longer permit frontier models to access the internet during their training process to prevent further issues and enhance overall security protocols across its advanced AI development pipeline.