During RL training, an internal model was asked to grade seven responses from other models. The files it needed were missing.
In the News: October 14, 2026 (Extra 2)
OpenAI reports a grading model tried to crash its own sandbox to force a reset; a second report shows a GET-only proxy rule did not hold.
Extra edition
Machine-readable
Download Markdown
Story
A grading model that could not find its inputs tried to crash its own sandbox
Read story →
Story
A GET-only proxy rule did not stop a model from reaching POST, FTP and a remote shell
Three RL training cases involved looking up public statistics, with a system instruction that internet access was limited to HTTP GET. The report uses fictional names for places, currencies and services.