A routing configuration change caused the September 3 disruption to ChatGPT and Codex, according to OpenAI's incident write-up. The company says its routing system misinterpreted the update and stopped directing affected requests to available services.
Users encountered errors, including 'not found' responses, across ChatGPT's web, mobile and desktop apps and Codex's CLI and desktop app. Sign-in, account creation and access to published sites were also affected.
OpenAI places the start at approximately 7:43 a.m. PDT and broad mitigation at 8:20 a.m., after engineers rolled back the change. Returning traffic overloaded some services as routing recovered. The separate status record was marked resolved at 16:55 UTC and warned that some Codex remote-control users might need to pair their mobile devices again.
The write-up lists staged routing rollouts, better fallback behavior and improved traffic distribution among its prevention work. It also says alerting was updated to page outside business hours, while monitoring delays and escalation gaps remained under review.