Codex went down for 56 minutes on 25 September 2026, with ChatGPT also listed as affected. OpenAI's own status page says the fault was internal and its five updates show the company working through root cause to a mitigation.
Started
22:58 UTC
Duration
Lasted 56m
Source
IsDown
Next time
OpenAI is degraded. Is your product working?
An incident on their side does not always mean an outage on yours, and the only way to know is to already be watching your own endpoints. Set that up in under a minute and you will be notified the next time your own site is affected. And there will be a next time.
50 monitors for your own websites. No card.
Affected components: ChatGPT, Codex
What happened?
OpenAI's status page opened this incident at 22:58 UTC under the title "Issues with Codex", saying users were experiencing a Codex outage and that an internal issue had been identified. At 23:03 the wording shifted to "elevated errors for the impacted services", which is how OpenAI described the ChatGPT component also being listed as affected. By 23:19 OpenAI posted a workaround: logging in via API key would unblock access while the main fix was still in progress. At 23:34 they said they had found the root cause and were moving towards mitigation, and at 23:45 they said the mitigation was applied and they were monitoring recovery. No update after that was posted, and the incident is marked resolved with a total duration of 56 minutes. That is a single incident touching two components, Codex and ChatGPT, out of OpenAI's full component list. No user reports were filed during this window, so there is no outside account of what the outage looked like from a Codex session or a ChatGPT chat while it was happening. The whole account of what broke comes from OpenAI's own five updates, nothing more. OpenAI has filed 96 incidents on this service in the past 90 days, with a median duration of 1 hour 52 minutes, which puts this 56 minute incident on the shorter side of its own recent history.
Learning
OpenAI's status page describes this as an internal issue inside Codex that also spilled into ChatGPT, which points to a shared backend layer rather than two separate client-side bugs. If you were only watching your own Codex integration endpoint from outside, you might have seen nothing unusual, because the fault here sat inside OpenAI's infrastructure rather than at the edge your checks would normally hit. That is exactly why external, outside-in monitoring and a vendor's own incident log can tell different, equally correct stories: your probe only sees whether a request completes, not whether the completion took a route through a broken internal service. The workaround OpenAI offered, logging in via API key, is itself a clue that the normal login path was part of what broke, something a simple uptime check against an API endpoint would not expose. With no user reports in this window and only OpenAI's own account to go on, there is no independent confirmation of exactly how the errors presented to end users. Your monitoring covers your half, UptimeRobot now watches the provider's half too, so next time you know which side broke without guessing.
