OpenAI's own advice during its October 9 Android outage: use ChatGPT on the web
If you opened ChatGPT on an Android phone on the evening of October 9 and got "This content is unavailable in this version of the app" where an answer should have been, the fault was OpenAI's. Its suggested workaround was to stop using the app: "Users can use ChatGPT on the web while we work to restore the experience for older app versions."
OpenAI's own card for its plugin management documentation, the admin area the second incident affected. Image: OpenAI
That wording comes from the first update on the incident, which OpenAI logged as major impact. OpenAI's status page records it as opening at 21:00:39Z on October 9 and closing at 00:26:44Z on October 10 with "All impacted services have now fully recovered", a run of three hours and 26 minutes. The figures here come from the status page's public incident feed, read at 21:42 UTC on October 10.
OpenAI said a fix was "rolling out gradually in the latest Android version", which left anyone on an older build with the browser or nothing. No cause was published, and the incident carries only two updates across its whole life.
A second major incident locked admins out of plugin settings
Forty-eight minutes into the Android fault, OpenAI opened a second incident of major impact. Workspace administrators could not "view and manage plugin availability and settings", and OpenAI said some users "may also see plugins as unavailable". That one ran from 21:48:34Z to 23:26:59Z on October 9, one hour and 38 minutes, with a recovery note at 22:52:35Z.
The two overlapped completely. From 21:48:34Z until the admin fault cleared at 23:26:59Z both were open, and the Android fault then ran on alone for another hour before OpenAI closed it. OpenAI has published no cause for either and no figure for how many people were affected by either.
A third incident was still open on October 10
At 21:42 UTC on October 10, the status page listed the two major incidents as resolved and a third, minor one as still unresolved. "Delayed Costs data in the Compliance API" opened at 17:23:18Z on October 9, more than three hours before the Android fault, and its most recent update at that moment was timestamped 22:03:59Z on October 9.
In it OpenAI says it has "implemented a mitigation, and delayed Costs data is now processing through the Compliance API", but that "some customers may continue to see incomplete Costs data while the backlog catches up". The company expects "this catch-up process to continue through Sunday, October 11". ChatGPT usage and API Platform requests are unaffected, and the overall status at that read was "Partial System Degradation" at minor severity.
Costs data is an administrator's billing view rather than anything a ChatGPT user touches. OpenAI graded it minor while it graded the Android rendering fault major.
September 29 was longer, and it reached the API
OpenAI's run of incidents this autumn has been uneven rather than constant. The September 29 outage that ran five hours and 22 minutes across 30 components was far longer and far broader than anything on October 9, and neither of the two major incidents here touched the API.
What they did touch was the two groups least able to work around a fault: people on a phone running an Android build OpenAI had not yet updated, and the administrators who decide what their colleagues are allowed to connect.
