Rare Outage Hit Four Major AI Models All at Once

On Thursday morning, the cloud-based AI services of four of the biggest companies — OpenAI, Anthropic, xAI and Google — experienced noticeable disruptions almost simultaneously. As Ars Technica notes, a simultaneous outage across all four market leaders is practically unheard of.

Anthropic was the first to report trouble: at 9:23 a.m. (Eastern Time) the company logged a “partial outage” with an elevated number of errors when calling the Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5 models. The cause was found roughly fifteen minutes later, and the problem was fixed by 12:16. A little after noon, it separately noted brief errors on requests to Claude Sonnet 5.

At 10:43, OpenAI reported an “elevated number of errors” and “degraded performance” in ChatGPT and Codex. About half an hour later the company rolled out a temporary fix, and by 12:55 the incident was marked as resolved.

xAI's Grok was still showing users a message about the trouble at the time of publication, along with a promise to restore service as soon as possible. According to DownDetector, complaints about the service jumped from fewer than a dozen around nine in the morning to 1,365 by 9:45, then dropped to 273.

Google did not publicly acknowledge any problems with Gemini, but complaints on DownDetector rose from 23 around 10:30 to 412 just after eleven. The monitoring service StatusGator also recorded a “likely outage” of the Gemini API between 10:45 and 11:15, after which the status returned to normal.

Other major internet services — Amazon Web Services, Microsoft Azure and Cloudflare — reported no serious problems, though complaints about them also ticked up somewhat that morning.

Cutting-edge models go down individually from time to time, but for all four to fall within one short window is an exceptional case. Claude claims 99.4% availability over the past 90 days; the previous similar three-hour “partial outage” fell on August 24. OpenAI reports 99.63% availability for ChatGPT and 100% for Codex over the same period.

The reasons for the coincidence remain unclear. The comments on the piece offer a theory attributed to an OpenAI employee named Thibault: a “domino effect” may be at play — many customers keep backup providers, and when one service fails the load shifts onto the others, spreading like a chain reaction.

Source: Ars Technica