ChatGPT, Grok, and Claude all went down at the same time

Emma Roth reports that around 11AM ET on Thursday ChatGPT began returning errors, with OpenAI's status page citing elevated errors across ChatGPT and Codex. Logins, file uploads, voice, search, deep research, and image generation were affected. OpenAI said it had applied a mitigation and was monitoring recovery, with performance still degraded. The outage landed as the company teases Astra.

Anthropic's Claude, Claude Code, and API went down around the same time. Staffer CJ Avilla called it an infrastructure issue and a partial outage; Anthropic said it was resolved around 12:15PM ET. Grok on Android, iOS, and web started failing at 9:30AM ET, returning an overloaded-model message, with xAI saying it was working to restore service. Roth is clear that it is not known whether the incidents are related.

Three frontier chatbots failing in a four-hour window is either a shared dependency or a very loud coincidence. The timestamps do not line up cleanly — Grok at 9:30AM ET, ChatGPT at 11AM, Claude recovered by 12:15 — which argues against a single blast radius. It also does not rule out a common upstream: a cloud region, a DNS provider, a certificate, a peering path, or a traffic spike that hit everyone because everyone is now a consumer product with the same busy hour.

Signal Desk's read: treat 'related' as unproven and treat 'concentrated' as proven. A working day in which the three products people actually type into were all degraded is a reliability story the industry has been postponing. Consumer chat is now infrastructure. Infrastructure that cannot publish a shared postmortem will keep producing mornings like this, and each one trains users to keep a second tab open.

The Astra timing is a sideshow until it is not. Launch windows are when you least want Codex and ChatGPT degraded. If this was a load test going wrong, that will leak. If it was unrelated, OpenAI still has to explain why 'elevated errors' lasted long enough to be a news story on a day it is selling a more capable model.

Claude coming back in about an hour, with a named infrastructure cause, is the grown-up version of this incident. Grok's 'overloaded' copy is the least informative. Overloaded by what is the only question that matters, and it is unanswered.

Partial, overlapping outages among AI chatbots have happened before and are usually separate. What is new is the expectation that these products stay up the way email stays up. They do not, yet.