On 3 September 2026, ChatGPT and Claude both had real outages, within a few hours of each other, and Grok and Gemini were reportedly affected too. At least one contemporaneous report raised a shared cause: a failure inside Microsoft Azure. We read OpenAI’s own written incident report, Anthropic’s status page, and Google Cloud’s own status history directly. None of them names Azure. OpenAI’s own explanation is the opposite of a vendor-side outage. We found no matching entry on Google’s infrastructure status page, though that page may not cover the consumer Gemini app the way OpenAI’s and Anthropic’s pages cover their own products. This is a case where a plausible-sounding shared explanation did not hold up as well as the one report we checked seemed to suggest.
Key takeaways
- ChatGPT and Claude each had a real, separately confirmed outage on 3 September 2026, roughly overlapping in time.
- OpenAI’s own published write-up says the cause was “a routing configuration change” inside its own systems, not an outside vendor failure.
- Anthropic’s status page confirms an incident the same day but states no cause at all.
- Google’s own infrastructure status page shows no matching 3 September entry, though it may not cover the consumer Gemini app directly. One contemporaneous news report says Microsoft itself denied the Azure theory it had raised. We checked one report, not several.
- We could not check Grok’s status directly; that part relies entirely on the one news report we read.
Last reviewed 1 October 2026. We read OpenAI’s incident write-up, Anthropic’s status history, and Google Cloud’s status page directly in a browser on that date, plus one news article published the same week as the event. We did not verify Grok’s status directly, since xAI does not appear to publish a public status-history page we could read the same way.
What actually happened, by each company’s own account
OpenAI. Its own published write-up for the incident (read it here, retrieved 2026-10-01) says: “On Sep 3, 2026, beginning at approximately 7:43 a.m. PDT, users experienced elevated errors while accessing ChatGPT and Codex. A routing configuration change prevented affected requests from reaching our services.” It states the disruption was mitigated by about 8:20 a.m. PDT, and that “as routing was restored, returning traffic overloaded some services, prolonging recovery.” Nothing in OpenAI’s own account mentions Azure, Microsoft, or any outside vendor. The cause it describes is internal: a configuration change OpenAI’s own engineers made.
Anthropic. Claude’s status history lists “Elevated errors for multiple models” on 3 September 2026, 13:26-16:23 UTC (retrieved 2026-10-01), affecting Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5, and says only that “the issue affecting Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5 has been resolved.” No cause is given anywhere on the page. The time window overlaps with OpenAI’s incident (7:43-8:20 a.m. PDT is 14:43-15:20 UTC, inside Anthropic’s stated 13:26-16:23 UTC window), which is consistent with, but does not prove, a shared cause.
Google. We opened Google Cloud’s own public status history (retrieved 2026-10-01) and read every incident listed for the first week of September. The only entry near that date is “Multiple products in us-central1-b are experiencing network service degradation,” dated 1 September 2026, two days before the OpenAI and Claude incidents, affecting Cloud Run, BigQuery, Compute Engine and similar infrastructure products. We found no Google Cloud incident dated 3 September 2026, and no entry mentioning Gemini, Bard, or the consumer AI product by name. This page covers Google Cloud infrastructure, not necessarily the consumer Gemini app, so an absence here does not confirm the Gemini app itself had no problem that day, only that nothing matching showed up on this particular page.
What the press reported, and what it didn’t confirm
We read one contemporaneous report (9to5Google, 3 September 2026, retrieved 2026-10-01) that covered the ChatGPT, Claude and Grok outages as they happened, and raised the Azure theory explicitly, then noted a denial in the same piece: “The underlying problem might have something to do with Microsoft Azure, which is also seeing a spike in outage reports,” followed by “Microsoft says that this is not the case.” The article also says Google Gemini “was not officially confirmed down,” despite a spike in user reports, and describes the outage as recovering by 12:38 p.m. Pacific. We read only this one report, not a cross-section of outlets, so we cannot say every outlet covering the story ran the same Azure theory or the same denial; we can only say this one did, and reported Microsoft’s denial of its own theory in the same article.
So was it one shared cause, or three separate ones?
Based on what we read, the honest answer is: we don’t know, and the companies’ own pages suggest it probably wasn’t one shared cause. OpenAI states its own internal cause directly. Anthropic states no cause, so we cannot say whether its incident was related to OpenAI’s, coincidental, or something else again. Google’s own status page shows nothing matching. The Azure theory, as reported in the one article we read, was raised as a possibility by its reporter, not confirmed by any of the three AI companies, and the same article reports a denial attributed to Microsoft, though we did not find an on-the-record Microsoft statement ourselves beyond that secondhand account. We are not saying the three outages were definitely unrelated. We are saying that the specific claim we set out to check, that one Azure failure explains all of them, has no support in any of the primary sources we read, and a reported contradiction from one secondary source.
Why we checked this at all
“Multiple AI tools down at once” is the kind of story that invites a simple, shareable explanation, and “it was Azure” is simple and shareable. We’ve found this pattern before: in our count of every OpenAI status incident in September 2026, a multi-day Codex problem users described did not match any single officially logged incident. In our look at Anthropic’s own attribution of a different bug to a Microsoft update, the Microsoft page it linked to never actually named Claude. Each time, going to the primary source changed, or at least complicated, the story that was circulating.
What this means for how you read outage news
When several tools break at once, the instinct is to look for one tidy cause, and “a cloud provider went down” is the tidiest one available, because it is simple, shareable, and does not require reading any single company’s own technical write-up. In this case the tidy explanation was raised in the same article that reported it had been denied, which is easy to miss on a skim. The pattern worth taking from this: when a single shared cause is offered for several companies’ outages at once, check whether any of those companies’ own status pages actually says so, rather than assuming the simplest explanation is the correct one. In our experience across the status-page pieces linked above, the companies’ own pages are usually more specific, and sometimes contradict, the shared narrative that circulates afterward.
What we could not verify
- We could not read a public, primary status page for Grok (xAI), so we rely entirely on the news report’s description of Grok’s outage.
- We could not confirm whether Google’s consumer Gemini app had any user-facing problem on 3 September 2026, only that we found no matching entry on Google Cloud’s own infrastructure status page, which may not cover the consumer app the same way OpenAI’s and Anthropic’s pages cover theirs.
- We cannot rule out a shared root cause that none of the three companies has disclosed. Absence of a stated shared cause is not proof there wasn’t one.
- We read one news article’s account of Microsoft’s denial; we did not find a direct, on-the-record Microsoft statement ourselves to quote beyond what that article reports.
How we researched this
On 1 October 2026 we read OpenAI’s published write-up for its 3 September incident, Anthropic’s status history entry for the same date, and the first week of September on Google Cloud’s own status page, all in a browser, recording exact quotes and timestamps. We then read one news article published the week of the event covering the same outages, to see what was reported publicly versus what each company’s own page states. Where we quote a cause, it is copied from the source named. Where we note an absence (no stated cause, no matching incident), that reflects what we found on that specific page on that date; it is not proof nothing happened.
Frequently asked questions
We found no support for that in the three companies’ own status pages we read. OpenAI states an internal cause unrelated to Azure. Anthropic states no cause. Google’s infrastructure status page shows no matching incident, though it may not cover the consumer Gemini app directly. One news report we read says Microsoft itself denied the Azure theory.
Its own write-up says a routing configuration change inside OpenAI’s systems prevented requests from reaching its services, and that recovering traffic then overloaded some services, extending the disruption.
No. Its status page confirms an incident affecting several models on 3 September 2026 but states no cause.
Their stated time windows overlap, which is consistent with a shared cause but does not prove one. Neither company’s own page states a connection to the other.
