September 2026 AI Service Outage Explained

Three of the world’s largest artificial-intelligence services experienced overlapping outages on September 3, 2026, disrupting OpenAI’s ChatGPT and Codex, Anthropic’s Claude and xAI’s Grok. The near-simultaneous failures prompted speculation that the companies had been hit by a shared cloud-provider problem or coordinated cyberattack, but later reporting indicates that at least some of the services suffered different technical problems. OpenAI’s outage was attributed to a routing error, Anthropic described its Claude disruption as an infrastructure issue, and xAI said Grok went down because of an outage at its Memphis compute center. All three services were restored later that day.

ChatGPT And Codex Experience Elevated Errors

OpenAI reported elevated errors across ChatGPT and Codex on September 3. The incident affected numerous ChatGPT components and several Codex services, producing failed requests and problems accessing features for users. [OpenAI Status]

Reports of ChatGPT problems increased sharply during the morning. Users encountered errors involving conversations, logins, file uploads, image generation and voice features. [The Verge]

Codex was disrupted at the same time. That expanded the impact beyond ordinary ChatGPT conversations to OpenAI’s coding and agentic-development tools. [OpenAI Status]

OpenAI Says Routing Error Caused Outage

Subsequent reporting identified a routing error as the cause of OpenAI’s September 3 disruption. Wired reported that OpenAI attributed its outage to a routing problem rather than a cyberattack or broader failure at a shared cloud provider. [Wired]

The incident began at approximately 10:43 a.m. Eastern time and the most severe disruption was relatively brief. OpenAI nevertheless continued mitigation and monitoring for several hours before formally declaring the incident resolved. [Wired] [OpenAI Status]

A routing error generally means traffic is being directed incorrectly through a service’s infrastructure. OpenAI has not published a detailed technical postmortem explaining the exact configuration or system that caused the routing failure.

OpenAI Fully Resolves September 3 Outage

OpenAI declared the ChatGPT and Codex incident resolved at 4:55 p.m. Eastern time on September 3. The company said the issue causing elevated errors had been fixed. [OpenAI Status]

OpenAI warned that some Codex remote-control users might need to pair their mobile devices again. That was the only continuing user action specifically mentioned when the company closed the incident. [OpenAI Status]

Claude Goes Down

Anthropic experienced a separate Claude outage during the same morning. Its status system reported elevated errors across several models beginning around 9:26 a.m. Eastern time. [Claude Status]

Affected models included Claude Mythos 5.1, Claude Fable 5.1, Claude Opus 5, Mythos and Fable 5, Opus 4.8 and Opus 4.6. Anthropic gradually restored most models while continuing to work on Opus 4.8 and Opus 5. [Claude Status]

Claude.ai, Claude Code and the Claude API were all affected to some degree. [Wired]

Anthropic Blames Infrastructure Issue

Anthropic described its outage as an infrastructure issue. The company said it had identified the source of elevated errors and deployed a fix. [Wired]

Anthropic did not initially release a detailed public explanation of which infrastructure component failed. The status updates focused on affected models and restoration rather than a full technical postmortem. [Claude Status]

Anthropic Resolves Claude Outage

Anthropic declared the larger Claude incident resolved at 12:16 p.m. Eastern time. The company said impact had ended and affected models had returned to normal error rates. [Claude Status]

Anthropic currently lists Claude.ai, Claude Code and the Claude API as operational. [Claude Status]

Grok Goes Down

xAI’s Grok also experienced an outage beginning around 9:30 a.m. Eastern time. The failure affected Grok on the web, mobile apps, X and API infrastructure. [xAI Status]

Users encountered unavailable models and failed requests while xAI investigated. [xAI Status]

xAI Blames Memphis Compute Center

xAI later identified its Memphis compute center as the source of the Grok outage. The company apologized to users for problems caused by the data-center failure. [Engadget]

xAI also apologized to unnamed “compute partners” affected by the Memphis outage. That suggests the failure affected infrastructure being used by organizations beyond Grok itself. [Engadget]

The company did not publicly identify those compute partners. It is therefore speculative to connect any unnamed partner directly to the simultaneous OpenAI or Anthropic outages.

Grok Outage Lasts More Than Three Hours

The Grok web outage lasted approximately three hours and 37 minutes. xAI’s status history lists the September 3 problem as resolved. [xAI Status]

Grok on X, Android and xAI’s API experienced similar disruptions. Those services were also restored later in the day. [xAI Status] [xAI Status]

Were The Three AI Outages Connected?

There is still no evidence that a single failure brought down ChatGPT, Claude and Grok. Later information instead points toward different immediate causes for at least some of the companies. [Wired]

OpenAI reported a routing problem, while xAI identified a failure at its Memphis compute center. Anthropic separately described an infrastructure problem. [Wired] [Engadget]

The timing remains unusual but does not prove a shared cause. Independent technical failures can occur at overlapping times, particularly as AI providers operate increasingly complex and heavily utilized infrastructure.

Microsoft Azure Theory Weakens

Early reporting raised Microsoft Azure as a possible shared point of failure. That theory circulated because major AI companies depend on large cloud and networking providers. [Axios]

Later evidence does not support blaming Azure for the multi-company outage. Wired reported that Azure, Amazon Web Services and Cloudflare did not report corresponding major incidents during the outage period. [Wired]

OpenAI’s own explanation pointing to a routing error also makes a broad Azure failure less likely as the direct cause of its outage. [Wired]

Was The AI Outage A Cyberattack?

There is no evidence that the September 3 outages were caused by a cyberattack. None of OpenAI, Anthropic or xAI has attributed its outage to hackers, a denial-of-service attack or coordinated malicious activity.

The technical explanations that have emerged instead point toward ordinary infrastructure failures. OpenAI reported a routing error and xAI identified a data-center outage. [Wired] [Engadget]

Gemini Was Not Confirmed Down

Some users reported problems with Google Gemini during the same period. Downdetector-style reports led several outlets to initially describe Gemini as part of the wider AI outage.

Google did not log a comparable major Gemini outage in its public cloud incident history. Wired reported that Gemini appeared generally available while the confirmed ChatGPT, Claude and Grok outages were occurring. [Wired]

It is therefore more accurate to describe September 3 as a simultaneous outage involving ChatGPT, Claude and Grok rather than saying every major AI service went offline.

OpenAI Launches GPT-6 Astra The Same Day

The outage occurred on the same day OpenAI unveiled GPT-6 Astra. The unusual timing caused immediate speculation that the service disruption was connected to a major model deployment. [The Verge]

There is no evidence that Astra caused the outage. OpenAI’s reported routing error provides a more specific explanation for the disruption. [Wired]

There is also no evidence that OpenAI intentionally took ChatGPT offline to launch Astra. The company treated the event as an unplanned service failure and issued status updates while engineers mitigated it.

OpenAI Extends WebMCP Challenge

The September 3 outage disrupted participants in OpenAI’s WebMCP Challenge. OpenAI extended the competition deadline by 12 hours because users lost access to relevant tools during the disruption. [OpenAI]

That provides direct evidence that the outage affected developer workflows in addition to ordinary chatbot users.

Separate Overnight ChatGPT Display Outage

OpenAI experienced another separate incident beginning late September 2. Some users saw blank ChatGPT responses even though the underlying system had successfully generated answers. [OpenAI Status]

OpenAI determined that incorrectly configured DDoS protection caused that earlier problem. Automated security systems blocked JavaScript files required by the ChatGPT web interface.

That incident should not be confused with the later September 3 routing outage. They occurred within hours of one another but had different documented causes.

ChatGPT Work Mode Also Goes Down

ChatGPT Work Mode experienced a separate short outage early on September 3. OpenAI began investigating high error rates shortly after midnight and said all affected services had recovered by 12:10 a.m. [OpenAI Status]

There is no evidence the brief Work Mode incident caused the later ChatGPT and Codex outage.

New OpenAI Outage On September 4

OpenAI experienced another service disruption the following morning. Beginning at 7 a.m. Eastern time on September 4, users in the Asia-Pacific region experienced increased errors involving ChatGPT, Work, image generation, file uploads, Voice and Codex Cloud. [OpenAI Status]

OpenAI applied a mitigation at 9:47 a.m. Eastern time. The company then monitored recovery. [OpenAI Status]

The September 4 APAC outage was fully resolved at 10:46 a.m. Eastern time. OpenAI said all affected services had recovered. [OpenAI Status]

OpenAI has not said that the September 4 APAC incident was connected to the September 3 routing outage. They should be treated as separate service events unless the company says otherwise.

Multiple OpenAI Incidents In Early September

OpenAI recorded several service incidents in the opening days of September. New-account creation experienced elevated errors on September 2, while the Responses API and some ChatGPT conversations experienced problems on September 1. [OpenAI Status]

Those incidents varied substantially in scale. A status-page event affecting one feature or subscription tier does not necessarily mean ChatGPT as a whole was unavailable.

Why The September 3 Outage Matters

The incident demonstrated how dependent users have become on AI services that increasingly function as infrastructure rather than novelty chatbots. ChatGPT, Claude and Grok are used for coding, research, writing, customer service, data analysis and automated business tasks.

Coding services were particularly exposed. Codex and Claude Code problems can interrupt work being performed autonomously in the background rather than merely preventing someone from opening a chatbot.

The simultaneous failures also showed that using several competing AI platforms does not guarantee uninterrupted access. Even without a common cause, independent AI providers can become unavailable during overlapping periods.

September 2026 AI Outage Status

The September 3 OpenAI incident is resolved. OpenAI formally closed the ChatGPT and Codex outage at 4:55 p.m. Eastern time. [OpenAI Status]

Anthropic’s September 3 Claude outage is resolved. Claude.ai, Claude Code and the API are currently listed as operational. [Claude Status]

xAI’s September 3 Grok outage is resolved. Grok’s web, mobile, X and API services are currently listed as operational. [xAI Status]

OpenAI’s separate September 4 APAC outage has also been resolved. [OpenAI Status]

There remains no evidence that one shared technical failure caused the September 3 ChatGPT, Claude and Grok outages. The available evidence instead points toward separate infrastructure problems occurring during an unusually narrow window. [Wired]