ChatGPT, Claude and Grok suffer simultaneous outages as GPT-6 Astra rumors intensify

Major artificial intelligence platforms including ChatGPT, Claude and Grok experienced widespread service disruptions on Thursday, September 3, 2026, leaving users unable to access some of the most widely used AI tools. The outages affected multiple services and came at a particularly notable time, as speculation surrounding OpenAI's next-generation AI model, reportedly codenamed Astra, intensified.

OpenAI confirmed that ChatGPT and Codex were experiencing elevated error rates. Its status page showed the incident as an ongoing issue affecting ChatGPT and Codex, while users reported problems across the web, mobile and desktop versions of the service.

Anthropic's Claude was also affected by elevated errors across several models. The company's status page said it had identified the cause of the problem and was working on a fix. The affected models initially included Mythos/Fable 5.1, Mythos/Fable 5, Opus 5, Opus 4.8 and Opus 4.6.


Grok, the AI chatbot operated by xAI, was experiencing an outage at roughly the same time. xAI's status page said Grok was experiencing issues and that the company was working to restore the service. The incident affected Grok on the web as well as related services.

The simultaneous disruptions quickly attracted attention because ChatGPT, Claude and Grok are operated by separate AI companies. The Verge reported that the outages began around the same period, with ChatGPT experiencing problems with functions including voice, search, deep research, image generation, logins and file uploads.

The reason behind the simultaneous outages was not immediately clear. Reports raised questions about whether the incidents could have been connected through shared infrastructure, although there was no confirmation that the companies' problems had a common cause. An earlier report noted that Microsoft Azure was also seeing a spike in outage reports and that several of the affected AI services use cloud infrastructure from multiple providers.

Google's Gemini was also reported to have problems for some developers and API users. Google AI Studio's status information indicated that the Gemini API was experiencing issues involving recently created API keys, including keys used through OpenAI-compatible libraries.

The disruptions highlighted how heavily other websites, applications and developer tools now depend on AI platforms. When a major model provider experiences an outage, the impact can extend beyond users directly visiting an AI chatbot and reach products that use its API in the background.

The ChatGPT outage was particularly disruptive for developers because Codex was also affected. Codex is OpenAI's programming-focused tool, meaning that the incident potentially affected people using AI assistance for software development as well as those using ChatGPT for general tasks.

OpenAI's status page subsequently showed that the company had restored affected services in at least some incidents during September 3. OpenAI's incident history also shows that individual availability can vary by subscription tier, model and specific feature, meaning that users may experience different levels of disruption even during the same incident.

ChatGPT outage coincides with GPT-6 Astra speculation

The timing of the ChatGPT outage generated additional speculation because OpenAI had been teasing what appeared to be an upcoming announcement.

On September 3, OpenAI posted on X: "The stars are almost aligned."


ChatGPT's official account posted the same message, prompting users to speculate that the reference to "stars" could be connected to Astra, the rumored codename for OpenAI's next major model.

The speculation surrounding Astra had already been building before the outage. Online reports have claimed that an OpenAI API probe returned unusual behavior when tested against the identifier "gpt-6-astra," fueling speculation that the model may have been staged internally. However, such reports do not establish that OpenAI has officially named its next model GPT-6 Astra or that a public launch is imminent.

The name Astra itself has become closely associated with OpenAI's next-generation model in recent discussions, but the precise relationship between Astra and GPT-6 remains unconfirmed in public reporting. Reports have described Astra as a possible successor to the current generation, while noting that OpenAI has not necessarily confirmed the commercial branding or release date.

Were the AI outages connected?

There was no immediate evidence establishing that the ChatGPT, Claude and Grok outages were caused by the same technical problem.

The fact that several major AI services experienced problems within a similar period nevertheless made the incident unusual. ChatGPT, Claude and Grok are operated by different companies, although modern AI services rely on extensive networks of cloud computing, content delivery, networking, authentication and other infrastructure.

The outages therefore demonstrated both the growing importance of AI services and the complexity of the infrastructure supporting them.

For businesses that have incorporated AI models into everyday workflows, even a short interruption can affect customer-service systems, coding tools, research workflows and applications built around model APIs.

Services began recovering for some users

The outages did not affect every user in exactly the same way. As per reports, services began returning after roughly half an hour, while some others continued to experience errors or degraded performance.

Anthropic's status updates showed that the company moved from investigating the Claude problems to identifying the cause and implementing a fix. The company subsequently reported that some affected models had returned to normal levels, although the status of individual models could differ during the recovery process.

OpenAI likewise reported recovery for affected incidents in its status history. The company's records show that ChatGPT and other OpenAI services can experience incidents affecting specific components rather than the entire platform, making the user experience dependent on which feature or model is being accessed.

As services returned, users were advised to refresh pages or retry their requests. Because recovery was not necessarily simultaneous for everyone, some users could continue seeing errors even after a provider began reporting that its systems were recovering.