Three of the most-used AI chatbots in the world broke at roughly the same time on September 3, 2026. Grok went down first at 9:30AM ET, Anthropic's Claude followed at 10:30AM ET, and ChatGPT began returning errors at 11AM ET — a 90-minute cascade that briefly left millions of users without access to the tools they had built into their daily workflows.
OpenAI's status page reported 'elevated errors across ChatGPT and Codex,' with knock-on effects across logins, file uploads, voice mode, search, deep research, and image generation. The company said it had 'applied a mitigation' and was 'monitoring recovery,' but flagged continued 'degraded performance' for hours after. Codex, OpenAI's coding product, was affected alongside the consumer chatbot, meaning developers pushing code and end users asking for recipes were locked out of the same platform simultaneously.
xAI's Grok returned a blunt overload message to users prompting it inside X and across its Android, iOS, and web apps. A notice on the Grok website said xAI was 'working on restoring service as quickly as possible.' No cause was given.
“This model is overloaded right now. Please try again shortly or pick a different model.”— Grok, xAI chatbot error message
Key facts
- 01ChatGPT, Claude, and Grok all experienced outages within a 90-minute window on September 3, 2026, starting with Grok at 9:30AM ET.
- 02OpenAI's status page cited 'elevated errors across ChatGPT and Codex' beginning around 11AM ET, affecting logins, voice mode, file uploads, and image generation.
- 03Anthropic's outage began at 10:30AM ET and narrowed to Opus 5 and Opus 4.8; the company blamed an 'infrastructure issue.'
- 04The outages arrived as OpenAI teases its next model, Astra.
- 05No company publicly linked the three incidents, and none provided a shared root cause.
Anthropic's outage was narrower in scope but hit its most capable models. The company's status page initially listed Mythos/Fable 5.1, Mythos/Fable 5, Opus 5, Opus 4.8, and Opus 4.6 as impacted before narrowing the incident to Opus 5 and Opus 4.8 — the two models most Claude Code and Claude API customers actually depend on for production workloads.
CJ Avilla, an Anthropic technical staff member, said on X that an 'infrastructure issue' caused a partial outage across the company's services. Beyond that, Anthropic offered no timeline.
“We're working on it but don't have an ETA yet”— CJ Avilla, Anthropic technical staff member
None of the three companies has publicly linked the incidents. The obvious question — whether a shared upstream provider, DNS layer, or CDN sits behind more than one of them — went unanswered as of press time. OpenAI, xAI, and Anthropic did not immediately respond to requests for comment.
The timing is awkward for OpenAI, which has been teasing the launch of Astra, its next model. A daylong reliability incident on the existing product line is not the backdrop the company wanted for a launch cycle, and the outage came less than a week after safety researchers raised concerns about Astra's opaque reasoning architecture, which AI Chat Daily covered earlier.
Simultaneous outages across independent AI providers are rare but not unprecedented. What is new is how much of daily work now flows through these three endpoints. When ChatGPT, Claude, and Grok all return errors in the same hour, the fallback for many users is not a fourth chatbot — it is going back to Google search, or to the pre-2023 workflow. That is a data point every enterprise buyer negotiating an AI contract should have in front of them.
The correlation could still be coincidence. Three large distributed systems each running load-balanced inference across dozens of data centers will occasionally have bad mornings, and the fact that Anthropic named an 'infrastructure issue' while OpenAI cited 'elevated errors' suggests different failure modes rather than a shared cause. Neither company has ruled a common vendor in or out.
The lesson for the AI market is that redundancy across model providers is not the same as redundancy across infrastructure. Enterprises that architected their AI stack to fail over from Claude to GPT or Grok discovered on September 3 that all three fallbacks can go dark inside the same 90-minute window. Expect procurement teams to start asking harder questions about which hyperscaler, which CDN, and which identity provider actually sits underneath the label on the invoice.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
