September 4, 2026, (Inside AI) — A cascading authentication failure inside Microsoft's Exchange Online infrastructure knocked several frontier AI platforms offline this week. The disruption hit Anthropic's Claude, OpenAI's ChatGPT, xAI's Grok, and Google's Gemini, according to outage reports and official status pages.
The incident began on August 31, when Microsoft first acknowledged problems with Exchange Online. The company later tracked the event under reference MO1465074. Downdetector logged thousands of user reports at the peak of the outage, though those figures only capture submitted complaints, not the total number affected.
Claude's status page confirmed that its Opus 4.8 and Opus 5 models were impacted. Issues affecting Claude Mythos 5.1, Fable 5.1, and Opus 5 were later resolved. ChatGPT's Work Mode also experienced a lengthy stretch of elevated latency on the same day.
Authentication Glitch Rippled Across Microsoft 365
Microsoft traced the root cause to a core authentication configuration problem. This configuration is used by several internal services within the Exchange Online infrastructure. In simpler terms, the fault prevented authentication components from deploying correctly across part of the system. Users struggled to sign in and access their services.
Engineers responded by conducting manual tests on individual servers. They then expanded remedial measures across the affected infrastructure gradually. The company acknowledged the fix would take time and warned users to expect varying levels of disruption and email delays during recovery.
The blast radius extended far beyond email. While Outlook received most attention, many other services failed too. The outage affected OneDrive for Business, SharePoint Online, Microsoft Teams, Microsoft Purview, and Microsoft Defender XDR. Notably, Microsoft's own Copilot AI assistant was among the affected tools.
Rare Simultaneous Failure for Frontier Models
It is rare to see all four affected frontier models experiencing interruptions at the same time, even if they do go down occasionally. Claude has been doing well, boasting a 99.4% uptime for its services over the past 90 days. Its last notable hiccup was a three-hour partial outage on August 24.
OpenAI has reported an impressive 99.63% uptime for ChatGPT and a perfect 100% for ChatGPT Codex during that same timeframe. However, the August 31 incident shows how dependent third-party AI tools can be on underlying cloud infrastructure.
Microsoft has confirmed the outage had been fully mitigated. The company said all affected services were restored. It noted the final Outlook impact occurred in less than 1% of its infrastructure. Microsoft added that it would continue monitoring telemetry to ensure services remained healthy going forward.
The simultaneous disruption raises questions about concentration risk in AI infrastructure. When a single authentication layer fails, multiple competing AI platforms can go dark at once. That is a lesson enterprise customers are unlikely to forget.
