Microsoft 365 falls for five hours after a maintenance bug cut West US off the network

On July 23 2026, Microsoft 365 and a broad slice of Azure went down for nearly five hours, from 10:44 a.m. to 3:41 p.m. ET, in incident MO1437424. Teams, SharePoint Online, OneDrive, the Admin Center, Power Automate, Copilot Chat and Loop all degraded, alongside Azure App Service, Application Gateway, Cosmos DB, AKS, ExpressRoute, Virtual WAN and more.

The root cause was a decision encoded in automation. A software layer translates human maintenance requests into machine-readable instructions, and a bug in that conversion incorrectly marked additional network devices as part of the maintenance scope. The result removed IP routes from more devices than were authorized, severing a West US datacenter from the WAN.

The failure survived Microsoft's own safety gate: the check evaluated the buggy conversion's output rather than the engineers' original intent, so it validated the wrong scope and passed. Users filed 2,403 Downdetector reports — 83 times the baseline of 29 — and at least 19 downstream SaaS services fell with cascading outages, 16 of which were still open after Microsoft announced recovery.

The incident capped a poor quarter: Microsoft 365 delivered 99.526% uptime in Q1 2026, the lowest since reporting began in 2013. The most telling gap was not the bug but the safety check, which verified what the machine produced instead of what the human meant — an automation failure that no checklist caught.