A Windows Server Bug Behind the Nearly 24-Hour Microsoft 365 Outage

A Windows Server Bug Behind the Nearly 24-Hour Microsoft 365 Outage

Arkadiy Andrienko

Microsoft said its testing setup missed a Windows Server defect introduced in a July 2026 update, allowing the bug to lurk for more than three months before it brought down parts of Microsoft 365. The outage disrupted services from November 12 to 13, 2026.

Under certain workloads, the defect caused an internal Windows process to compete for resources until it exhausted CPU and memory, leaving servers unable to respond. Microsoft's tests missed the problem because its laboratories mainly used less powerful machines with fewer CPU cores, where the issue was less likely to appear.

The incident escalated after Teams traffic was redirected from an alternative gateway to the primary one shared with Exchange Online, around 15 hours before the outage. About 14% of gateway server capacity was already unavailable because of an unrelated issue from the previous day. More than 80% of the gateway servers stopped responding within two hours, and regional safeguards failed to contain the disruption.

At the moment, full functionality has been restored
At the moment, full functionality has been restored

The outage affected Exchange Online and features in Outlook, Teams, SharePoint, OneDrive, Copilot, and other Microsoft 365 services. Users encountered delays in email delivery and lost access to some features. Engineers prepared a special fix and restarted servers in stages to restore the infrastructure.

Microsoft said the defect was specific to its Microsoft 365 gateway infrastructure and didn't affect broader Windows deployments. The company characterized the incident as a technical failure, with no indication of malicious activity or AI-agent involvement, according to details from the preliminary incident report.

Microsoft acknowledged shortcomings in update testing and resilience measures. Planned steps include testing across a wider range of hardware configurations, improving monitoring, and preventing cascading traffic redistribution between servers. The company plans to publish its final post-incident report by November 20, 2026.

Is testing updates on less powerful hardware enough to prevent another outage?

    About the author
    Comments0