Some messages sent to Outlook.com mailboxes that had been inactive for more than 30 days expired during the Microsoft 365 outage and never reached their recipients. Microsoft confirmed the issue in its final incident report, published on November 20; it did not specify how many emails or users were affected.
The incident began on November 12 and affected customers in different regions and services. The final report, according to 365Fanatic's summary, places the start of the impact at 16:20 UTC and its end at 16:10 UTC the following day. Microsoft declared the incident resolved on November 14 at 04:15 UTC, after a period of monitoring and clearing the mail queue.
Business emails delivered, exceptions on Outlook.com
Exchange Online could not process accumulated messages normally while the servers were failing. To allow more time for delivery retries, Microsoft temporarily extended the window from 24 to 48 hours before generating a non-delivery notice. The company says all queued business emails were eventually delivered and that it found no corporate messages that were permanently lost.
The situation was different for certain consumer Outlook.com mailboxes. Pending messages addressed to accounts that had been inactive for more than 30 days expired, and senders received non-delivery notifications. Computer Hoy reports that Microsoft did not disclose the number of affected messages or how many people received those notices.
365Fanatic summarizes the report by saying that customers' stored data was not affected. That statement coexists with the exception of expired messages: the loss refers to emails that did not reach their destination, not to a general alteration of data stored in accounts.
A July fix triggered the failure
The report attributes the outage to the combination of four factors: a latent defect in connection processing, reduced server capacity, redirected Teams traffic, and a spike in activity in the United States. None of these factors alone would have caused the same impact, but together they caused the network gateway servers to stop responding.
The defect stemmed from a change prepared for Microsoft 365 gateway servers. The rollout began in July 2026 and was gradually extended worldwide over more than three months without incident. IT Pro reports that Charles Lamanna, Microsoft's executive vice president for Copilot, Agents, and Platform, described the issue as a “latent defect” and explained that the fix had been prepared specifically for those servers, not for Windows as a whole.
“In less than two hours, the previously unknown defect was encountered across our fleet and rendered them unresponsive. At its lowest, more than 80% of servers were unresponsive.”
— Charles Lamanna, Microsoft's executive vice president for Copilot, Agents, and Platform
The technical cause was lock contention: tasks within a Windows process waited on one another until they consumed CPU and memory, and the machines stopped processing requests. The effect was worse on newer servers with more cores, because they could receive more concurrent requests. Standard load tests did not reproduce the combination of hardware and traffic that occurred during the outage; they were run on older machines with a different workload.
That weakness was compounded by the fact that almost 14% of gateway capacity was out of service for routine operations related to deploying a service version. Microsoft said there was enough spare capacity and that this reduction alone would not have caused the incident.
Teams traffic rerouting extended the cascade
Microsoft maintains two independent gateway systems to isolate services and keep them running when problems occur. On November 12, at 01:07 UTC, a routing issue in the US East region caused timeouts in Teams. The service automatically switched to the gateway that also supported Exchange Online, shifting load and changing the traffic pattern.
The infrastructure continued to operate for about 16 hours. The problem became visible when North American traffic approached its daily peak. The first widespread failures began around 16:00 UTC; the Exchange Online alert came at 16:20. Microsoft confirmed issues with several access methods, including Outlook, Outlook on the web, Exchange ActiveSync, and the messaging and REST APIs.
When some servers failed, the traffic management system shifted requests to healthy units. But those units received more load and were exposed to the same defect. The process repeated across regions: in less than two hours, a large proportion of North American servers were experiencing problems and, at the worst point, more than 80% were unresponsive. The rerouted traffic eventually affected servers in every region.
Recovery was not immediate either. Servers that came back online were overwhelmed by waiting requests and could fail again. In addition, the defect had been rolling out for months, so stopping an update or rolling back a recent change was not enough to remove it from the entire fleet.
An outage broader than Outlook
Exchange Online was the most affected service. Many users could not access email through Outlook, the web, the iOS and Android apps, Exchange ActiveSync, or the APIs. Searches, Exchange integrations, and message delivery were also degraded. The level of impact varied by region, connection method, and timing; some users had partial or intermittent access.
The list of affected services included Teams, Copilot, and SharePoint, as well as OneDrive, Defender, Purview, Fabric, Intune, Planner, Power Apps, To Do, Power BI, and Universal Print. Teams, SharePoint, and OneDrive were affected mainly at first and recovered sooner, after their traffic was moved to another gateway. Features that depended on Exchange data—such as search, calendar, presence, and people—continued to experience problems for longer.
Security and administration features also failed. In Defender for Office, Zero-hour Auto Purge (ZAP), which removes malicious messages after delivery when a threat is detected, was delayed. Some users had problems with Safe Links, which analyzes suspicious links, and administrators had difficulty releasing or deleting quarantined messages, managing allow and block lists, and carrying out remediation actions.
In Purview, the portal could be unavailable, and certain data loss prevention, audit, eDiscovery, and insider risk management tasks were delayed. The report also notes incidents in Copilot, where many requests failed and applications would not start; in Teams, affecting calendar, presence, Teams Rooms, and desk phones; and in Universal Print, where print jobs could not be created or retrieved.
Recovery and outstanding changes
Microsoft began applying a corrective update at 07:20 UTC on November 13. The change reverted the defective connection management setting and was rolled out in groups, prioritizing regions that were at or approaching peak hours. The company also removed groups of capacity units from traffic, restarted the affected servers, and brought them back online after the fix.
The service began to stabilize at 10:00 UTC. According to the timeline reported by 365Fanatic, Asia-Pacific recovered around 11:00 and Europe at 11:35. By 13:00, recovery was continuing in all regions and mail queues were shrinking; webmail access returned to expected levels at 14:17. The active impact ended at 16:10 UTC.
Measures already completed include incorporating the fix into the Windows image used on the servers, adding capacity, improving traffic and connection monitoring, and establishing temporary limits to prevent the network from automatically distributing load in a way that could spread the failure. Microsoft also extended the email retry window and returned servers that were undergoing maintenance to service.
The outstanding plan extends through 2027, and several tasks do not yet have a date. Microsoft plans to limit Teams rerouting to the affected region, separate gateway capacity by workload type, and reduce the size of failure zones. It also wants to improve the gradual reintroduction of servers, control and retry mechanisms, testing with higher-core hardware and representative traffic, and detection of prolonged capacity losses.
The company also plans to shorten the time to the first customer notification, improve diagnostics and auditing of Windows changes, and speed up recovery from failures that affect the entire fleet. The report rules out a cyberattack and found no evidence of malicious activity; 365Fanatic adds that Microsoft said the change that caused the issue did not come from artificial intelligence tools. The company says administrators do not need to make any changes in their tenant.
Sources: Computer Hoy, 365Fanatic, IT Pro