
Microsoft now blames a code change it shipped in July. That change took Outlook offline for most of two working days. The defect sat inside its Microsoft 365 networking gateway servers for about four months and did nothing at all.
Charles Lamanna, Microsoft’s executive vice president for Copilot, agents and platform, posted an abridged account on X on Monday, Nov. 16. The full preliminary report went to administrators through the Microsoft 365 admin center the day before, The Next Web reported.
It fills a gap we flagged when the outage itself eased on Nov. 13. At the time Microsoft would say only that something went wrong during a network capacity upgrade.
The defect shipped in July and did nothing for four months
“First, a latent defect caused servers to become overwhelmed and unresponsive under certain load conditions,” Lamanna wrote, Fox Business reported. The fault was lock contention in an internal process on the gateway servers.
Lock contention happens when too many threads queue for the same resource. Here the queue ate the server’s CPU and memory until it stopped answering at all.
The change came from a fix written for those gateway servers alone. Lamanna said the defect does not reach broader Windows deployments. It went out through Microsoft’s safe deployment process from July 2026 and saturated globally over several weeks, ITPro reported. It then ran worldwide for more than three months with nothing reported against it.
Microsoft’s test machines were weaker than its real ones
Load testing never caught it. Lamanna said the labs ran a different traffic and hardware mix from production. Contention gets worse on machines with more cores and newer silicon, he said. The tests leaned on lower-core hardware.
That is the finding worth sitting with. The test did not fail to run. It ran on equipment too slow to produce the fault. So the safe deployment process did its job and still let the defect through.
Two routine events switched the bug on
Microsoft runs two independent gateway systems so that services can move between them when one struggles. At 01:07 UTC on Nov. 12, Teams moved to the primary gateway. That gateway also carries Exchange Online. Lamanna called that normal operations.
Separately, an unrelated problem the previous day had taken nearly 14% of the gateway’s server capacity out of service. Microsoft said the rest had enough headroom, so that shortfall did no damage on its own.
Then US peak traffic arrived. More requests landed on fewer servers, the queues built, and the machines stopped answering. Microsoft logged its first support case at 16:33 UTC. By our arithmetic on the report’s own timestamps, that is 15 hours and 26 minutes after the Teams failover.
Health checks spread the failure instead of containing it
When a server stops answering, Microsoft’s traffic management sends its users to a healthy endpoint. That design protects customers from an isolated hardware or software fault.
Here it did the opposite. Every redirect added load to the next server, which then met the same defect. The pool of working machines shrank with each step.
Lamanna said the defect crossed the fleet in under two hours. “At its lowest, more than 80% of servers were unresponsive,” he wrote.
A four-month-old change was beyond the usual rollback
Age made the fix hard. The defect came from a months-old update, not a live deployment. “Our normal rollback, flighting, and feature flags were unable to quickly revert the change,” Lamanna said.
Engineers wrote an update that returned the code path to a known good state. They applied it to one group of servers after another. Machines that came back met the traffic queued behind them and fell over again, so the work went slowly.
Teams, OneDrive, SharePoint and the Office apps moved off the damaged gateways and recovered earlier. Exchange Online could not move, for reasons of scale, which is why Outlook stayed dark longest, The Next Web reported.
Where it stands, and what customers can still ask for
Microsoft started applying the fix at 07:20 UTC on Nov. 13. Its final customer update went out at 04:15 UTC on Nov. 14. That is 35 hours and 42 minutes from the first support case, by our count, longer than the near-24-hour recovery Microsoft described at the time.
Both service health incidents are now closed.
Business customers who lost email can ask for a service credit under Microsoft’s Service Level Agreement for Online Services. It pays a percentage of the month’s fee for the affected service. Microsoft does not apply it automatically, and a November claim must reach the company by Dec. 31. Admins can pull the incident IDs from the Service health page in the admin center.
Lamanna said a final review would add more detail later this week, Fox Business reported. House Homeland Security Committee chair Andrew Garbarino has asked Microsoft and CISA for more information, The Next Web reported. The final document will be read well beyond the admin center. Days earlier, for unrelated reasons, the FAA halted every US flight for four hours.


