November 18, 2026
Microsoft’s Outlook Outage Started With a Bug It Shipped in July
Exchange Online was down for 23 hours and 50 minutes across November 12 and 13, 2026. Microsoft’s preliminary review blames a months old…

By Shivashish Yadav
14 min read
Exchange Online was down for 23 hours and 50 minutes across November 12 and 13, 2026. Microsoft's preliminary review blames a months old defect on its gateway servers, a routine Teams failover and missing capacity. I walk through the timeline, why restarts failed, what happened to mail sent during the outage, the phishing wave that followed and what to check in your own setup.
On Thursday, November 12, 2026, Outlook stopped loading for a large part of the working world. I want to walk through what Microsoft has published about why, minute by minute where the record allows, why the usual fixes failed, what happened to the mail people sent into the void, how attackers used the outage within a day, and what an IT team can take from it. The FAA grounded every departing flight in the United States the next afternoon, and I cover that too, because a lot of people assumed the 2 events were linked.
Most of the technical detail here comes from Microsoft's preliminary post incident review for incident EX1550543, dated November 15 and published to administrators in the Microsoft 365 admin center. I have not read it directly. I rely on detailed summaries from IncidentHub, UC Today and SMTPedia, and on the account Charles Lamanna, Microsoft's Executive Vice President of Copilot, Agents and Platform, posted on X on Monday, November 16. He opened it with an apology, writing "We are deeply sorry for the impact to all our customers."
How big it was
The impact window in the review runs from 16:20 UTC on November 12 to 16:10 UTC on November 13, which is 23 hours and 50 minutes, per IncidentHub. That is about 11:20 a.m. Eastern on Thursday to 11:10 a.m. on Friday. Outlook, Outlook on the web, Exchange ActiveSync for phones and the messaging and REST APIs other apps use all failed. A second incident, MO1550821, covered Teams, Copilot, Defender and other services. IncidentHub lists the services the review names as affected, and the list runs from Exchange, Teams and SharePoint through Intune, Planner, Power BI and Universal Print.
Downdetector gives a sense of scale. Reports of Outlook problems climbed to about 130,000 shortly after 1 p.m. Eastern, Al Jazeera reported in a story credited to AP and Reuters. Latin Times counted over 38,000 further reports for Microsoft 365 as a whole and roughly 4,500 for Teams. Within about an hour of the start, the site had logged reports from 74 countries and territories, according to a follow up, and by about 07:00 GMT on Friday it still had at least 43,000 Outlook reports from 72 countries. Those are complaints submitted by users. The number of affected people is unknown and certainly higher.
The list of those hit is long. Al Jazeera reports disruption to government systems in Singapore and Japan, to communications at the Supreme Court of the Philippines, and to Australia's spy agency and federal police, along with integrations used by OpenAI and Anthropic. Its London correspondent described doctors' surgeries falling back on phones to reach patients. Latin Times reports that IndiGo and Akasa Air saw their email support disrupted while flights kept operating normally.
You will see 2 durations quoted for this outage, about 24 hours and about 36. Both have a basis. The review's impact window is 23 hours 50 minutes. The public status page kept the incident open until 05:45 UTC on November 14, per IncidentHub, while queued mail drained and monitoring continued. IncidentHub also flags a third figure. DMARCguard reported a 10:05 UTC start taken from Microsoft's final EX1550543 update, which the review does not mention.
3 harmless things at once
Microsoft's review names 3 factors. Each was harmless alone, and together they took down the gateway that sits in front of Exchange Online.
The first is a latent defect. UC Today reports that a fix for Microsoft 365's networking gateway servers began deploying in July and ran worldwide for more than 3 months without problems. Inside it was a lock contention bug, which IncidentHub and SMTPedia place in an internal Windows process. Under certain load, threads waited on each other until the server ran out of CPU and memory and stopped responding. Lamanna said neither the defect nor the fix applies to broader Windows deployments, so the bug does not affect ordinary Windows machines.
The second is extra load. At 01:07 UTC on November 12, about 15 hours before impact, an automated failover moved Teams traffic from an alternative gateway onto the primary one, which also serves Exchange Online. IncidentHub says load then rose gradually through the day, with isolated server failures that automated systems handled. The failover raised load and changed the traffic mix on the shared servers.
The third is reduced capacity. Nearly 14% of gateway server capacity was out of service because of an unrelated issue the previous day. Microsoft says there was enough headroom for that alone to cause no impact.
At US peak hours on November 12, those 3 lined up. In less than 2 hours, the defect spread across the fleet, and at the lowest point more than 80% of the gateway servers were unresponsive. A fourth problem made it worse. The bulkheads between the gateway's capacity units, which exist to stop a failure in one place from spreading, did not contain it because of a separate issue the review has not described yet.
Why testing missed it
The review gives a specific reason. Lamanna said the contention gets worse on newer machines with more CPU cores, and Microsoft's load testing focused on weaker hardware with fewer cores. IncidentHub adds that the traffic mix and server models used in the labs also differed from production. A bug that only shows up on the biggest servers under an unusual blend of Teams and Exchange traffic passed every test it was given.
I think this is the most useful line in the whole review. Load tests are usually run on whatever hardware is cheap to spare. The production fleet gets the newest machines. When those 2 differ, the test proves the code works on hardware you no longer run.
The automation that was supposed to help spread it
The first minutes of the incident show a pattern worth knowing. IncidentHub's timeline from the review has anomaly detection flagging connectivity problems at 16:03 UTC. At 16:17, automation began redirecting traffic to other healthy endpoints. At 16:20 the incident opened.
Redirecting traffic away from a sick server is the textbook response. Here it fed the cascade. Every healthy server that took on extra connections was a server running the same defective code, now under more load, and it hit the same bug. UC Today reports that traffic redirection made failure more likely on the remaining servers, and that Microsoft's automated recovery was never designed for an outage that broad. At 19:20, Microsoft attributed the elevated load to the number of servers that were unavailable. At 01:53 on November 13, engineers disabled some of the automated traffic redirection.
Customers learned about it late. IncidentHub recorded the first public status page update at 18:10 UTC, 1 hour and 50 minutes after the review's start time. The first support case had arrived at 16:33.
Why turning it off and on again failed
The recovery took most of the 24 hours, and the timeline shows why. Microsoft paused all deployments and configuration changes at 19:00 UTC. At 20:00 it deployed a change to bypass the components failing health checks, at 20:30 it started restarting servers, and at 21:00 an alternative health check returned some of them to service. By 23:55 those servers had become unresponsive again under traffic.
The reason is in the mechanism. A load balancer pulls failed servers out of rotation and sends their traffic to the rest. With more than 80% of the gateway down, every server that came back met a queue of mail clients from around the world trying to reconnect, and it hit the same defect within minutes. Statusstack's log of the public updates shows Microsoft saying at 01:06 UTC that a process responsible for establishing secure connections had become overwhelmed, so servers stopped processing requests. A partial recovery in some regions around 04:00 UTC did not hold, and at 05:00 machines stuck in bad states were keeping changes from syncing, per IncidentHub.
Rolling back did not help either, because the change was months old. UC Today reports that standard rollback, flighting and feature flags could not quickly revert it. Engineers built an update that returned the code path to its last known good state and applied it to large groups of servers at once, such as more than half of a region. IncidentHub dates the first deployment to part of the capacity at 07:20 UTC on November 13, about 15 hours after impact began. Stability on updated servers was confirmed at 08:00, and the rollout went first to regions entering their business day. Parts of Asia Pacific recovered by 11:00, Europe was back to healthy levels at 11:35, and the incident ended at 16:10.
Why Teams escaped and Outlook stayed down
Teams chats and meetings worked for most of the outage. IncidentHub reports that Teams, OneDrive, SharePoint and the Office apps could fail out of the affected gateway and partly recovered earlier. Exchange Online could not. Lamanna wrote that not all services, like Exchange Online, can fail out of this network gateway, and IncidentHub says the review gives scale as the reason.
There is an irony in the sequence. Teams failing over onto the shared gateway was one of the 3 triggers, and Teams's ability to fail over again is what let it escape.
Teams had its own, earlier problem that morning. Incident TM1550488 ran from 9:45 to 10:53 a.m. Eastern, which IncidentHub attributes to EpiFive, citing BleepingComputer. Microsoft has not said whether it was related, and the review does not mention it.
What happened to mail sent during the outage
Email is built to survive short outages. A sending server that gets a temporary failure keeps the message and tries again, and it gives up only after its own retry window runs out. The question for this outage was whether 24 hours would outlast those windows.
Microsoft acted on its side. At 09:10 UTC on November 13 the status page said the email timeout window was being extended to 48 hours, per IncidentHub, and the review lists that as a completed repair item, so delayed messages would not be returned as undeliverable. Senders outside Microsoft depended on their own settings. IncidentHub reports that SparkPost saw temporary 4xx replies from Microsoft's servers, and that DMARCguard quoted Mailgun saying some messages may have expired with status code 602 after its default 8 hour retry window.
Delayed queues began shrinking at 13:00 UTC. Webmail returned to expected levels at 14:17 and was stable by 14:44, and most access methods were back to normal traffic by 15:54. If your business sends invoices, alerts or receipts through a third party mail service, its bounce log for those 24 hours is worth reading this week. Some of those messages may never have arrived, and the recipients will not know to ask.
What the status page said, and what the review says
The public updates during the incident and the review do not tell the same story. At 19:57 UTC on November 12, Statusstack's log shows Microsoft saying no recent changes correlated with the start of impact. At 01:06 UTC on November 13, the status page said Microsoft had begun reverting a recent network configuration change that correlated with the start of impact. In its 17:57 UTC update on Friday, Microsoft apologized and said the incident came from an issue during an earlier network capacity upgrade that surfaced after a more recent network change, per IncidentHub and SecurityWeek. IncidentHub notes that the preliminary review mentions neither a network configuration change nor a capacity upgrade. It names the July fix, the Teams failover and the missing 14%.
I read these as a team updating its theory as it learned more, which is normal during an incident. The status page also named a different list of affected services from the review, adding the admin center, Microsoft Graph and Windows 365 Cloud PCs. The review is the document to rely on, and Microsoft says the final version, with the full list of repair items and the longer term changes, is due on November 20.
What Microsoft says it has fixed so far
The preliminary review lists 5 repair items as completed, per IncidentHub and SMTPedia. The connection handling code is back to its last known good state. The email delivery window was extended. Extra capacity went in to help recovery. Microsoft added monitoring for traffic redirection patterns and connection health, and it added guardrails on how traffic gets distributed so one failure cannot cascade across the fleet.
The last 2 items answer the automation problem directly. The bigger structural questions are still open. UC Today lists 3 that the final review should answer. Is the bulkhead issue fixed? How will load testing cover high core hardware from now on? Will Exchange Online keep depending on a single gateway it cannot fail out of?
The phishing wave arrived within a day
Attackers used the outage almost at once. Fortra reports the campaign began appearing in inboxes early on November 13, while services were still down. Microsoft Threat Intelligence published its warning on November 14 at 06:32 UTC, per Suped, describing messages that hit organizations across multiple US sectors.
The emails were built to pass a quick check. SecurityWeek reports they posed as Microsoft 365 service health notices and cited EX1550543, the genuine incident number. Subject lines began with "Outage -" followed by a 6 character code, and sender names combined the target organization's name with "Outage Notification" or "Service Health," so they looked like a message from internal IT. Some claimed held mail would be deleted within 24 hours unless the recipient clicked a link to restore it. SMTPedia says the clicks routed through personalized SendGrid tracking links, and that a later update added more destination domains.
Fortra saw a second lure in the same campaign, a "Support Desk" message warning of a security issue on the account. It also found messages going to corporate and consumer addresses at once, which suggests broadly sourced recipient lists. Microsoft describes the activity as consistent with credential phishing, says its security products are blocking it, and has not reported confirmed credential theft or a count of victims.
Field Effect saw the same theme in text messages and fake IT support calls, which skip email filtering entirely. It also notes that Microsoft was the most impersonated brand in Q3 2026, at 33% of brand phishing attempts. Its post recalls that after the CrowdStrike failure in 2024, attackers sent malware disguised as a fix. I expect the pattern after every large outage now. People who just lived through an actual outage will believe a follow up message about it.
The defenses the security firms recommend are plain. Check service status through a bookmarked Microsoft 365 service health page or a known IT contact, and treat any link inside an outage email as suspect. Search mail logs for the "Outage -" subject pattern. For anyone who clicked, reset the password, revoke active sessions and check the mailbox for new forwarding or inbox rules. Fortra adds phishing resistant MFA where possible. Suped warns against blocking SendGrid wholesale, since the tracking service itself is used by many legitimate senders.
The FAA ground stop was a separate failure
The next afternoon, November 13, the FAA stopped departures nationwide. The FAA's own Q&A says the problem began at about 3:30 p.m. Eastern and air traffic resumed at 8:15 p.m. It says operators canceled about 12,462 flights, including 9,082 by major airlines, though not all of those were caused by the ground stop, and that every aircraft landed safely. Implicator reports it was only the second nationwide ground stop of all departures since September 11, 2001. The first followed the NOTAM system outage on January 11, 2023.
The FAA's account of the cause centers on its telecommunications network, run by L3Harris. Legacy FAA equipment became unstable when a flood of connection attempts hit the network, reportedly about 1,000 every 4 to 5 seconds. The initial investigation points to a radar site in Amarillo, Texas, and a radio site in El Dorado, Arkansas, where L3Harris had been deploying a new router. Blocking both locations stabilized the network. The FAA calls these potential causes, says the investigation continues, and says its own upgrades did not cause the outage.
FedScoop reports that L3Harris owns the FAA's copper network and has been converting it to fiber, and that Transportation Secretary Sean Duffy said the router being replaced in that switch started flooding other routers. Duffy said officials were working out whether L3Harris made an installation error or installed a faulty router. FedScoop also reports the FAA spent about 45 minutes hunting for the cause before the stop, slowed because maintenance is logged by hand. Officials said there was no sign of a cyberattack, and no public statement links the FAA failure to Microsoft's.
The 2 do share a shape. In each, routine work on a network exposed a weakness in shared infrastructure that everything else depended on.
Microsoft has seen this pattern before
Several of Microsoft's largest recent cloud outages started with routine work. On January 25, 2023, the Azure WAN review says, a network engineer adding capacity to the global WAN in Madrid ran a command that behaved differently on one manufacturer's routers, and Practical365 puts that Microsoft 365 incident at 07:05 to 12:43 UTC. On October 29, 2025, valid customer configuration changes exposed a latent bug in Azure Front Door's data plane, per the final review, with impact from 15:41 UTC to 00:05 UTC the next day. That defect also passed validation because some features were never tested across versions.
The closest match came this January. On January 22, 2026, Microsoft 365 went down across North America for about 10 hours, from 19:37 UTC to about 05:33 UTC the next morning, The Register reported. Microsoft's root cause statement, as IsDown archived it, blamed elevated service load caused by reduced capacity during maintenance. Reduced capacity and extra load were 2 of the 3 ingredients again in November.
The November 2026 outage ran longer than the 2023 and 2025 incidents together. A latent bug in a gateway that one of the most used services cannot route around is the reason.
What Microsoft ruled out
Microsoft said at 04:11 UTC on November 13 that there was no indication of malicious activity, per IncidentHub. SecurityWeek reports a later Microsoft statement calling it a technical issue unrelated to cyber activity or to AI agents.
The outage also came 17 days after Microsoft opened a voluntary exit offer to about 60,000 employees in Canada, India, the United Kingdom and the United States, about 27% of its global workforce, NBC New York reported. Chief People Officer Amy Coleman announced it on October 26. That timing drew speculation online. The review describes a defect shipped in July and a testing gap on hardware, and nothing in it connects the outage to staffing.
2 things from this generalize
The first is that a shared front door is a single point of failure, however many servers sit behind it. Teams, SharePoint and OneDrive could route around the broken gateway. Exchange Online could not, and it was down for 24 hours. Any service that cannot fail out of a dependency inherits every bug in that dependency, and automation that shifts load between identical servers only spreads that bug faster.
The second is that a test only proves the system it ran on. Microsoft tested a fix on fewer cores than production used, and the bug waited 3 months for the right load. If your staging servers are older or smaller than production, your load tests cover a fleet you do not run.
This week, I would write down the backup channel my team uses when email and chat both fail, send everyone the address of the Microsoft 365 service health page so the next outage notice gets checked there first, and pull our mail provider's bounce log for November 12 and 13.
Sources
- The November 12, 2026 Microsoft 365 Outage, IncidentHub
- Microsoft traces Exchange Online outage to a July bug, UC Today
- Microsoft 365 outage root cause, SMTPedia
- Microsoft 365 and Exchange Online outage, Statusstack
- Microsoft outage hits tens of thousands of users, Al Jazeera
- Microsoft hit by global outages, Al Jazeera
- Microsoft 365 global outage, Latin Times
- Microsoft rules out cyberattack, warns of phishing, SecurityWeek
- Microsoft warns of phishing exploiting the outage, Suped
- Fake Microsoft 365 outage notices, SMTPedia
- Threat actors exploit Microsoft 365 outage, Fortra
- Microsoft 365 outage theme drives phishing, Field Effect
- Nationwide Ground Stop Questions and Answers, FAA
- FAA lifts ground stop, Implicator
- FAA ground stop coverage, FedScoop
- Azure WAN post incident review VSG1-B90, LeanIX status page
- Microsoft 365 outage January 2023, Practical365
- Azure Front Door post incident review, Rise Up status page
- Microsoft 365 outage drags on for nearly 10 hours, The Register
- Microsoft 365 incident of January 22, 2026, IsDown
- Microsoft offers buyouts to a quarter of its workforce, NBC New York