Microsoft 365 outage: Outlook and Teams error screen during the November 2026 global disruption

Microsoft 365 outage: what happened and how to avoid it

•9 min read

The Microsoft 365 outage of November 12–13, 2026 cut off Outlook, Teams and Copilot for users in 74 countries for more than a day, and Microsoft says it was not a cyberattack: a defect left behind by an earlier network capacity upgrade stayed hidden until a later network change exposed it. What turned that defect into a global blackout was a chain reaction, in which servers dropped out of service and piled their load onto the survivors until those failed too.

What happened

On the afternoon of Thursday, November 12 (UTC), people around the world suddenly could not open their Outlook mailboxes. According to Al Jazeera, the problems began at about 16:00 GMT, and within roughly an hour Downdetector had logged reports from 74 countries and territories. TechRadar reported a peak of around 120,000 concurrent Downdetector reports in the US alone.

The email problem spread. TechCrunch reported that the outage also hit Copilot, Defender security screening, and the Intune dashboards administrators use to manage devices. Mexican outlet N+ also listed SharePoint Online, Planner and Power BI among the services that might be affected.

Timeline at a glance

WhenWhat happenedSource
Nov 12, ~16:00–16:17 UTCExchange Online and Outlook access problems beginAl Jazeera, Postbox Services
Nov 12, 18:26 UTCMicrosoft opens the main incident, MO1550821Control Plane
Nov 12, after 5 p.m. ETMicrosoft says initial fixes are not delivering the expected relief and that a core process on affected servers is being overwhelmedBetaNews
Nov 13, 04:36 UTCMicrosoft reports "partial recovery" in some regionsWikipedia (sourced summary)
Nov 13Microsoft apologizes, blames an earlier network capacity upgrade and rules out cyberattacks and AICNBC, Inc.
Nov 13–14Microsoft says service is restored to the remaining customers and regionsBrightDefense, The Times of Israel

The numbers don't line up perfectly, and that matters. Al Jazeera describes a 24-hour outage; Trending Topics puts it at about 30 hours. On-Site Technology cites Microsoft's final service health record as showing impact from 5:05 a.m. ET on November 12 to 11:10 a.m. ET on November 13, which starts earlier than the first user reports. Each source is most likely measuring from a different starting point. The official reference is Microsoft's record in the admin center; its post-incident report, which On-Site Technology says is still expected, should settle the final root cause.

The impact reached well beyond office inboxes. Per the Wikipedia summary, US House of Representatives staff told Politico the outage ended a day of legislative work, and UK minister Luke Pollard called for "greater resilience" against technology failures. Al Jazeera notes this was Microsoft's second service disruption this year and one of the longest in its history.

How it happened

It helps to separate what Microsoft has confirmed from what is analysis.

What Microsoft has confirmed

  • The origin: in a Friday statement reported by CNBC and Inc., Microsoft said the incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change, and that it was unrelated to cyberattacks or to AI and agents.
  • The failure pattern: Microsoft's early updates, per CNBC, described a number of servers becoming unexpectedly unavailable and straining the remaining infrastructure, with some servers failing automated health checks. Microsoft later clarified that the load was not driven by more traffic but by the servers that had dropped out, according to BetaNews.
  • The hard recovery: Microsoft rolled back a configuration change and patched a component responsible for authentication and routing, per Trending Topics. Restored servers kept failing again as soon as traffic hit them, which Macworld also reported.

How the chain fits together (our analysis)

What follows is our reading of the confirmed facts, not an official explanation. The pattern is a classic cascading failure:

  1. A latent defect. The earlier network upgrade left something wrong that caused no visible symptoms. Microsoft has not said what it was.
  2. A trigger. A second, apparently routine change activated that defect.
  3. Safety mechanisms amplify the damage. Health checks exist to pull a sick server out of rotation. But when many servers fail at once, the system removes a big slice of its own capacity, concentrating traffic on the remaining machines, which then overload and get pulled as well.
  4. Recovery becomes the problem. A server coming back online immediately absorbs the pent-up demand of millions of Outlook and Teams clients retrying at once. Postbox Services says Microsoft's updates pointed to a process that sets up secure connections as the one being overwhelmed. Encrypted connections are expensive to establish, so a reconnection wave can knock a freshly restored server straight back down. That is consistent with Microsoft restarting infrastructure and managing traffic, per Macworld, and adding capacity with virtual machines, per On-Site Technology.

Still unknown: what exactly the network upgrade changed, why testing didn't catch it, how many organizations were affected, and whether any email was lost. Reporting so far describes delayed mail, not lost mail. Microsoft has not published its final report.

How it could have been prevented

Nobody outside Microsoft can say for certain which control would have stopped this specific incident. But the practices that contain this kind of failure are well understood, and they apply to a mid-sized company's software as much as to a global cloud.

1. Staged rollouts with automatic rollback

What it is: apply every change to a small slice of infrastructure first (a "canary"), watch the metrics, then expand. If metrics degrade, the change reverts itself. Why it works: it caps the blast radius, so a defect hits one region or a small percentage of users instead of 74 countries. Effort: moderate; it needs deployment automation and trustworthy metrics.

2. Treat changes as a system, not one at a time

What it is: keep an infrastructure change history that lets engineers ask, "which earlier changes interact with this one?" Why it works: this failure came from two changes separated in time. Testing a new change against the real current state, not the assumed one, is how latent defects get caught. Effort: low to moderate; mostly discipline.

3. Health checks with an ejection limit

What it is: configure load balancers so they never remove more than a set share of servers for failing health checks. If "everything looks sick," the likely culprit is the check or a shared dependency, not every server. Why it works: it breaks the eject-overload-eject loop. Effort: low; many modern load balancers support it out of the box.

4. Load shedding and backoff on retries

What it is: an overloaded server quickly rejects part of its requests instead of trying to serve them all, while clients retry with growing, randomized delays. Why it works: it keeps reconnection waves from flattening servers that have just recovered, which is exactly what slowed Microsoft's recovery. Effort: moderate; it has to be designed into both servers and client apps.

5. Communication channels outside the failing system

What it is: status pages and alerts that don't run on the affected infrastructure. Microsoft partly did this, pointing administrators to an alternate page when the admin center was unavailable, per BetaNews. Effort: low, and just as relevant for a company talking to its own customers.

What to do today

No Microsoft customer could have prevented this outage. What you can control is how much the next one hurts.

For your company

  1. Map your dependencies. List which processes stop if Microsoft 365 (or Google Workspace) goes down: sales email, approvals, sign-in to other systems, invoices sent by email.
  2. Agree on a backup communication channel from a different provider, with key contacts stored outside Outlook, and make sure everyone knows when to switch.
  3. Chase delayed email. Ask your team to look for non-delivery notices from November 12–14 and resend anything important, especially quotes, invoices and orders.
  4. Write a one-page runbook for vendor outages: who decides, how customers are told, and what gets done by phone or on paper in the meantime.
  5. If you run your own software, ask your team or vendor whether deployments are staged, whether a change can be rolled back in minutes, and what happens when an external service stops responding. If the answers are vague, it's worth reviewing with a custom software development team.
  6. Read Microsoft's post-incident report when it lands (incidents EX1550543 and MO1550821 in the admin center, per BetaNews).

For everyday users

  • If an important email didn't arrive during those days, check with the sender or resend it.
  • Be wary of "verify your account" or "restore your access" emails after the outage: Microsoft doesn't need your password to fix its servers.

Frequently asked questions

Was the Microsoft outage a cyberattack?

No, according to Microsoft. The company said it was a technical issue unrelated to cyberattacks or to AI and agents, as reported by CNBC and SecurityWeek.

What caused the Outlook and Teams outage?

Microsoft's preliminary root cause is an issue introduced during an earlier network capacity upgrade that surfaced after a recent network change. It led many Exchange Online servers to drop out of service and overload the rest (Control Plane).

How long did the Microsoft 365 outage last?

More than a day. Estimates range from 24 hours (Al Jazeera) to about 30 hours (Trending Topics), with full restoration announced on November 13 or 14 depending on time zone.

Was the US flight ground stop related to Microsoft?

No. US officials said the November 13 ground stop was a telecommunications outage blamed on old hardware, not a cyberattack or the Microsoft outage, according to The Times of Israel.

Comments

No comments yet. Be the first to share what you think.

Leave a comment

It is never published.