On Thursday 12 November 2026 at 16:17 UTC, users around the world started losing access to their Exchange Online mailboxes. The problem spread to parts of Teams, SharePoint, Copilot and several security and management services, and Microsoft's own end time for the incident is 16:10 UTC on Friday 13 November. That is about 24 hours of degraded email for a large part of the business world.

Microsoft says the outage was not an attack. Its preliminary explanation is a problem introduced during an earlier network capacity upgrade that surfaced after a recent network change. A full post-incident report is still pending. This article covers what is known, what your organisation can claim, and what to set up so the next outage costs less.

Timeline of the Microsoft 365 outage (UTC)

  • 12 Nov, 14:45 to 15:53: a separate Teams incident (TM1550488) delays chat messages for users in the Americas.
  • 12 Nov, 16:17: users cannot open mailboxes. Incident EX1550543 (Exchange Online) starts.
  • 12 Nov, 18:19: Microsoft reports elevated load across multiple regions affecting Outlook on the web, desktop Outlook, mobile mail and REST access. Outlook problem reports on Downdetector peak at around 130,000.
  • 12 Nov, 18:26: a second incident, MO1550821, is opened for dependent Microsoft 365 services.
  • 12 Nov, evening: traffic is rerouted, deployments and configuration changes are paused, servers failing health checks are removed and restarted. The workarounds bring limited relief.
  • 13 Nov, 00:00 to 02:14: Microsoft limits traffic in some locations and starts reverting a recent network configuration change.
  • 13 Nov, 09:10: the email timeout window is extended to 48 hours so delayed mail is not lost.
  • 13 Nov, 17:57: Microsoft apologises and states the preliminary root cause.
  • 14 Nov, 03:51: the incident is declared fixed, with a stated end of impact at 16:10 UTC on 13 November and the mail backlog processed.

What caused the Microsoft outage?

Microsoft's statements during the incident tell a consistent story. The extra load did not come from more traffic. A number of servers became unexpectedly unavailable, which pushed work onto the remaining ones. A process that establishes secure connections was overwhelmed, and affected servers stopped processing requests. The trigger, according to the preliminary statement, was an issue introduced during an earlier network capacity upgrade that surfaced after a recent network change.

Two points are worth noting. Microsoft repeatedly said it saw no sign of malicious activity and that the incident was unrelated to AI agents. And the Azure status page stayed clear: this was a Microsoft 365 incident, not an Azure platform failure, even though some outlets reported Azure complaints.

Why the fault spread so widely and why recovery took a full day are questions only the post-incident report can answer.

What stopped working, beyond email

The mailbox failure was the visible part. The dependent services are where many organisations were caught off guard:

  • Teams: calendar scheduling, meeting invites and edits, and live events could fail. Chat and running meetings mostly kept working.
  • Microsoft 365 Copilot: chat could fail to load or to reach mail, calendar and files.
  • Security and compliance: Defender for Office 365 and Purview reported delayed or failed enforcement of data loss prevention and governance policies.
  • Microsoft Graph: mail, calendar and contact requests timed out, which breaks every in-house application and integration built on them.
  • Others: SharePoint external sharing, Intune, Planner, To Do, some Windows 365 and Power Platform functions.

The lesson in that list: an email outage is also a scheduling outage, an automation outage and, for a while, a gap in security controls.

SLA service credits: what you can claim

Microsoft's Online Services SLA promises 99.9% monthly uptime for Exchange Online. The credit tiers for the monthly fee of the affected service are:

  • below 99.9% uptime: 25% credit
  • below 99% uptime: 50% credit
  • below 95% uptime: 100% credit

A 30-day month has 43,200 minutes, so 99.9% allows about 43 minutes of downtime. Twenty-four hours without service would be roughly 96.7% uptime for a fully affected user. The real figure for your tenant depends on how many users were affected and for how long, because the calculation uses user-minutes.

  • Credits are not automatic. You have to file a claim through the Microsoft 365 admin center and reference the incident IDs EX1550543 and MO1550821.
  • There is a deadline. According to current reporting, claims for services other than Azure must be filed by the end of the month after the billing month of the incident. Check your own agreement.
  • Bought through a reseller or CSP? Then the partner files the claim for you.
  • Collect evidence now: affected services, start and end times, number and location of affected users, screenshots and ticket numbers. Service health only keeps 30 days of history.

Seven lessons for IT and platform teams

  • 1. Keep one communication channel outside the suite. If email, calendar and chat all live with one provider, you need a way to reach staff when that provider is down: an SMS list, a status page on separate hosting, a second messaging tool.
  • 2. Know where to look without signing in. Bookmark status.cloud.microsoft and subscribe to service health alerts so the first signal does not come from users.
  • 3. Map what depends on Graph and Exchange. Ticket systems that read a mailbox, invoice flows, approval workflows, monitoring alerts sent by email. Each needs a defined behaviour when the API times out: queue and retry, not fail silently.
  • 4. Send critical alerts on more than one path. If your on-call alerts arrive only by email or Teams, an outage like this one hides every other incident.
  • 5. Treat degraded security controls as an incident of their own. When policy enforcement is delayed, note the window and review what passed through it afterwards.
  • 6. Freeze your own changes and stop local repairs. During a provider outage, reinstalling Office, deleting Outlook profiles or resetting passwords only adds work. Tell users early that the fault is upstream.
  • 7. Rehearse it. A one-hour tabletop exercise, "email is gone for a day", shows which processes have no fallback. That is cheaper than finding out on a Thursday afternoon.

The same thinking applies to AI assistants. An assistant that only works inside one productivity suite is unavailable together with that suite. When we plan AI integrations, provider outages are part of the design: a fallback model, a queue, or a privately run model for the workflows that cannot wait.

Frequently asked questions

How long was the Microsoft 365 outage in November 2026?
About 24 hours. Impact started at 16:17 UTC on 12 November 2026, and Microsoft's stated end of impact is 16:10 UTC on 13 November. Microsoft declared the incident fixed early on 14 November.

What caused the Microsoft 365 outage?
According to Microsoft's preliminary statement, an issue introduced during an earlier network capacity upgrade that surfaced after a recent network change. Servers became unavailable and the remaining ones were overloaded. The full post-incident report is pending.

Was the Microsoft outage a cyberattack?
Microsoft says no. It stated repeatedly that there was no indication of malicious activity and that the incident was not related to AI agents.

Was Azure affected?
The incident was scoped to Microsoft 365, with Exchange Online at the centre. The Azure status page stayed clear, although some outage trackers showed Azure complaints.

Can we get compensation for the Microsoft 365 outage?
Business customers can claim SLA service credits: 25%, 50% or 100% of the monthly fee for the affected service, depending on monthly uptime. Credits are not automatic, they must be claimed through the admin center or your reseller.

Bottom line

Large cloud platforms fail rarely, and when they do, they fail for everyone at once. You cannot prevent that. You can decide in advance how your organisation communicates, which integrations queue instead of break, and who files the SLA claim. Teams that had answers to those three questions lost a day of email. Teams that did not lost a day of work.

Want a second pair of eyes on your dependencies? In Cloud Management and DevOps & Site Reliability Engineering we map single points of failure, build alerting that survives a provider outage and write the runbooks for days like 12 November. Our incident response playbook is a good starting point. Talk to us.

Sources