The Exchange Online outage of 12 and 13 November: timeline, Microsoft's stated cause and what reached us
On 12 November 2026 at 16:20 UTC Microsoft reported that users could not reach their mailboxes in Exchange Online. By Microsoft's account incident EX1550543 ran until 13 November 16:10 UTC and took Teams, Copilot, Defender, Purview and Intune down with it. We looked at what reached two of our own systems: the Zammad ticket inbox, which reads over Microsoft Graph, and the mail gateway that delivers to our tenant.
Exchange Online is Microsoft's hosted mailbox layer. Outlook, mobile devices, Graph calls and mail to and from on-premises servers all go through the same servers that accept connections. When those fail, every access path fails at once, even if the mailboxes are untouched. That is how Microsoft describes this incident.
Timeline from Microsoft's posts
The times come from the update history of EX1550543 and MO1550821 in the Microsoft 365 admin center. We read them on a University of Pennsylvania mirror page (reported, state of 15 Nov, 10:55 UTC); our own tenant was not a source. All statements are vendor-claimed.
| Time (UTC) | Microsoft's post, paraphrased |
|---|---|
| Thu 12 Nov, 16:20 | Investigation of a possible problem starts; start time per the final text |
| Thu 12 Nov, 17:45 | Connectivity errors on part of the infrastructure that accepts access requests |
| Thu 12 Nov, 19:57 | No recent change found that matches the start; health-check component is bypassed |
| Thu 12 Nov, 22:00 | A core process on the servers is overwhelmed; part of the infrastructure is isolated |
| Thu 12 Nov, 23:59 | Servers returned to service fail again once traffic arrives |
| Fri 13 Nov, 01:06 | The process that sets up secure connections is overwhelmed; a network change that correlates with the start is rolled back |
| Fri 13 Nov, 07:25 | Stability improves after a patch to an upstream authentication and routing component |
| Fri 13 Nov, 09:03 | Microsoft extends the mail timeout window to 48 hours |
| Fri 13 Nov, 14:01 | Mail queues much shorter |
| Fri 13 Nov, 16:10 | End time of the incident |
| Sat 14 Nov, 04:15 | Final status with preliminary root cause |
BleepingComputer, The Register and SecurityWeek report the same sequence, some with US Eastern times. The Register adds that Intune and the Microsoft 365 Apps Admin Center went down too, which left admins short of diagnostic tools. MO1550821 also lists Graph calls on Exchange data that timed out, and Intune enrollment, check-in and compliance policies that were intermittently affected.
Microsoft's stated cause
The final status (vendor-claimed) gives as preliminary cause a failure in the systems that establish connections to Exchange Online. Traffic shifts during routine network maintenance exposed a latent bug, requests piled up, the servers used up their capacity and stopped accepting new connections. The incident was introduced during an earlier network capacity upgrade and surfaced after a recent network change. Microsoft rules out a cyberattack and AI agents in so many words.
Its statement about a change moved during the incident: at 19:57 UTC Microsoft said it had found no matching change, at 01:06 UTC it rolled back a network change that correlates with the start. That proves nothing either way. It shows how incomplete such posts are early on. A preliminary post-incident report is promised within two business days and a final one within five; neither was out when we read. Entra is not in the service list of MO1550821, and the Azure status history had no entry for these days on 15 November (verified).
What reached us
Both measurements were read-only and ran on 15 November 2026 around 16:50 CET. The hosts log in CET, and the table is converted to UTC.
| Measurement | Carrier and method | Result |
|---|---|---|
| Graph polling of the ticket mailbox | Zammad host, docker logs of the scheduler container, line "Unable to list emails from Microsoft Graph" | 1,636 errors from 12 to 14 Nov, 1,634 of them between 12 Nov 17:42 and 13 Nov 13:39 UTC; 1,566 "UnknownError", 60 HTTP 500, 6 HTML error pages, 3 gateway errors, 1 timeout |
| Delivery to the tenant | Mail gateway, mail.log, lines with status=deferred to the tenant host under protection.outlook.com | 3,198 deferrals for 87 queue IDs; all 87 later sent, no bounce caused by the outage |
The Zammad channel polls about 116 times an hour. On 13 November between 04:00 and 12:00 UTC, 112 to 119 polls an hour failed, which is practically every one. The channel shows "ok" today, and the mail stays in the mailbox because keep_on_server is set.
On the gateway the timing is the surprise. Thursday evening from 17:00 UTC delivery still worked, and the first seven deferrals came between 20:00 and 22:00 UTC. The bulk followed on Friday between 06:00 and 17:00 UTC, and in every hour from 08:00 to 12:59 UTC deferred attempts outnumbered delivered ones. The peak was 11:00 UTC with 600 against 182. The replies were 451 4.7.0 and 451 4.3.2 "Temporary server error", and 421 4.3.2 "maximum number of concurrent server connections has exceeded". Why the evening ran smoothly is unknown to us. Thursday evening also had little mail volume (35, 18 and 12 deliveries an hour), so nothing follows from it.
Admin checklist
- Set up a status source outside your tenant. The admin center partly failed to load. Microsoft points to the status page and the X account @MSFT365status, and the service health view is documented on Learn.
- Poll service health over Graph:
GET /admin/serviceAnnouncement/issueswith the permissionServiceHealth.Read.All(Learn). We did not test whether that call answered during the outage. - Do not force anything at the gateway. The 421 reply reports exhausted connections on the other side, and a permanent flush (
postqueue -f) adds load. We checkpostconf maximal_queue_lifetimeinstead: five days here, the default per the Postfix documentation. A 24-hour outage stays well inside that. - Review integrations that use Graph: does the client retry by itself, and does it alert after n failures in a row? Zammad resumed polling on its own after the outage.
- Test alert paths that themselves run through Exchange Online or Teams. They go quiet in exactly this case.
- Hold the post-incident report against the timeline and compare its cause with the statements of 12 November.
If you want these dependencies inventoried for your own operation, see our IT services. Another date in the tenant is the EWS shutdown in Exchange Online, and more operations posts are in the servers and network hub.
Open points
- We did not read Microsoft's admin center directly; the timeline comes from the mirror page. Its header gives the start as 12 Nov 10:05 UTC, the final text says 16:20 UTC. We used the text.
- Post-incident report: not published, so the cause stays preliminary.
- Zammad: whether every ticket arrived is unchecked. Two errors on the morning of 12 Nov predate the incident and are unassigned.
- Gateway: only our delivery to the tenant is measured, not the opposite direction.
- Entra sign-ins and Teams usage are not measured for the period.
- Reach, Downdetector figures and consequences at third parties are unchecked. We did not read BBC and Guardian; the fetch was not possible.
Read next: Stalwart 1.1.0 and the missing role rights · the Railway breach and which secrets to rotate · scaling at Slack and GitHub · the EWS shutdown in Exchange Online and Apple Mail · Exchange Server end of support in October 2026 · Zammad as a helpdesk
Further Reading
- Mirror of Microsoft admin center posts EX1550543 and MO1550821 (reported, read 15 Nov 2026)
- BleepingComputer, Massive Microsoft 365 outage (reported, bot wall 403, read through a rendered fetch)
- The Register, Microsoft struggling to fix massive cloud outage (reported)
- SecurityWeek, Microsoft 365 outage (reported, bot wall 403, read through a rendered fetch)
- Microsoft Learn, Service health (verified)
- Microsoft Learn, serviceAnnouncement issues (verified)
- Azure status history (verified)
- Postfix, postconf(5) (verified)
How long was Exchange Online disrupted on 12 and 13 November 2026?+
According to Microsoft's final status for EX1550543, from 12 November 16:20 UTC to 13 November 16:10 UTC, about 24 hours; the header of the mirror page we read gives the start as 10:05 UTC, which would make it about 30 hours. Processing of the queued mail continued afterwards, and the final status was posted on 14 November at 04:15 UTC. All times are vendor-stated.
What was the cause?+
Microsoft's preliminary root cause is a latent bug in the systems that accept connections to Exchange Online. Traffic shifts during network maintenance triggered it, requests queued up and the servers stopped accepting new connections. That is a vendor claim, and the post-incident report is still outstanding.
Was Entra ID affected?+
Entra does not appear in Microsoft's text for MO1550821, and the Azure status history has no entry for those days. We did not measure Entra sign-ins in this window ourselves, so the answer rests on Microsoft's text.
senn-tech