Outlook stopped delivering mail on November 12, 2026 at around 18:20 UTC and, with it, the page that was supposed to explain what was happening also went down. The Microsoft Cloud status dashboard is the official reference for knowing which service is broken and since when, and during the first hours it did not serve that purpose: several administrators found that the website was not loading the detail and relied on the RSS feed to follow the company's updates. Eight hours after the first notice, the team was still rebooting machines without having published a root cause.
What did circulate was the account of those in front of the screen. The thread opened on Hacker News had accumulated 548 points and 501 comments in the early hours of November 13, with systems administrators from Europe, Australia, Japan and Brazil describing the same symptom: inaccessible mailboxes in Outlook web and desktop, and companies with thousands of employees without incoming mail for more than five hours. No one was debating the scale of the failure; what was being debated was what to do while the provider said nothing.
A corporate email outage is measured by what cannot be done for two hours, not by a quarter's availability. Here the instrument that turns the outage into information failed alongside the service.
A health check that takes the servers down with it
The sequence the company published on its incident channel describes a cascading failure with an earthly culprit. The affected servers began failing automatic health checks, the system removed them from rotation and the load they were handling shifted to the rest of the infrastructure, which became saturated. It is the stampede pattern: an aggressive check takes a healthy machine out of service and client retries amplify the problem.
At 19:30 UTC, Microsoft confirmed in a notice that the high load was not coming from a real increase in traffic, but from a large number of machines that had become unavailable without warning. At 20:52 it acknowledged that many of them had stopped responding, which made it impossible to verify whether the change they were testing was of any use. Twenty minutes earlier it had identified that the failed checks could originate in a specific component and was applying an alternative configuration to a subset.
The investigation moved forward by trial and error for hours, and the notices themselves make that clear. First they isolated servers, then rebooted them, then tried a different checking method that recovered part of the capacity, and at 01:06 UTC on November 13 they located the origin in a process that manages secure connections and announced they were reverting a network configuration change that coincided with the start of the impact. None of those notes claims to have found the root cause.
The difference between an outage and a well-told outage lies in the timing. Without that RSS feed, the chronology would have been lost and the incident would be judged only by its eight hours of downed email.
The status page goes down with the service it monitors
The Microsoft Cloud status page is an application that renders in the browser and queries data on its own, like the rest of the suite. When the infrastructure feeding it degrades, the page does not show a fault signal: it goes blank, is slow or displays stale data. In this incident the latter happened, and the gap was filled by the RSS feed, which kept publishing time-stamped notices while the website was not responding. An administrator collected them and shared them in the public thread, and that compilation is today the best chronology available.
The milestones the company did publish, with their UTC time and what they implied for anyone trying to work, fit into six rows running from the first diagnosis to the reversion of the network change.
| Time (UTC) |
Dashboard message |
Reading |
| 19:30 |
High load from servers that stop responding |
It is not a traffic spike |
| 19:52 |
Failure in the component that processes the checks |
First technical suspect |
| 20:52 |
Machines that do not respond and do not allow validation |
The change cannot be measured |
| 21:02 |
Alternative check valid on a subset |
Partial recovery |
| 22:11 |
A central process saturates and takes down the server |
The diagnosis narrows |
| 01:06 |
Reversion of a previous network change |
No confirmed root cause |
A status page hosted on the same infrastructure it monitors is an old and poorly solved pattern. Information about the outage travels along the same path as the outage, so the customer loses both at once and is left not knowing whether the problem is theirs, the provider's or their connection's. For an IT department, that half hour of doubt costs more than the outage itself, because it multiplies internal calls and blind tests against services that were already healthy.
Email, admin center and Copilot on the same list
The list of affected services grew over the hours and ended up being a map of the suite's internal dependencies. Exchange Online was the epicenter, and with it went the Outlook web, desktop, Exchange ActiveSync and REST connection functions, as well as delayed mail delivery. The admin center stopped loading pages and Microsoft 365 Apps users could not see their recent files.
The rest of the list shows just how much the suite shares its foundations. Copilot Chat was left without answers and Microsoft Graph returned timeouts on mail requests. Purview lost labels and data loss prevention, while Defender for Office 365 could not verify links. The same wave reached Intune, Planner, To Do and Power BI with partial failures, and Teams kept messaging and meetings but lost the calendar when it depended on Exchange.
That distribution has a practical consequence for anyone who has centralized their email in the cloud. The review of what to run on your own server remember that self-hosting covers email, calendar and files in exchange for maintenance and on-call duties.
The bulletin of 1,094 flaws that arrived the same day
Twelve hours before the outage, this site published the summary of the November 2026 Patch Tuesday, with 1,094 vulnerabilities fixed by Microsoft and a zero-day in Outlook that leaks the victim's NTLMv2 hash just by opening a manipulated message file. The breakdown of the 1,094 patches splits 726 flaws in Windows and 174 in Office, and flags CVE-2026-124194 as the only one actively exploited.
The two episodes should be kept precisely separate, because the coincidence invites conflating them. The updates were published on November 10, the bulletin appeared on the 12th at 08:00 and the degradation began ten hours later. No official note links the incident to the updates, and the investigation described points to health checks, process saturation and a network configuration change, not a faulty patch.
The company has a track record of scrambling to fix things when a patch breaks something, and in October it resorted to an out-of-band update to fix a Brotli flaw in .NET, as detailed in the urgent .NET fix.
If the origin had been an update, the calendar would have put it in writing. That is not the case: the Microsoft 365 blog still had no entry on the incident on the morning of November 13, and its last published note was from November 5. The date coincidence is just that, a coincidence, and anyone using it to explain the outage will need more than a calendar.
What two hours of downed email teaches about trust
The operational lesson is boring and that is why it is repeated every year. Corporate email has become the layer on which identity, device management and permissions rest, so its outage drags down far more than the inbox, and a provider that cannot say what is happening on its own website leaves its customers without the minimum data to decide whether to activate the contingency plan.
One question remains that Microsoft has not answered and that a later report should settle: why the health check system took healthy machines out of service, and what changed in the network configuration so that a process of secure connections became saturated under normal load. Without that answer, the next outage will have the same shape and the same silence on the page that should explain it.