NEWS
The Microsoft 365 Outage Followed a Familiar Slow Restore
A latent network upgrade fault knocked Exchange Online down, then Microsoft had to drip servers back because live mail kept taking recovered machines offline.
Microsoft spent Friday putting Exchange Online back in stages after a network capacity upgrade knocked Outlook, Teams, and Copilot offline around the world. The company said the Microsoft 365 outage was a technical fault, not a cyberattack and not an AI agent.
Servers that failed their own health checks dropped out of rotation on Thursday, November 12, and the machines left standing then drowned under mail load. Recovered nodes later went dark again once live traffic returned, which is why Redmond refused a fast cutover.
Servers Failed Their Own Health Checks
Microsoft’s own dashboard timed the start of incident EX1550543 at 16:17 UTC on Thursday, November 12, first as mailbox access failures in the Americas. Thirteen minutes after the first public note, Microsoft 365 Status said users in any region could be hit. The companion ticket was MO1550821.
The early story was load. The later story was missing machines. Status wrote that a slice of infrastructure was under elevated load across several regions, then walked that back: the extra strain was not a spike in customer traffic. A number of servers had become unexpectedly unavailable, and the rest were carrying the mail.
Health checks meant to protect the fleet pulled those boxes out of rotation. Each dropout left fewer nodes to handle Outlook on the web, MAPI, Exchange ActiveSync, and REST connections, so the survivors failed next. Copilot, Teams calendars, SharePoint, Intune, and Defender tooling that depend on Exchange Online went with them. Microsoft paused all deployments, shifted traffic, and started restarting the sick machines.
THE THURSDAY NIGHT CASCADE
- November 12, 2026, 16:17 UTC: Exchange Online mailbox access fails in the Americas, tracked as EX1550543.
- 17:54 UTC: Microsoft 365 Status posts that it is investigating those mailbox errors.
- 19:12 UTC: The company pauses all deployments, moves traffic off the damaged pool, and restarts affected servers.
- 19:31 UTC: Status says the elevated load is not extra customer traffic, but servers that became unexpectedly unavailable.
- November 13, 00:06 UTC: Servers already returned to service become unresponsive again once they receive traffic.
- 03:13 UTC: Status says the serviceability changes “are not taking as expected” and starts looking for other ways to add capacity.
By late Thursday, engineers were testing a configuration change aimed at the component that processes server health, and they were weighing a revert of recent network updates. An alternative health-check method got a subset of boxes back. Then those boxes fell over again.
Why the Restore Took Longer Than the Break
The overnight failure was not a missing reboot. It was recovered capacity dying under the same mail it was supposed to carry. Microsoft 365 Status put that in public language before the apology landed.
The servers we returned to service can become unresponsive again once they receive traffic; we’re analyzing diagnostic data to determine what’s triggering the behavior.
Microsoft 365 Status, on X, November 13, 2026
So the company throttled which traffic could reach the damaged pool, rebuilt some servers on a previous configuration, and only then let users back in waves. It also extended the email timeout window to 48 hours so delayed messages would not expire while queues were still stuck. That is a mail-preservation move, not a claim that nothing was lost from the user’s point of view. Delivery was late, and some people could not open a mailbox at all.
On Friday, Microsoft said it had fixed the underlying issue and was seeing broad recovery, while still restoring remaining customers and regions. The working cause, in the company’s words, was “an issue during an earlier network capacity upgrade that surfaced following a recent network change.” It ruled out cyber activity and AI agents. A full post-incident report is still pending.
We are deeply sorry for the disruption this incident has caused our customers and partners. We know the trust they place in our services and take that responsibility extremely seriously. We are seeing broad recovery and stability across our services.
This was a technical issue…
— Microsoft (@Microsoft) November 13, 2026
The slow walk-on was the point of the method. “Given the scale of demand, we restored services gradually, validating stability at each stage,” the company wrote. “While this took longer, it helped ensure a stable restoration.” Charles Lamanna, executive vice president for Copilot, Agents and Platform, said teams were reviewing how to stop a repeat and that he regretted the hit to people who depend on the suite every day.
Last October’s Front Door Failure Used the Same Playbook
This was a different product from Azure Front Door, and it was still the same recovery logic. After the October 29, 2025, Front Door crash, Microsoft’s own review said bringing edge sites back too fast would overload the ones that were already healthy. The published fix was gradually rebalancing Front Door traffic onto a small set of good sites, then opening the rest.
That incident ran from 15:41 UTC on October 29 to 00:05 UTC on October 30, a span of 8 hours 24 minutes. Incompatible configuration metadata had passed health gates because the crash showed up late, then poisoned the “last known good” snapshot. Microsoft 365 was among the downstream casualties, logged then as MO1181369. Abhishek Tiwari, vice president of engineering for Azure Networking, later set a target of about 10 minutes for automated rollback at every edge site and listed the Azure Front Door resiliency repairs that followed.
November 12 rhymed with that night. A network change exposed a latent capacity-upgrade fault. Automated health handling emptied the pool. Live traffic then punished the machines that came back. The safe answer, again, was to drip load in and watch each stage hold.
HOW THE LAST YEAR’S NETWORK FAILURES CAME BACK
| Date | What broke | How Microsoft brought it back |
|---|---|---|
| January 25, 2023 | A planned WAN router change made routers recalculate routes and interrupt forwarding | Most services recovered that day; some Outlook on the web and ActiveSync trouble ran into January 26 |
| October 9, 2025 | A manual cleanup let bad Front Door metadata past canary sites | Protection had worked until engineers bypassed it; impact was heaviest in Europe and Africa |
| October 29, 2025 | Incompatible Front Door metadata passed health checks, then crashed edge sites and DNS | Staff edited the last-known-good snapshot and reloaded sites in stages to avoid overload |
| November 12, 2026 | A latent fault from a network capacity upgrade, exposed by a later network change, took Exchange Online into a health-check cascade | Servers returned to service failed again under traffic, so restore ran in validated stages |
July 23, 2026, added a smaller rhyme: a West US network path failure that Microsoft said had downstream Microsoft 365 impact under MO1437424. September 30, 2026, brought another gateway control-plane stumble during a rolling update. The November event was larger because Exchange Online is the shared mail and calendar hub, not an edge fabric a portal can fail away from.
Hospitals, Courts, and a 48-Hour Mail Queue
Ookla’s Downdetector desk called this the largest outage of the past year on its tracker, and the longest. Reports stayed high for about 20 hours, roughly twice the plateau of any other large outage in that window. Other big incidents faded in a few hours once the fault was cleared.
THE DOWNTIME IN NUMBERS
- Report volume: People filed more than 7 million reports on Microsoft services by 16:00 UTC on November 13.
- Peak rate: Combined reports on five Microsoft services hit 19,650 a minute at 18:25 UTC on November 12, against fewer than 15 a minute on a normal Thursday.
- Spread: The outage reached 74 countries and territories in about an hour; more than 90 countries later logged at least 1,000 reports.
- Outlook: Outlook reports alone peaked at more than 2,000 times their normal level.
Microsoft's outage reports on November 12 reached 74 countries and territories in about an hour on Downdetector.
Microsoft says its Exchange Online incident began at 16:17 UTC. Reports to Downdetector about Outlook began climbing at about 17:25 UTC, 12:25 in New York. By 18:25,… pic.twitter.com/7FQ4dDsfbm
— Downdetector (@downdetector) November 12, 2026
Microsoft’s investor figures, as cited by Ookla, put paid commercial seats above 450 million. That is the installed base, not a count of people who lost a full day’s work. Patrick Anderson, president of Anderson Economic Group, treated a widely circulated 430 million “affected users” figure as mostly annoyance and estimated that perhaps 26 million people on Thursday and 51 million on Friday had operations disrupted enough to cause a real loss. His starting productivity bill was about $7 billion, with room to rise, plus a notional $800 million a day if customers pressed for subscription refunds.
The public-sector list is shorter and uglier than a seat count.
WHO LOST THE INBOX
- Japan: A Japan Digital Agency email outage left the agency unable to receive mail; deputy chief cabinet secretary Masanao Ozaki said government IT was disrupted.
- New Zealand: Immigration New Zealand paused visa processing while the suite was down.
- United Kingdom: NHS staff lost email and calendars, with some c-sections cancelled because booking sits on Microsoft; Rachel Harrison, GMB national secretary, said ambulance and NHS trusts took a serious hit while life-threatening callouts still ran. Police forces could not take online crime reports. Courts warned judgments might slip.
- United States: The House of Representatives told staff it was affected and that engineers were working with Microsoft.
People who still had to meet did what offices always do. Ookla recorded Zoom reports rising more than 50 times usual, Gmail more than 40 times, and GoDaddy, which resells Microsoft 365 mail to small firms, more than 60 times. Some of that is substitution. Some of it is other people’s mail bouncing and looking like a Gmail fault.
Banks Already Treat Microsoft as Critical Kit
Regulators had already written this dependency down. In November 2025, EU financial authorities named Microsoft a critical technology provider to banks and insurers. In July 2026 the UK Treasury issued critical technology provider safeguards that put the company in the same bucket for British finance.
Those labels did not shorten Thursday night. They exist because a mail and identity outage at this scale is no longer a desktop inconvenience. Ciaran Martin, former head of the UK National Cyber Security Centre, called the event one of the biggest outages of all time and said it might be worse than the recent Amazon and CrowdStrike shocks. James Ferryman, a computer science professor at the University of Reading, called it one of the longest and most disruptive Microsoft 365 failures the suite has had.
The comparison that matters inside Microsoft’s own paper trail is narrower. Front Door can fail a portal away from a sick edge. Exchange Online is the thing hospitals, courts, and legislatures use to send the next instruction. When its health checks empty the pool, there is no quiet failover that still delivers the inbox.
Still No Full Incident Report From Redmond
Microsoft has not closed the story with a root-cause paper. Friday’s account is the working version: a fault introduced during an earlier network capacity upgrade, exposed by a later network change, then a staged restore because demand was too large to slam traffic onto recovered boxes. Admins were told not to rebuild Outlook profiles, chase local DNS, or rewrite tenant routing while regions were still coming back.
WHAT WE KNOW
- Trigger: Microsoft says a latent network-capacity upgrade issue surfaced after a recent network change, and that cyber activity and AI agents were not involved.
- Failure mode: Servers failed automated health checks, dropped out, overloaded the rest, then failed again when live traffic returned.
- Mail handling: The timeout window was stretched to 48 hours to keep delayed messages from expiring during the backlog.
WHAT IS UNCONFIRMED
- Final engineering cause: The post-incident report has not been published, including which upgrade and which later change lined up.
- Lost mail: Microsoft has described delay and extended timeouts, not a confirmed purge of messages.
- Customer credits: Anderson’s $800 million-a-day refund sketch is a scenario, not a Microsoft offer.
Until that report is public, the October Front Door review is the closest signed manual for what operators just watched on Exchange Online. Microsoft already knew that recovered capacity can kill itself if you feed it the full queue at once. On November 12 it had to prove that again, in front of hospitals, courts, and a 48-hour mail timer.
-
NEWS3 months agoGeneration Lab’s Secret Youth Shot Has a Copycat Problem
-
AUTO3 months agoTesla Raises Dual Motor Prices as Texas Builds Cybercabs
-
LIFESTYLE3 months agoThe Nantucket Friendship Basket Boom Meets a Maker Shortage
-
BUSINESS3 months agoBristol Myers Quits Cellares as Autoimmune Doses Proceed
-
NEWS3 months agoOkta Stock Jumps 20% on McKinnon’s Identity Bet
-
BUSINESS3 months agoPoland Closes Visa-Free Work for Three Fast-Growing Nationalities
-
LIFESTYLE3 months agoLabor Day Mattress Sales Repeat a Familiar Holiday Discount
-
GAMING3 months agoGTA 6 Bets Its Last Look on Netflix and $100
