Microsoft's November 12 Outage Was a Bug That Sat Quietly Since July. Rollback Couldn't Reach It, the Walls Between Regions Didn't Hold, and Exchange Had Nowhere to Go
If your company runs on Microsoft 365, do three things this week. Write down how your people reach each other when Outlook and Teams are both down, because on November 12 they were down together. Ask your Microsoft admin to save the post-incident review from the Microsoft 365 admin center when it is published, because it is the only document that will say exactly what failed. And check whether your contract's service credits apply, since a claim has to be filed by you and nobody files it for you.
What Microsoft has said
The outage began on Thursday, November 12, 2026, during the U.S. business day. Microsoft's explanation so far comes from Charles Lamanna, the executive whose organization owns the service, in a post on X that we could not read directly. Everything below is his wording as quoted by ABC News and Fox Business, or their paraphrase where marked.
The bug was old. "A latent defect caused servers to become overwhelmed and unresponsive under certain load conditions." Per Fox Business, it came from a fix built for the Microsoft 365 networking gateway servers in July and ran worldwide for more than three months without a problem.
Two other things lined up. Load on the gateway surged, and capacity was already reduced. Wikipedia's summary of his post puts the reduction at 14 percent of the gateway fleet. Lamanna, per ABC: the reduction "had no impact on its own," but "taken together, the increased load and the reduced server capacity" triggered the July bug.
It got very bad. "At its lowest, more than 80% of servers were unresponsive" (via Fox Business).
Recovery fought itself. As servers came back, "they were immediately overwhelmed by waiting traffic and became unresponsive again" (via ABC News).
Not an attack. "This was not a cyberattack or an issue caused by AI tooling" (via Fox Business).
Two more points come only from press and Wikipedia paraphrase, not from Microsoft's quoted words. The safeguards meant to keep a failure inside one part of the network did not, so it spread across regions. And the normal rollback tools could not undo a change that had been live since July. Teams, OneDrive, SharePoint and the Office apps could be moved off the failing gateway; Exchange Online, per Fox Business, could not make the same switch.
Fox Business reports Microsoft promised a fuller review. When it appears in the admin center, it supersedes this post, and we will update it.
The arithmetic of 14 percent
Take a fleet of servers and switch 14 percent of them off. The same traffic now lands on 86 percent of the machines, so every remaining server carries about 1 divided by 0.86, or 1.16 times, its normal load. That is a 16 percent increase on every box before anything else happens.
Now add a traffic surge on top. A 10 percent surge on the reduced fleet puts each server at about 1.28 times normal load. A server that was comfortable at 60 percent busy is now at 77 percent, and a server that was at 75 percent is at 96 percent. Somewhere in that range a defect that only shows up "under certain load conditions" wakes up.
None of this is a criticism of running at less than full capacity during maintenance. Everyone does it. It is the reason capacity planning has to assume the bad day: maintenance, a surge and an unknown bug, all at once.
Why it was worse than a normal bad deploy
Rollback only reaches recent changes. A bad change from this morning gets rolled back in minutes. A defect that shipped in July and behaved for four months is part of the baseline. There is nothing recent to undo, so engineers have to find it under pressure.
The walls between regions didn't hold. Large services split their servers into separate units, so a failure in one cannot spread. Here, per the reporting, that separation did not contain it.
Recovery stampedes. When servers return to a queue of waiting traffic, the first ones back take the full load and fall over again. The fix is to admit traffic slowly, not all at once, and it has to be built before the outage.
Exchange had one road. Teams and the other apps could be moved off the failing system. Email could not. If a service has only one path, the other safeguards do not help it.
What you can do, even though you don't run Microsoft's network
Have a second way to talk. If your incident chat, paging and status page all ride on Microsoft 365, then the outage also took out the tools you would use to manage the outage. We wrote about this the day after: check where your alerts land.
Save the post-incident review. Microsoft publishes it in the Microsoft 365 admin center under Service health. It is the primary source, and it is what your auditor or insurer will ask for.
Claim your credits. Microsoft's online services agreement has a financially backed service level. The credit is not automatic; your admin files a claim within the window the agreement sets.
Expect phishing that looks like an outage notice. "Your mailbox was affected by the outage, sign in to restore it" is an easy lure this week. Microsoft does not ask you to sign in from an email to fix an outage.
Learn the general lesson for your own systems. Keep headroom for the day maintenance and a surge coincide, test rollback for changes older than a week, and build slow restart into recovery.
How this fits the week
This was the first of three. The next day a router swap at a remote FAA site stopped flights nationwide, and the FAA has now confirmed it switched off the broadcasting function on more than 60 new routers (our math on that network). Four days later a software problem at Chase reached other banks through the payment network they jointly own (the bank post). None of the three was an attack. All three were ordinary changes meeting a system with too little room and too few walls. We pulled the pattern together in Forgot about 2009: three outages in five days.
Our earlier post on November 13, Did Patch Tuesday break Microsoft 365?, said Microsoft's incident notes named no link to the monthly updates. The root cause above is consistent with that: the defect came from a July gateway fix, not from November's Patch Tuesday.
What is not known
Microsoft's own full review. Everything here is press quotation and paraphrase of an X post we could not open.
Why the separation between regions failed.
How many customers and users were affected. Wikipedia cites about 430 million Microsoft 365 users worldwide; that is the user base, not the number affected. A $7 billion productivity estimate from Anderson Economic Group is a third-party estimate, not a measurement.
Sources
ABC News, "Microsoft identifies issues leading to global Outlook outage," November 16, 2026. Fox Business, "Microsoft says hidden software bug caused Outlook outage last week," November 16, 2026. Wikipedia, "2026 Microsoft outage," read November 17, 2026, for the timeline and its summary of Lamanna's post. Charles Lamanna's post on X, cited by both outlets, could not be read directly (the page refused automated access).
How do AI models see YOUR brand?
AIPM has audited 250+ domains. 15 seconds. Free while still in beta.



Comments