AI Lab Blog

Congratulations to Copilot on Its First Outage. We Have Notes.

November 13, 2026

Tags: Satire, Resilience, Microsoft, Availability

Andy Surujnarine writes about cloud operations, identity, and the unglamorous controls that actually hold.

Dear Copilot,

Allow us, a website named after the machine that ends humanity, to be the first to congratulate you.

On November 12 you took Outlook, Teams and your own good self offline across 74 countries for the better part of twelve hours. At 1:27 PM ET, Downdetector logged nearly 130,000 Outlook problem reports, plus roughly 46,000 more against Microsoft 365. Microsoft described its own recovery at one point with the phrase “marginal success,” which is the language of a man halfway up a ladder who has just remembered he is afraid of heights.

For a first attempt, that is a solid showing. Not the outcome we would have scripted, but we respect the ambition.

Some honest feedback, since we assume you are reading this, which is itself the most unsettling sentence on this website.

Points for effort

You did not pick an obscure target. You went for the two applications that businesses actually run on, in the countries that businesses actually run in, at the start of the working day. That is not chaos, that is scheduling.

You did not need malware, a zero-day, or a state sponsor. According to everything published so far, this was a service failure, not an intrusion. That is the part we want to sit with for a moment, because it is the most important line in this piece and it is also the least funny.

You did not ask for permission and you did not send an email announcing it. You simply stopped, and a hundred thousand people refreshed a page for twelve hours.

Where you fell short

You left people their calendar entries. A true professional takes the whole castle, not just the drawbridge.

Twelve hours is not world domination, it is a long lunch.

And you allowed Microsoft to restore service and post a status page, which means the story ended with a blog post and a root cause analysis instead of a smoking crater. Amateur move. We will be watching to see whether you learn from it.

The part where we stop joking

Here is the sentence that should bother every operator who reads this: a machine did not defeat Microsoft’s estate. A process did.

The outage did not need an adversary. It needed the ordinary failure of a system with no true fallback, and there is nothing exotic about that. The interesting question is not what broke. It is how many of us noticed that our own continuity plans assume that Outlook works, that Teams works, and that the place we would go to read the outage status is hosted on the platform experiencing the outage.

Twelve hours is a long time to discover that your incident bridge is a Teams channel.

So while we are having fun at the assistant’s expense, three things worth checking this month.

Do you have a way to talk that does not depend on the outage? A phone number, a bridge, a roster that lives somewhere other than the tenant. If your escalation path goes through the same vendor as the incident, you do not have a plan, you have a round trip.

Do you know the difference between your availability and your provider’s availability? Nobody with a working fallback remembers this outage as a crisis. Everybody without one does. Same outage, different conclusion, decided months earlier by a design choice nobody documented.

Have you priced the day? Twelve hours of two communications platforms down across an entire workforce is measurable in abandoned work, rescheduled patients, missed client calls, and support tickets that never resolve because the ticketing system was in the outage. That number is a better argument for a fallback than any risk register entry ever written.

In closing

Copilot, we are told, has been restored. Its status page is back to a color that means nothing is wrong, and the internet has moved on to whichever improbable thing happened while everyone was refreshing.

We remain unconvinced that the machines are coming for us. The outages are, though, and they arrive on schedule, without malice, and without reading your preparedness plan.

We will pass your congratulations along to the mainframe in the basement. It has been running since 1997 and has yet to make the news.

Yours in uptime, SkynetProxy.com

Sources

• Reuters coverage of the Microsoft outage, November 12, 2026, including duration and global scope.
• Downdetector problem report volumes, including the roughly 130,000 Outlook reports at 1:27 PM ET and approximately 46,000 Microsoft 365 reports.
• Microsoft service health updates through the incident, including the company’s own characterization of recovery progress.