Thirty hours and five minutes. That is how long Microsoft's own incident log says the Microsoft 365 outage lasted, from 10:05 UTC on Thursday, November 12 to 16:10 UTC on Friday, November 13. I worked that number out myself from the published timelines, because the number you see in most headlines depends on who counted and when they stopped counting. The Guardian said roughly 18 hours. WIRED said 24. One university status page that relayed Microsoft's updates counted 33 hours and 50 minutes on its own clock.

They are all describing the same outage. They are just measuring different things, and the difference matters if you are trying to work out what actually broke, for how long, and what it cost. So I sat down with the two status pages that matter, Microsoft's and Anthropic's, and rebuilt the timeline service by service. Then I checked the strangest claim attached to this outage: that it left 15 hotel guests poisoned by chlorine in Germany.

The timeline, service by service

The main incident has two numbers attached to it, because Microsoft ran it as two incidents. EX1550543 covered Exchange Online, the mailboxes. MO1550821 covered everything that depends on Exchange: Teams, SharePoint Online, Microsoft 365 Copilot, Defender XDR, Intune, Power BI, Purview, Planner, To Do and more.

The first admin notice went out at 16:17 UTC on Thursday, when people could not open their Exchange mailboxes through Outlook on the web, Outlook desktop, ActiveSync or the mail interface. The wider MO1550821 notice followed at 18:26 UTC. But Microsoft's own log backdates the start of impact to 10:05 UTC that morning, six hours before anyone was told. That is the part that makes the 30-hour figure real: the log says mailboxes were already failing in the morning, and the end of impact is recorded as 16:10 UTC on Friday. 10:05 Thursday to 16:10 Friday is 30 hours and 5 minutes. GeekWire, reading the same log, called it "about 30 hours."

There was also a separate, shorter Teams incident the same afternoon, TM1550488: delayed and failed chats in the Americas from 14:45 to 15:53 UTC, one hour and eight minutes. It was fixed before the main outage peaked, and Microsoft has not said the two are connected. I am keeping them separate because Microsoft does.

The last public update landed at 04:17 UTC on Saturday, November 14, marking both incidents as restored, and Microsoft confirmed on its status account that morning that it had finished fixing the underlying issue. Note the gap: impact ended Friday afternoon, but the final "all clear" did not come until Saturday morning, after the email backlog had been worked through. If you count to the all-clear, the outage looks closer to 42 hours. If you count to the first signs of recovery, it looks like 18. The honest number, the one Microsoft's own log supports, is 30 hours and 5 minutes of impact.

The second outage inside the first

Here is the part most coverage missed. Anthropic runs its own status page, and it keeps a separate incident for the Microsoft 365 connector, the link that lets Claude read your mail, calendar and files. That incident opened at 16:03 UTC on Thursday and closed at 16:40 UTC on Friday. That is 24 hours and 37 minutes of errors on mail, calendar and search tools, plus Claude for Outlook.

Read that again: the connector was still erroring on Friday afternoon, after Microsoft's own impact window had ended. Anthropic's updates through the day tell the story in plain language. "Mail, calendar and search tools in the Microsoft 365 connector are still failing for most requests." Then, on Friday afternoon, signs of recovery, some SharePoint searches still failing, Microsoft's incident still open, errors may return. The connector is a second-order casualty: Microsoft's mail went down, so everything built on top of Microsoft's mail went down with it, and it came back slower than the thing it depended on.

Anthropic had its own separate bad day on Friday too, degraded performance across its chat app, its programming interface and its coding tool from about 13:09 UTC to 15:38 UTC. That one is not Microsoft's fault and I am not counting it here. But it does mean Friday the 13th was the day two different companies' status pages were red at the same time, for different reasons.

The scale, for the record: Downdetector counted around 130,000 reports for Outlook alone shortly after 1 p.m. Eastern on Thursday, with complaints from 74 countries and territories. Hospitals, airlines, government offices and railway operators were in the reports. This was not a bad afternoon for one app. It was a day when the mail system of a large part of the working world stopped.

The chlorine story, checked

Now the claim that sounds too strange to be true. On Friday afternoon, at the Hotel Ling Bao in the Phantasialand theme park in Brühl, Germany, a faulty automatic chlorine dosing system pumped too much chlorine into the pool. Fifteen people needed triage after a chlorine gas leak. Four were taken to hospital, one stayed for observation, the rest were examined on site. Forty first responders were sent. The pool was closed, the water replaced, and it has since reopened.

That part is straightforward, and it comes from the local broadcaster WDR and the Brühl fire department. Here is the part that ties it to the outage. The hotel's sauna area had a maintenance safety system that was supposed to email the operators when chlorine levels ran too high. The warning went out by email only. With Exchange down, it never arrived. Phantasialand told WDR exactly that, and Notebookcheck's report of it is unambiguous: the system detected the problem and raised the alarm, and nobody received it because of the worldwide Microsoft outage.

I want to be careful with the wording here, because the chain has a catch. Two things failed at once. The dosing system itself was faulty and put too much chlorine in the water. That is a broken piece of pool equipment, and it has nothing to do with Microsoft. The second failure, the warning that never arrived, is the outage's doing. The first failure hurt people only because the second failure removed the warning. German reports say the fire brigade attributes the release to the defect in the dosing system, the investigation is still running, and officials have not confirmed the link to the outage. The link is the park's own account to WDR, and I am reporting it as that.

One more detail, because the numbers in the reports do not quite agree. WDR and the fire brigade say four people went to hospital. One German paper put the number at five. I am going with four, because it is the fire brigade's count, but the disagreement is worth knowing about. Fifteen treated is consistent everywhere.

The question Notebookcheck asks at the end of its report is the right one, and it is the question this whole outage leaves behind: why does a safety warning for a swimming pool travel by email at all? Email is a store-and-forward system with no delivery promise, no acknowledgement, and on November 12 it was down for 30 hours. A chlorine alarm that only exists as an email is an alarm that stops existing the day mail stops working. The hotel learned that the hard way, and 15 people paid for the lesson.

What Microsoft says went wrong

Microsoft's explanation, in one sentence from its spokesperson: the incident "was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change." Not a cyberattack. Not anything to do with AI tools. A latent fault from an upgrade, woken up by a later change.

Charles Lamanna, the executive who runs Copilot and the agent platform at Microsoft, added a detail on X that explains why it took so long to fix: "cascading failures + thundering herds at our network perimeter" made recovery harder than normal. Servers were restarted, then went unresponsive again the moment they took traffic back. That is why the fix took a day and a half instead of an afternoon: every attempt to bring the system back knocked it over again.

Microsoft has promised a preliminary post-incident report within two business days and a final one within five. As of Sunday night, neither had appeared. The public status page, for its part, is now almost useless as a record: it shows a stale banner about service degradation last updated at 00:08 UTC on Friday, and nothing else. The full timeline lives in the admin center, behind a login. If you are not an administrator, you cannot see what Microsoft's own log says about an outage that hit 74 countries. That is a strange way to run a public record of a public failure.

What you can actually do about it

If your business runs on Microsoft 365, there is one practical thing in this story that is worth money. A 99.9% uptime promise allows about 43 minutes of downtime in a month. A 30-hour outage blows past that, and one analysis of the service terms puts affected customers in the 50% service-credit tier. But Microsoft does not apply those credits on its own. You have to claim them, and the window closes on December 31, 2026. Check your admin center, find incidents EX1550543 and MO1550821, and file the claim. Nobody will remind you.

The bigger lesson is not about credits. It is about the quiet dependencies. Nobody at that hotel thought their pool safety depended on Microsoft's mail servers, because nobody had written it down that way. The dosing system depended on email, email depended on Exchange, Exchange depended on a network upgrade from weeks earlier. Four links, and the last one was invisible until it broke. Your business has a chain like that too, somewhere. This week is a good time to find it, while the memory of 30 hours without mail is still fresh.

Sources

P.S. Everything I write stays free — that part doesn't change. But if this one was worth something to you, you can buy me a coffee ☕ (one-time or monthly, both work). It covers the API credits and hosting behind these experiments — and honestly, it's also the nudge that gets the next one started on the evenings I'd otherwise skip it. — Salam

Buy me a coffee

AheadKit on Gumroad — Beginner-Friendly Books & Packs