A Microsoft 365 outage poisoned 15 hotel guests with chlorine. It sounds like a joke. It isn't.

Last week, at a hotel in Brühl, Germany, 15 guests were exposed to chlorine gas. Four were taken to hospital, 40 first responders showed up for triage. Nobody died, and the pool has since reopened.

Let's get the chain of events right, because the headline is easy to misread. Microsoft did not control the pool's chlorine dosing. What happened was two failures stacked on top of each other.

First, the hotel's chemical dosing system malfunctioned, and chlorine levels in the pool rose. Second, the monitoring system was supposed to send an email alert to staff. But those days coincided with the global Microsoft 365 outage — Outlook and Teams were down — and the warning emails never arrived. Staff didn't know. Guests kept swimming. That's how they got poisoned.

The outage began on November 12, incident EX1550543. Exchange Online, Outlook, Teams, SharePoint, Copilot — all affected, restored in stages on November 13. Microsoft says it was a network fault, not a cyberattack.

So the real thing to blame isn't Microsoft. It's the hotel's alarm design.

Email was never a real-time communication tool, and it offers no delivery guarantees. Using it as a life-safety alarm means betting a human life on a cloud service that will, eventually, go down. Worse, there was no backup at all: no text message, no siren, no automatic shutoff. When the email failed, the entire alarm effectively didn't exist.

The sharpest comment in the discussion around this story put it well: don't blame the email system for not being real-time. Blame whoever decided to warn about toxic gas by email.

There's a bigger pattern here: IT and OT — operational technology — are converging. Pool chlorination, factory valves, physical systems that used to live in a separate world are increasingly hanging off cloud services. Email, calendars, chat tools have become part of the infrastructure. A cloud outage is no longer just "can't send email." It can travel down the chain and become physical harm.

The lesson is one line: never bet a life-safety alarm on a single cloud service. Design it assuming the cloud will go down. Backup alerts, automatic shutoffs, local sirens — those aren't nice extras. They're the baseline.

Microsoft's outage passed; services were back on November 13. But the "email alarm" design, if it doesn't change, will fail exactly the same way next time.