Last week Microsoft 365 was down for about a day. Most of the news was about email, because that's what everyone noticed first. We wrote up the timeline on Friday, while it was still going on. You can read that here: Microsoft 365 outage, 12 November: what happened.
This post is about something that got less attention. The outage also affected some of Microsoft's security tools. Defender, Purview and Intune were all on Microsoft's list of affected services. So for about a day people couldn't use email, they found other ways to send files, and the tools that normally check that kind of sharing weren't fully working either.
There is no sign that anyone was attacked. But it does leave some checking to do, and most companies haven't done it yet. Below is a short recap of the outage, and then the list of things we would check.
What happened
- It started on Thursday 12 November at 16:17 UTC. People couldn't open their mailboxes in Outlook. It was first reported in North and South America and then spread to every region.
- About two hours later Microsoft widened the incident to include Teams meeting scheduling, Copilot, parts of SharePoint, Planner, Power Platform, Defender, Purview, Intune and some admin pages. Azure was not affected.
- The night was difficult. Microsoft restarted servers, and they stopped responding again as soon as users reconnected. More than 12 hours in, Microsoft said it had only had "marginal success".
- Things improved on Friday morning after Microsoft applied a patch. Asia recovered first, then Europe and North America.
- Microsoft's own record says the impact lasted roughly 24 hours. On Saturday 14 November it confirmed that the last affected customers and regions were back.
Microsoft has not said how many people were affected. There were reports from 74 countries, and Downdetector had about 130,000 reports about Outlook at the peak. In the UK, parts of the NHS went back to paper and cancelled some operations. Staff at the US House of Representatives lost a day of work. Richemont delayed its quarterly results. One consultancy, Anderson Economic Group, estimated the cost at around 7 billion dollars in lost productivity. That is an estimate, not a measured figure.
This was not the first big outage this year. In January, Microsoft 365 was down across North America for almost ten hours.
What caused it
On Friday evening Microsoft apologised and gave a cause. It said the problem came from "an issue during an earlier network capacity upgrade", which only showed up after a later change to its systems.
From Microsoft's status updates, this is roughly what happened:
- Some of the servers that run Exchange Online failed their automatic health checks and were taken out of service.
- The servers that were left had to handle everyone. Microsoft said clearly that there was no increase in traffic. There were simply fewer servers.
- Those servers couldn't keep up with setting up secure connections and stopped responding.
- Each time servers were restarted, every Outlook client in the world tried to reconnect at once, and the servers went down again. This only stopped once Microsoft added capacity, limited traffic and applied a patch.
Microsoft has promised a full report. If you are an admin, download it when it appears in Service health, because that history is only kept for 30 days. The incident numbers are EX1550543 and MO1550821.
Was it a cyberattack?
Microsoft says no. Its statement was that this was "a technical issue not related to cyber or AI / agents". No group has claimed it, and nothing has been published that points to an attack. Independent analysts quoted in the press said they found Microsoft's explanation believable.
Still, a lot of theories went around while there was no official cause. These are the ones we heard most:
| Theory | Why people believed it | What we know now |
|---|---|---|
| It was a cyberattack | There was no explanation for most of the first day, and a US congressman asked publicly whether something "nefarious" was going on | Microsoft says no. An attack that floods a service shows up as extra traffic, and there was none. |
| AI or AI agents overloaded the service | Copilot stopped working too | Copilot stopped working because it reads email and calendars, and those were down. |
| It was caused by Patch Tuesday | Microsoft released nearly 1,100 security fixes two days earlier | Microsoft has not linked the two. The cause it gave was a network upgrade. |
| It was a bad update, like CrowdStrike in 2024 | An Australian minister said Microsoft told him an update had gone wrong | This is the closest to the official explanation. The difference is that the change was on Microsoft's own network, not on customers' computers. |
| AWS and Google were down as well | Reports for both went up on Downdetector at the same time | Amazon said its services were working normally. People often report the wrong service when something stops working. |
One fair question is still open. TechCrunch asked Microsoft how it can be sure there was no malicious activity, and did not get an answer before publishing. We expect the full report to cover that.
Which security tools were affected
According to Microsoft's advisory and the admin blogs that followed it, these were affected at some point during the outage:
- Safe Links. Links in emails that Safe Links had rewritten would not open. This was fixed as the service recovered.
- Quarantine. Releasing a message from quarantine could fail. Also fixed during recovery.
- Data loss prevention (DLP) and sensitivity labels. Microsoft warned that policies relying on Purview DLP and information protection "may experience delays or failures in policy evaluation and governance action enforcement". In plain terms, a rule that normally blocks or flags the sharing of a sensitive file may have run late or not at all.
- Purview auto-labelling, communication compliance and data classification.
- The Defender and Purview portals might not load, so security teams had less visibility than usual.
- Intune. Some admins couldn't open their dashboards.
One detail shows the kind of choices people faced. Some labelled items in Microsoft Fabric could not be exported during the outage. The workaround Microsoft suggested was to "remove the label from the item, if your data protection policies allow it". That is a reasonable suggestion when people need to get work done. It also means some files may have had their label removed last week, and somebody needs to check that the label went back on.
To be fair to Microsoft, none of this is surprising. These tools are built on top of email, so when email is down they are affected too. Microsoft said so openly while it was happening.
What people did while email was down
Email was down, but SharePoint and Teams chat mostly kept working. So people did the obvious thing and used what still worked:
- They sent files as sharing links instead of attachments. When someone is in a hurry they often choose "anyone with the link", because it always works for the person receiving it.
- They sent work documents from private email accounts, or through WhatsApp and similar apps.
- They downloaded files to their own computers so they could keep working.
- Some used personal laptops and phones.
None of this is bad behaviour. People were trying to do their jobs. But it means company files may now be in places where your normal rules don't apply, and some of the links made last week will stay open until someone removes them.
What to check now
Look at the period from Thursday 12 November, 16:17 UTC, until your users were working normally again. It helps to compare it with a normal day before the outage.
| What to check | Why | What to look for |
|---|---|---|
| Sharing links created during the outage | People used links instead of email attachments | "Anyone with the link" links, links sent to outside companies you don't normally work with, and links to files that contain sensitive data |
| Sensitivity labels that were removed or lowered | Removing a label was a suggested workaround | Any label change during the outage, and whether the label has been put back |
| Unusually large downloads | People copied files so they could work offline | Users who downloaded far more than they normally do. Ask them why. |
| Email forwarding rules | Sending work mail to a private address was a quick fix | Forwarding rules created last week that are still active |
| New apps given access to your tenant | People connected other tools in a hurry | Apps that were given access to email or files last week, and who approved them |
| Sign-ins from new devices or countries | People used personal devices, and that makes a real intruder harder to spot | New devices and locations for each user. Check anything that doesn't fit. |
| Security settings changed by admins | Rules sometimes get relaxed to unblock people | Any policy that was changed during the outage. Make sure it was changed back. |
| Emails that arrived late | Hours of delayed email arrived while services were still recovering | Suspicious messages that users have reported since |
If you only have time for two, do the first two. Those are the cases where a sensitive file may still be open to people who shouldn't see it.
Also warn your users about phishing. Attackers will use the outage as an excuse. Expect emails that say something like "your mailbox will be blocked, sign in again to restore it". Tell people that Microsoft will never ask them to sign in through a link in an email.
How 1Security helps with this
1Security works alongside Defender and Purview. It does not replace them. It reads your tenant through Microsoft's own APIs and the audit log, so during the outage our data for those hours was delayed too. We want to be clear about that.
What we do have is a record from before the outage. 1Security keeps its own copy of who has access to which files, which files contain sensitive data, which sharing links exist and what users did, going back up to three years. That record is stored outside Microsoft 365, so it was still available during the outage.
That makes the checks above much faster, because you can compare today with last Wednesday:
- On the sensitive info and files screens you can see which sensitive files got a new public or external link since Thursday.
- In the activity logs you can compare what each user did during the outage with what that same user normally does.
- You can set up a policy that keeps checking the links created last week, and asks the file owner to confirm or remove any that are still open.
NIS2 and DORA
If your organisation falls under NIS2, an outage like this is a practical test of two requirements in Article 21(2): business continuity (point c) and supply chain security (point d). An auditor is unlikely to ask whether you were affected, because almost everyone was. They are more likely to ask what you did while email was down, and how you made sure your security controls were working again afterwards. DORA asks financial companies similar questions about their IT suppliers. We wrote more about that in the DORA articles auditors open first.
If you do the checks in this post, write down what you checked and what you found. That record is useful evidence.
Common questions
Was any customer data exposed by the outage? Microsoft has reported no unauthorised access. The services were unavailable, but nothing suggests they were broken into. The risk is in what your own users did to keep working.
Can we get money back from Microsoft? Possibly. The Microsoft 365 service agreement gives service credits when uptime in a month falls below 99.9 percent. You have to ask for them. Open a service request in the admin center and include the incident numbers, the times your users were affected and how many users that was. If you buy your licences through a partner, ask the partner.
Was the Teams problem on Thursday morning part of the same outage? We don't know. Microsoft recorded it as a separate incident and has not connected the two.
Do we need a backup email provider? That is a big decision, and for most companies it is not the first step. A simpler first step is to agree on a backup way to communicate that does not depend on Microsoft 365, and to have your security team approve it in advance. Then next time people won't have to improvise.


