procedural

Microsoft 365 outage (November 2026): what it means for resilience, NIS2, DORA and the EU AI Act

15 min read

The 12 and 13 November Microsoft 365 outage was not a cyberattack, but it was a governance test. What NIS2, DORA, FINMA and the AI Act ask of you now.

What happened in the Microsoft 365 outage on 12 and 13 November 2026?

At about 17:15 Swiss time on Thursday 12 November, Exchange Online started failing. Within hours, Outlook, Teams, SharePoint, Copilot and Defender XDR were unreliable or unreachable worldwide. Downdetector logged outage reports from 74 countries and territories within an hour. By 09:00 Swiss time on Friday, more than 15 hours later, Microsoft was reporting better server stability but still had no full recovery.

Microsoft has given no time frame for a full recovery. Its updates on Friday morning reported better server stability and partial recovery in some regions, but no estimate of when Outlook, Teams and the other services will be back to normal for everyone.

  • 15+ hourswithout a full recovery, by 09:00 Swiss time on Friday
  • 74countries and territories with outage reports on Downdetector within an hour
  • 6services hit: Exchange Online, Outlook, Teams, SharePoint, Copilot, Defender XDR
  • 0indications of malicious activity, according to Microsoft

The list of those affected reads like a resilience exam paper. Government offices, courts and banks in several countries lost email and fell back on phone, paper and fax. Closer to home, Richemont delayed its half-year results, scheduled for 07:00 Swiss time on Friday, before publishing them later that morning. In the US, Workhorse moved its earnings release by four days.

  1. Exchange Online starts failing (Swiss time)
  2. Outlook, Teams, SharePoint, Copilot and Defender XDR unreliable or unreachable worldwide
  3. Richemont delays its half-year results, then publishes them later that morning
  4. More than 15 hours in: better server stability, no full recovery and no time for one
  5. Microsoft’s full report, with the root causeawaited

Microsoft has said there is no indication of malicious activity. It has described servers becoming unexpectedly unavailable, which strained the rest; a process that sets up secure connections becoming overwhelmed; restored servers failing again once traffic returned; and a patch to a configuration component for authentication and routing. It has promised a full report. Until that report lands, nobody outside Redmond knows the root cause, and this article does not guess at one.

That is the point. For most organisations, the root cause will change very little about the decisions that mattered on Thursday night. Those decisions were either made in advance, or they were improvised at 23:00 by people who could not reach each other on Teams.

Status at the time of writing (13 November 2026, 10:30 CET): Microsoft lists Exchange Online (EX1550543) and the wider Microsoft 365 incident (MO1550821) as still open. In its latest update, reported by ITdaily at 10:21 CET, Microsoft says it is seeing positive results for service stability and higher capacity, and is applying its traffic measures more widely in the affected regions. It has given no time for full recovery. Check status.cloud.microsoft for the latest.

Why is the Microsoft outage a concentration risk issue under NIS2, DORA and FINMA?

The outage was not a security failure. It was a dependency made visible. When one supplier carries your email, your meetings, your documents, your identity layer and your security tooling, its bad night becomes your bad night, all at once.

One supplier under six layers of workUnreliable or unreachableSame supplier
  • EMAILExchange Online, Outlook
  • MEETINGSTeams
  • DOCUMENTSSharePoint
  • AICopilot
  • SECURITY TOOLINGDefender XDR
  • IDENTITYSign-in and accessSame supplier
Based on the services named in this article
Illustration: one supplier under six layers of daily work, with the services named in this article.

Regulators already have a name for this: concentration risk. Three regimes ask Swiss and EU organisations to manage it explicitly.

Three regimes that ask you to manage concentration risk
RegimeWho it touchesWhat it asks for, in plain terms
NIS2 (EU)Essential and important entities in 18 sectorsBusiness continuity, backup and crisis management, plus supply-chain security, as part of the risk-management measures in Article 21
DORA (EU)Banks, insurers, investment firms and their ICT providersICT third-party risk management, exit strategies and concentration-risk assessment. Microsoft is one of 19 providers the ESAs designated as critical in November 2025
FINMA Circular 2023/1 (CH)Swiss banks and securities firmsOperational resilience: identify critical functions, set a tolerance for disruption for each, and prove you can stay within it

None of these says "don't use Microsoft." All three say: know what depends on it, decide in advance how much disruption you can absorb, and test that you can.

Henri Haenni, President of Abilene Group, has spent three decades on business continuity.

"In three decades of continuity work, the recurring failure is not that organisations lack a plan. It is that the plan assumes the tools people need during a crisis will still work. When Outlook goes down, the first question is not which alternative channel to use. It is whether anyone can find the contact list, reach the decision-makers, and trigger the plan without the very system that failed. Thursday night tested all three."

Henri HaenniPresident, Abilene Group

Each of those three has a fix that costs almost nothing: a contact list held offline and on personal devices, a secondary channel agreed in advance that does not run on the primary supplier, and a named person, with a deputy, who has authority to invoke the plan. None of them depends on Microsoft's root cause.

The useful test is the tolerance for disruption. If your critical function is "client payments" or "regulatory disclosures", what is the maximum time it can be degraded before harm becomes unacceptable? Four hours? Twelve? This outage passed 15. If your number was lower and your plan assumed Microsoft would be back, the plan was a hope.

Your tolerance against this outage
  • Tolerance of 4 hoursExceededPassed at about 21:15 on Thursday
  • Tolerance of 12 hoursExceededPassed at about 05:15 on Friday
  • This outage: 15+ hoursStill openNo full recovery by 09:00 on Friday
Example tolerances; times in Swiss time
Illustration: two example tolerances against an outage that started at about 17:15 Swiss time on Thursday.

"A tolerance for disruption is often treated as an IT recovery target. It is not. It is a governance decision: how long can this critical function be unavailable before the harm to clients, regulators, staff or the business becomes unacceptable? In my experience, the organisations that recover best are not those with the most sophisticated technology. They are the ones where management has already decided that answer, rehearsed it, and knows who can invoke it at 23:00."

Henri HaenniPresident, Abilene Group

In practice, that decision fits on one page: each critical function, its tolerance in hours, its owner, and who can invoke the fallback. If that page did not exist on Thursday, writing it is the most valuable thing to come out of this outage.

How does a Microsoft 365 outage affect AI systems and the EU AI Act?

Copilot went down with Exchange. So did every workflow that quietly relies on it: the meeting summaries, the drafted client replies, the triage of the support inbox. Third-party AI tools that connect to Microsoft 365 also reported failures, because their connectors had nothing to connect to.

  1. 1The workmeeting summaries, drafted client replies, support inbox triage
  2. 2The AI systemCopilot or a third-party assistant
  3. 3What it readsExchange Online and SharePoint, direct or through a connector
  4. 4One supplierMicrosoft 365
The dependency chain most AI registers leave out: when the last link fails, the first one stops

Most organisations' AI inventories list the model and the vendor. Few list what the AI system depends on to work at all. On Thursday, that missing column mattered more than the model.

To be precise about the law: this outage did not trigger any obligation under the EU AI Act. Copilot drafting emails is not a high-risk system. But the Act's logic is already worth applying, and the deadline is closer than it looks. Following the Digital Omnibus, obligations for stand-alone high-risk systems in Annex III (recruitment, credit scoring, critical infrastructure and others) apply from 2 December 2027. Those systems must be robust and resilient against errors and faults, including in the environment they run in.

Alexis Hirschhorn, CEO of Acuna GRC, works on exactly this seam between third-party risk and AI governance.

"When I ask organisations how many AI systems they use, I usually get a number. When I ask how many of those systems stop working when a single cloud supplier fails, the room often goes quiet. That second question is the one that matters for resilience. An AI inventory that records the model and vendor but not its underlying dependencies gives you a false sense of control."

Alexis HirschhornCEO, Acuna GRC

The second number is hard because the dependencies are rarely direct. Copilot needs Exchange and SharePoint to have anything to work on. A third-party assistant reaches your mail through a Microsoft 365 connector. A screening tool may pull CVs from a shared mailbox. None of those links appears in a register that lists only the model and the vendor, and all of them broke on Thursday.

"The EU AI Act will not solve an outage, but it will make dependency visible. If a high-risk AI system relies on one cloud stack, its availability is part of its risk profile. You cannot credibly document robustness for a system whose dependencies you have not mapped."

Alexis HirschhornCEO, Acuna GRC

That matters most where the stakes are highest. A recruitment or credit-scoring tool listed in Annex III will, from December 2027, be expected to show robustness, and deployers will need trained people overseeing it. If nobody knows what those people do when the tool is unavailable, the oversight exists on paper only. So what would a register need to hold to answer that in ten minutes?

"The practical fix is not exotic. Add three fields to every AI system in your register: the supplier or platform it depends on, the business process it supports, and the manual procedure people follow when it is unavailable. If you cannot complete the third field, you have not mapped the risk. You have found your next tabletop exercise."

Alexis HirschhornCEO, Acuna GRC
AI registerSample
Three fields for every AI systemBeyond the model and the vendor3 new fields
AI-01CopilotExchange Online, SharePointMeeting summaries, client repliesDocumented
AI-02Third-party assistantMicrosoft 365 connectorSupport inbox triageTo be tested
AI-03CV screening toolShared mailboxRecruitment (Annex III)Missing
An empty third field is your next tabletop exercise.
Sample data
Illustration: the three fields to add to every AI system in your register, with sample data.

A sensible order of work: start with any AI system that touches an Annex III use case, then those embedded in client-facing processes, then the productivity assistants. Fill the three fields for each, then test the third one in your next continuity exercise rather than assuming it works.

Do you have to report the Microsoft outage under NIS2, DORA or Swiss law?

The honest answer is: it depends on the impact on you, not on what happened at Microsoft. Three myths came up repeatedly on Friday morning.

  • Myth 1: "It wasn't a cyberattack, so there's nothing to report." Partly true in Switzerland, wrong in the EU. The Swiss duty to report to the National Cyber Security Centre (NCSC, known in German as BACS) covers cyberattacks on critical infrastructure. A non-malicious outage most likely falls outside it. NIS2 and DORA are different: they look at incidents that compromise the availability of your services, whatever the cause.
  • Myth 2: "It's Microsoft's incident, so Microsoft reports it." Microsoft's obligations are Microsoft's. Yours depend on what happened to your critical services. If your client-facing operations were severely disrupted, the incident may be significant for you even though you did nothing wrong.
  • Myth 3: "We'll decide once we know the root cause." Clocks start from detection or classification, not from the vendor's post-mortem. Waiting for Redmond is not a classification method.
Reporting duties by regime, and whether they are likely to apply to this outage
RegimeReportable if…First deadlineLikely to apply here?
Swiss Information Security Act (NCSC)A cyberattack hits a critical-infrastructure operator24 hoursUnlikely: no malicious activity reported
NIS2The incident causes, or could cause, severe operational disruption or financial loss to your servicesEarly warning within 24 hoursPossible, if your essential services were materially disrupted
DORAThe ICT incident meets the major-incident classification criteria (critical services affected, plus clients, duration, geographical spread, data losses, reputational or economic impact)Initial notification within 4 hours of classification, 24 hours of detection at mostPossible for financial entities with critical functions on Microsoft 365
FINMAAn incident of substantial importance to supervisionImmediatelyAssess: depends on impact on critical functions
GDPR / nDSGPersonal data became unavailable in a way that creates a risk for individuals (GDPR) or a high risk (nDSG)72 hours (GDPR) / as soon as possible (nDSG)Usually document only; notify if the unavailability created real risk

Whatever you conclude, write it down: who assessed, when, against which criteria, and why you did or did not report.

A documented "no" is defensible. An undocumented one is a finding waiting to happen.

This section is general information, not legal advice. Check the thresholds that apply to your entity with your counsel.

How should a company respond to a Microsoft 365 outage? A case study

To make this concrete, picture two Swiss private banks of the same size, both running on Microsoft 365, both supervised by FINMA, both with an EU subsidiary. This is an illustrative scenario built from common patterns, not a real client.

Firm A has a continuity plan in a SharePoint folder. When Exchange fails shortly after 17:15, the plan is unreachable. The CISO texts the COO from memory. Nobody knows which client processes depend on Teams, so the team spends the evening discovering them. At 06:00, the compliance officer asks whether the incident is reportable. Nobody has the criteria to hand, so the answer is "we'll see."

Firm B made six decisions earlier in the year. Here is how its night ran.

Firm B's night, and the decision behind each step
Time (CET)What Firm B didThe decision made in advance
17:45Crisis team convened on a pre-agreed Signal group, with numbers held offlineA secondary channel that does not depend on the primary supplier
18:10Pulled the dependency map: 6 critical functions, 4 relying on Microsoft 365Critical functions mapped to the suppliers behind them
21:00Switched client payments confirmation to a documented manual procedureA manual workaround for each function, rehearsed once
23:30Checked elapsed time against each function's tolerance: 2 of 4 still within limitsTolerances for disruption signed off by the board
06:30Classified the incident against FINMA, DORA and NIS2 criteria, recorded the decision and rationaleReporting criteria and owners written down before they were needed
08:00Warned staff about outage-themed phishing ("re-enter your Microsoft password")A pre-drafted staff message for supplier incidents

Firm A: improvised on the night

  • Continuity plan in SharePoint, unreachable
  • Contacts texted from memory
  • Dependencies on Teams discovered during the evening
  • Reportable? "We’ll see"

Firm B: decided earlier in the year

  • A secondary channel off the primary supplier
  • Critical functions mapped to their suppliers
  • Manual workarounds, rehearsed once
  • Tolerances signed off by the board
  • Reporting criteria and owners written down
  • A pre-drafted staff message

Firm B did not have a better supplier. It had a better map. The difference was not technology but preparation that lived somewhere other than the system that failed.

That last point is practical. Firm B kept its supplier register, dependency mapping, continuity plans and reporting criteria in its GRC platform, hosted outside its Microsoft tenant. In our case, that is the setup we run on Acuna GRC, and it is the setup we recommend whichever tool you choose: your crisis playbook should never sit inside the service most likely to fail.

The test is simple: if your primary suite went dark tonight, could you open your supplier register, your continuity plan and your reporting criteria within ten minutes?

What should you do after a Microsoft 365 outage? A 10-point checklist

None of these depends on Microsoft's root cause. Run them while the memory is fresh and people still remember what they improvised.

  1. Map it. Which of our critical functions stopped or slowed on 12 and 13 November? List them by name.
  2. Trace it. For each one, which suppliers sit behind it, including identity, security tooling and AI assistants?
  3. Measure it. How long was each function degraded, and was that inside the tolerance the board approved? If there was no tolerance, set one.
  4. Reach each other. Did the crisis team have a working channel and an offline contact list within 30 minutes?
  5. Find the plan. Could people open the continuity plan while Microsoft 365 was down?
  6. Work around it. Which functions had a manual fallback, and did anyone actually use it?
  7. Classify it. Did we assess reportability under NIS2, DORA, FINMA and data protection rules, and record the reasoning?
  8. Warn people. Did staff get a phishing warning? Outage-themed credential lures typically follow within hours.
  9. Update the AI register. Which AI systems failed with the outage, and does the register now show what each depends on?
  10. Brief the board. Has the board seen a one-page account of impact, decisions and gaps, in plain language?

Two follow-ups should wait for Microsoft's post-incident report: whether your exit strategy for Microsoft 365 is realistic, and whether contract terms on availability and incident information need renegotiating. Diarise them now so they do not drift.

What should boards ask after a major cloud outage?

Boards do not need the technical detail. They need answers they can hold management to.

  1. Which of our critical functions depend on a single supplier, and do we accept that concentration knowingly?
  2. What disruption can each critical function tolerate, and did we stay within it this week?
  3. If this outage had lasted three days, what would we have done on day two?
  4. Who decided whether to report, on what criteria, and where is that decision recorded?
  5. Which AI systems do we rely on, and what happens to the work when they stop?

Thursday was a free rehearsal. The organisations that use it to answer these five questions will not need the next outage to be expensive.

Related reading: Freezers at 50 °C, on another night when a supplier-run dependency failed, and AI agents at the bank login, on authorization and evidence for AI systems.

Request a demo

Sources

Incident facts come from Microsoft's service health notices for incidents EX1550543 and MO1550821 and from the public sources below, all accessed on 13 November 2026. Henri Haenni's LinkedIn profile is named without a link.

Frequently Asked Questions

When will Microsoft 365 be fully back to normal?

Microsoft has not given a time frame. At the time of writing (13 November, 10:30 CET) it reported better stability and partial recovery but no estimate for full recovery. Check status.cloud.microsoft for the latest.

Was the Microsoft 365 outage on 12 and 13 November 2026 a cyberattack?

Microsoft has said there is no indication of malicious activity. It has not yet published a root cause and has promised a full report.

Do I have to report the Microsoft outage under NIS2?

Possibly. NIS2 looks at the impact on your services, not the cause. If your essential services suffered severe disruption, assess it against the significant-incident criteria and document the outcome.

Does DORA apply to a Microsoft 365 outage?

Yes, for financial entities. DORA covers ICT incidents originating at third-party providers, and Microsoft is a designated critical ICT third-party provider. Classify the incident against DORA's major-incident criteria.

Do Swiss organisations have to report it to the NCSC?

Probably not. The Swiss reporting duty covers cyberattacks on critical infrastructure, and no malicious activity has been reported. Other duties, such as FINMA's, may still apply.

What is concentration risk?

The risk that many critical functions, or many organisations, rely on the same provider, so that a single failure causes widespread disruption.

Does the EU AI Act apply to a Copilot outage?

Not directly. But from 2 December 2027, high-risk AI systems listed in Annex III must be robust and resilient, so mapping what your AI systems depend on is worth starting now.

Where should a business continuity plan be stored?

Somewhere that stays reachable when your primary productivity suite is down: offline copies, a separately hosted platform, or both.