An illustrated cloud network shows connected servers and buildings, with a central server warning and one blocked data connection.
A cluster of high-profile outages hit US organizations between November 12 and 17, 2026. The Microsoft 365 failure is the one best documented, and it matters most to IT administrators. The wider claim that AI, cyberattacks and corporate negligence "converged" is not supported by the public evidence. The better lesson is about shared dependencies and how fast systems are being changed.

An illustrated cloud network shows connected servers and buildings, with a central server warning and one blocked data connection. The short version​

  • Microsoft, the FAA, United Airlines, JPMorgan Chase, Apple, Marriott and the NSA each reported separate problems over about six days.
  • Every organization that has given a cause has pointed to something internal, such as a software defect, a failover, an update or maintenance.
  • A Cloud Security Alliance review of November 18 notes that none has blamed an adversary. It also says the Director of National Intelligence saw no sign of a coordinated cyberattack.
  • Temporal proximity is not evidence of a common cause.

Microsoft 365: the best-documented failure​

Microsoft's own account is detailed. Its preliminary post-incident report records the window as November 12, 2026, at 16:20 UTC through November 13 at 16:10 UTC, or 23 hours and 50 minutes. The incident IDs were EX1550543 and MO1550821.

Exchange Online took the worst of it. Most users could not reach their mailboxes through Outlook or other mail clients, and email delivery was delayed for many. The report lists many other affected services:

  • Copilot
  • Teams
  • SharePoint
  • OneDrive
  • Defender
  • Purview
  • Fabric
  • Intune
  • Power BI
  • Universal Print

Severity varied by region, connection method and time. Services that rely on Exchange data, such as search, calendar and presence, stayed affected longer than Teams, SharePoint and OneDrive. Those three recovered earlier because their traffic was moved off the affected gateway.

Three ordinary conditions, one outage​

Microsoft's preliminary report points to a combination of factors, none of which caused an outage alone:

  1. A latent defect. It came from a change enabled on Microsoft 365 networking-gateway servers. Reporting describes it as a months-old Windows server defect. The rollout began in July and had run for more than three months without visible impact.
  2. Reduced capacity. Nearly 14% of the gateway's server capacity was offline as part of routine maintenance.
  3. A Teams failover. At 01:07 UTC on November 12, automated failover moved Teams traffic onto the gateway front-end servers shared with Exchange Online. Load rose gradually, and broad failures began as US traffic approached its daily peak.

Microsoft says the defect and the update that introduced it do not apply to broader Windows deployments. Windows PCs were not the problem.

Why recovery was slow​

According to the report as summarized by 365Fanatic, the failure cascaded:

  • At the low point, more than 80% of North American servers were unresponsive.
  • Traffic then spilled into other regions that carried the same defect.
  • Servers that came back were immediately overwhelmed by waiting traffic and failed again.
  • Because the defect was months old, the usual safeguards could not undo it quickly. Those safeguards include stopping a rollout, rolling back and disabling a new feature.
  • Investigators found that a connection-establishing process on the servers was becoming overwhelmed, so the servers stopped processing requests.
  • Microsoft built an update that reverted the defective change and began applying it at 07:20 UTC on November 13.
  • The fix needed five steps per server: take it out of service, restart it, patch it, restart again, and return it to service.

Microsoft says its normal load testing missed the problem because test labs used older, lower-core machines and different traffic. The contention was worse on newer, higher-core hardware.

The final report and the repair list​

Microsoft's final report is dated November 20. According to 365Fanatic's summary, it says there was no evidence of malicious activity and that AI tooling did not make the change that caused the defect. No customer action is needed.

Completed items include:

  • Reverting the faulty change.
  • Baking the fix into the Windows image used for these servers.
  • Adding capacity.
  • Extending the email retry window from 24 to 48 hours.
  • Adding temporary guardrails against automatic traffic redistribution.

Planned items include:

  • Smaller failure zones.
  • Separate gateway pools for different workloads.
  • Failover for Teams only in the affected region.
  • Gradual traffic ramp-up when servers rejoin.
  • Load testing on the newest hardware at production-like peaks.
  • Detection of extended capacity loss.
  • Faster customer notification. The first notice came about 20 minutes after broad impact.

Dates for these run from Q4 2026 into Q1 2027, and some are still undated.

According to the same summary, no mail was lost and queued messages were delivered. Delayed messages to consumer mailboxes inactive for more than 30 days did expire, and their senders received non-delivery reports. Customer data at rest was not affected.

The other incidents​

The details below come from the Cloud Security Alliance review, which draws on secondary reports. Several causes are preliminary or only reported.

DateOrganizationStated or reported cause
Nov 13FAA ground stopA contractor-installed router during a copper-to-fiber conversion, together with a second site, flooded legacy equipment with connection attempts. The investigation was ongoing.
Nov 16JPMorgan ChaseA software bug in on-premise hardware systems. Delays reached other banks through transactions with Chase.
Nov 16United AirlinesBriefly grounded departures. No cause given in the review.
Nov 17AppleAn internal update.
Nov 17NSA classified networkDegradation during scheduled routine maintenance. The full cause was not established.

Marriott also reported sign-in problems on November 17. The review does not cover it, so no cause is given here.

The review also lists Workday and AWS incidents in the same period. They are outside the story as submitted, so they are not folded into this account.

What is and isn't supported​

Not supported:

  • That AI caused these outages. Microsoft explicitly said its change was not made by AI tooling.
  • That patches for AI-discovered vulnerabilities caused the cluster. The review calls that specific claim unsupported. It rests on anonymous sourcing, and several on-record causes don't fit it. Microsoft's July-origin defect is one example.
  • Claims about specific foreign operators exploiting these outages. No incident-specific evidence for them appears in the reporting reviewed.
  • "Corporate negligence" as a finding. It is a judgment, not something any investigation has established.
  • End-of-year budget or holiday pressure as explanations. These are plausible as general risk factors, but no source ties them to any of these incidents.

Supported as general lessons:

  • Failover and recovery mechanisms can expose or worsen faults. In Microsoft's case, failover worked as designed and exposed the dormant bug.
  • Shared dependencies carry failures outward. The review notes Anthropic's Microsoft 365 connector failed for as long as Microsoft's outage lasted.
  • The review argues that rising patch volume makes change velocity a security-relevant risk in its own right.
  • Early "no evidence of an attack" statements are not findings. Both premature suspicion and premature exoneration carry costs.

These are the Cloud Security Alliance's analytical points, not established causes of any individual outage.

Practical takeaways for admins​

  • Check your exposure from your own logs. Review authentication logs, emergency access grants and help-desk reset volumes for November 12 to 17. The review suggests treating spikes in account-recovery messages as possible phishing until shown otherwise.
  • Revoke temporary exceptions. Confirm any break-glass accounts or policy exceptions created during the outage have been removed.
  • Map failover headroom. Ask whether your failover targets can absorb redirected load when they are already partly degraded. Microsoft's report shows that condition mattered.
  • Test realistically. Match production hardware and traffic mix, including peak-load tests, as Microsoft now plans to do.
  • Give automation limits. Connectors and AI agents that call SaaS platforms need bounded retries, circuit breakers and defined degraded modes. Retry storms can add to the load during a recovery.
  • Stage security patches. Use canary populations and automated rollback for security changes, not only feature changes.
  • Ask providers for detail. The review argues contracts with critical SaaS providers should require technical post-incident reports detailed enough for customers to assess their own exposure.

The bottom line​

The strongest thread across these incidents is not a single villain. It is that ordinary changes, resilience mechanisms and shared dependencies can combine in unexpected ways under production conditions. Microsoft's report shows a months-old change surviving every safeguard until three routine conditions lined up.

Some of the other investigations, including the FAA's and the NSA's, were still unfinished at the time of the sources reviewed. Treat any claim of coordination, AI causation or negligence as unproven until they report.

 

References

  1. Microsoft 365 outage 12 November 2026: final report | 365Fanatic 365fanatic.com
  2. America's technology outages reflect a convergence of AI risks, cyberthreats, and corporate negligence - Milwaukee Independent Milwaukee Independent 2026-11-21T11:00:11+00:00
  3. The November Outage Cluster: Correlated Fragility and Attribution – Lab Space labs.cloudsecurityalliance.org