An IT operator monitors cloud servers, maintenance tasks, and service alerts in a control room.
Last week, critical systems failed one after another. Microsoft 365 went down for much of a day. The FAA stopped departing flights across the country. United Airlines, Apple and JPMorgan Chase had disruptions, and the National Security Agency lost an internal network it uses for classified communications. Governing called the stretch so bad that Mercury retrograde seemed like a fitting explanation.

The more useful finding is less mysterious. None of these failures has been reported as a cyberattack. Governing reports that many of them started while organizations were updating, maintaining or replacing systems. For state and local IT teams, the main lesson is that a routine maintenance window can still take down important services. The trick is getting ready for that before it happens.

An IT operator monitors cloud servers, maintenance tasks, and service alerts in a control room. The week in outages​

Governing gathered several separate incidents. None has been shown to share a cause with another:

  • Microsoft 365: Exchange Online and many other Microsoft 365 workloads were unavailable to many customers for almost a full day.
  • FAA: Problems with the agency's communications systems led to a nationwide ground stop.
  • United Airlines: A technology problem caused delays and cancellations, separate from the ground stop.
  • Apple: Service went down after an internal update.
  • JPMorgan Chase: The largest U.S. bank had trouble processing transactions.
  • NSA: An internal network used for classified communications went offline during scheduled maintenance.

The incidents happened close together, but timing doesn't prove a connection. As far as the public record shows, these were separate failures that happened in the same week.

Section summary: Six major organizations had outages in one week. None has been publicly tied to an attack, and several are linked to routine change work.

What actually broke at Microsoft​

Microsoft has published a preliminary post-incident report under incident IDs EX1550543 and MO1550821. 365Fanatic has also summarized it. The report says impact began on November 12, 2026, at 16:20 UTC and ended on November 13 at 16:10 UTC.

Exchange Online was hit hardest. Most affected users couldn't reach their mailboxes in Outlook or other mail clients, and many saw delayed mail. Microsoft also lists Copilot, Teams, SharePoint, OneDrive, Defender, Purview, Fabric, Intune, Planner, Power Apps, To Do, Power BI and Universal Print as affected. Most of those services stayed online, but features that depend on Exchange did not work properly. Impact varied by region, connection method and time, and some users had partial or intermittent access.

Governing describes the cause as several flaws combining, including a bug introduced in July. Microsoft's report identifies three factors:

  1. A latent defect from July. A fix built for Microsoft 365 networking gateway servers contained a bug. Under certain load conditions, it caused lock contention inside an internal Windows process and used up CPU and memory. The fix had been running worldwide for more than three months without trouble. Microsoft's test labs focused on weaker hardware with fewer cores, but the problem is worse on newer machines with more cores.
  2. A surge in load. About 15 hours before the outage, Teams automatically failed over from an alternate gateway to the primary gateway, which also serves Exchange Online. Failover is part of the resilience design, but it increased load and changed the traffic pattern.
  3. Reduced capacity. Almost 14% of gateway capacity was already out of service because of an unrelated problem from the previous day. On its own, that loss stayed within the available headroom.

These factors combined during U.S. peak hours. Servers stopped responding, and automated traffic management sent users to healthy servers, which then hit the same defect. Within two hours the failure had spread across the fleet. At the worst point, more than 80% of servers were unresponsive. The partitions that normally contain a failure like this didn't stop it, because of a separate issue.

Teams, OneDrive, SharePoint and the Office apps could fail away from the broken gateway and recovered earlier. Exchange Online couldn't do that at its scale.

Recovery was slow. Servers that came back were flooded by waiting traffic and failed again. Because the defect was months old, Microsoft couldn't quickly reverse it with its usual rollbacks or feature flags. Microsoft eventually built an update that returned the affected code path to its previous working state and started applying it at 07:20 UTC on November 13.

One scope note for Windows admins: Microsoft says the defect and its fix applied only to Microsoft 365 gateway servers, not to Windows deployments in general. There is nothing to patch on your own Windows Server fleet because of this incident. Microsoft also said tenants did not need to change anything.

The report is preliminary. Microsoft says a final report will replace it, and the full list of repair items was planned for November 20. Until then, treat the root-cause details as Microsoft's current account, not its final word.

Section summary: Microsoft's outage came from a months-old defect, a normal failover that increased load and an unrelated loss of capacity, all happening together. Ordinary Windows machines were not affected.

The FAA ground stop: one router, a lot of traffic​

Governing quotes the FAA as calling the disruption the result of a routine router replacement at a remote facility. According to the agency, a site in Texas or Arkansas, possibly both, flooded its telecommunications network with about 1,000 connection attempts every four to five seconds.

Other reporting fills in the picture. CNBC reported that the ground stop lasted for more than four hours Friday, marking the longest halt to U.S. flight departures since 9/11. It happened shortly after L3Harris Technologies, the primary contractor supporting the FAA telecommunications network, installed a new router at a facility in El Dorado, Arkansas. According to Axios, after L3Harris blocked network traffic from both facilities, the system stabilized, allowing normal flight operations to eventually resume, per the FAA.

The response so far:

  • Router installations paused. ABC News reported that the Federal Aviation Administration is halting the installation of Juniper routers following last week's rare nationwide ground stop. CNN reported that HPE, the parent company of the router maker, said the equipment was working as configured.
  • A large rollout under review. According to CNN, senators said FAA contractor L3Harris plans to install 3,648 Juniper routers, with 400 installed so far and just 69 operational.
  • Questions about oversight. In an exclusive based on unnamed sources, CNN reported that L3Harris notified the FAA on November 9 that it planned to carry out maintenance, replacing a router, at a remote FAA facility in El Dorado, Arkansas. The same sources said that by that Friday afternoon, the new router was connected and the L3Harris contractor had left the site with no idea of the chaos that was beginning to unfold. ABC News separately reported that FAA Administrator Bryan Bedford said agency staff were present during the work.
  • Progress on the investigation. CNBC reported that engineers have been able to replicate the situation that was overwhelming parts of the air traffic system, according to people familiar with the investigation. The root cause has not been officially determined.

A quick correction to Governing's account: Sean Duffy is the U.S. Transportation Secretary, not an FAA secretary. Governing reports that he has called for major system updates and described air traffic control systems as deeply outdated. CNBC reported that Bedford said that router updates were a matter of regular business for his agency and distinct from modernization allocations. In other words, this was not a failed modernization project. It was routine maintenance on an aging network that turned out to be fragile.

Section summary: A routine router swap in Arkansas is the main focus of the FAA investigation. The agency has paused installations, and the official root cause is still pending.

The scammers moved fast​

Governing reports that Randy Rose, vice president of security operations and intelligence at the Center for Internet Security (CIS), said attackers set up fake websites made to look like Microsoft URLs. Those sites offered "fixes" that installed malware if run. The report doesn't name domains or say how widespread the activity was.

The guidance is simple. The fix for a Microsoft cloud outage runs on Microsoft's servers, not on your PC. No legitimate repair for a service-side incident needs you to download and run an executable. Use these channels instead:

  • The Service health page in the Microsoft 365 admin center, including History for closed incidents such as EX1550543 and MO1550821
  • Microsoft's public Microsoft 365 status page
  • Your organization's normal internal IT communications

If a "fix" arrives by search ad, unsolicited email or an unfamiliar URL, treat it as hostile.

Patch or wait? Why AI makes that harder​

Rose said outages are a normal part of IT work: "Even just going in and doing a normal upgrade, because of the complexity of some of these environments, it can cause outages, and that's a very normal thing in the IT world." He added that they often happen during maintenance windows.

The difficulty is that waiting to patch has become riskier too. Rose said AI contributed to more than four times as many reported software vulnerabilities this October as in October last year. That figure comes from his briefing remarks and is not an independently published measurement. Phyllis Lee, vice president of content development at CIS, said many of these vulnerabilities are turning up in the kind of old, decades-old software that governments often run.

"Often organizations are told, 'If it's not broken, please don't fix it,' because we are providing critical services to our citizens, which is understandable," Lee said. But with AI speeding up vulnerability discovery, she argued, agencies need to work out how they will upgrade and patch their systems.

Both risks are real. Microsoft's July fix ran safely for three months and then failed under load. Deferring patches on legacy software leaves known holes open. Avoiding change doesn't remove risk. It means choosing a different risk.

Section summary: Updates can cause outages, and skipping them can expose systems to faster-found vulnerabilities. Agencies need to plan for both.

A practical checklist for state and local IT​

CIS gave several recommendations at the briefing. Here they are, expanded with practices drawn from general industry experience, which are noted as such:

  1. Inventory third-party dependencies. List which citizen-facing and internal services depend on outside software or SaaS. Lee's suggestion was to plan for a vendor or product becoming unavailable. Microsoft's incident shows why the list should include indirect dependencies: features that rely on Exchange broke even in services that otherwise stayed up.
  2. Build backup communications. Lee's example was running both Teams and Slack so one is available if the other fails. Industry practice: make sure the backup isn't hosted on the same platform it's meant to replace, and test it before you need it. A phone tree also counts.
  3. Keep manual fallback procedures. Rose pointed to the water sector, which often has manual processes it can switch to when digital systems go offline. Industry practice: write those procedures down, keep paper copies and run drills so staff know how to use them.
  4. Check what a patch could affect. Before an update, identify which critical services could be disrupted and how they will keep running. That's CIS guidance. Industry practice: use staged rollouts, pilot rings and a tested rollback plan, and avoid peak demand periods.
  5. Watch your contractors' change work. CNN's reporting on the FAA raises questions about how closely a contractor's change was monitored. Industry practice: require change notices, monitor during and after the change, and give someone clear authority to undo a change that goes wrong.
  6. Read vendor post-incident reports. Admins can find Microsoft's preliminary and final reports in Service health history. The final report should show what Microsoft is changing, which can inform your own resilience planning.

Lessons from the week​

The major failures last week were mostly ordinary changes that went wrong in complex systems: a months-old fix, a contractor's router swap, an internal update and scheduled maintenance.

That's good news and bad news for public agencies. There was no hidden attack to fear. The same kinds of risks exist in every agency's next change window. The governments that handle the next outage best will know which services they depend on, keep working alternatives ready and patch in ways that leave room to recover.

 

References

  1. State, Local Government Can Take Lessons From a Week of Tech Outages - Governing Governing 2026-11-18T00:00:00+00:00
  2. Microsoft 365 outage 12 November 2026: post-incident report | 365Fanatic 365fanatic.com
  3. FAA ground stop investigation: Engineers replicate network overload cnbc.com