Between Thursday and Monday, three things most people depend on stopped working. Microsoft 365 went down across dozens of countries. The FAA ordered a nationwide ground stop. Customers of several large banks could not log in or move money.

Nobody has found a thread connecting them, and no one has pointed at an attacker. That is not a disappointing answer. It is the interesting one. Each of the three started inside the organization it hit, and in two of them the starting point on the record is a change to the network.

What Microsoft says happened

Microsoft’s preliminary report traces the outage to an update applied in July to the networking gateway servers behind Microsoft 365. The update carried a defect that could overwhelm a server under certain traffic conditions. It passed load testing, because the test hardware had fewer cores than production and the defect gets worse with more cores. Then it ran without trouble for more than three months.

On November 12, Teams traffic automatically failed over onto the gateways that serve Exchange Online, while nearly 14 percent of their capacity was already offline for an unrelated reason. Together they set off the defect. At the low point, more than 80 percent of the gateway servers had stopped responding, and traffic redirected to the servers still standing overloaded them in turn. Customers were affected for just under 24 hours.

Two details are worth holding on to. The change that introduced the problem was made months before anything went wrong. And what spread the failure was the system’s own resilience: traffic moving off failing servers onto healthy ones until those failed too. Microsoft has promised a final report, so this account may yet be sharpened.

What the FAA says happened

A new router went into a radio site in El Dorado, Arkansas. Shortly after, the network saw about a thousand connection attempts every four to five seconds, and the FAA’s legacy equipment became unstable under that load. Communications and radar feeds were disrupted at 17 of the agency’s 22 en route centers. The fix was to isolate the two sites involved, after which the network stabilized.

The change was routine, the site was remote, and the reach was national. We wrote that one up on its own: the FAA ground stop began with a routine network change.

What Chase says happened

On Monday, customers of Chase, Bank of America, Wells Fargo, and U.S. Bank reported failed logins and delayed transfers, including Zelle payments. Some Chase customers could not use a card at an ATM or finish a transaction in a branch. JPMorgan Chase said the cause was “a software bug in on-premise hardware systems”, that there was no evidence of anything cyber-related, and that customers still had access to their accounts and their money. On Monday night it said transactions were processing normally again.

The bank also described how it spread, and that description is the part worth reading twice: other banks were affected only to the extent that they were party to a transaction with Chase. Zelle’s operator said its network was operating as it should and pointed back at the banks. U.S. Bank said its own systems were not directly affected, while warning that payments to or from another institution could be delayed. Nearly every system named in that paragraph was healthy, and people still could not move their money.

What the three have in common

None of them was an attack. None was a device that simply died. Each began inside the organization it hit: two of them with a change to the network, and the third with a software bug in a bank’s own on-premises systems.

In the two that began with a change, the change passed its own checks. Microsoft’s update cleared load testing. The FAA’s router came up, and its maker says it worked as intended. That gap is where the damage lives. A change can succeed on its own terms (the device comes up, the upgrade completes, the ticket closes) while the state of the network around it has moved. Sometimes the new condition is latent for months, as at Microsoft, and surfaces only when traffic shifts. Sometimes it takes minutes, as in Arkansas.

The second thing the three share is how far the trouble traveled. One router at a remote site in Arkansas disrupted 17 of 22 en route centers. Overload spread from one set of gateway servers to the next across dozens of countries. A problem at one bank stopped payments for customers of banks that were working normally. In each case, the reach was set by what depended on the thing that broke, not by the size of the fault. Most teams have a clear picture of the first and a vague one of the second.

The question a change needs to answer is not “did it work?” It is “is the network still doing what it did before, and can I show that?”

Six more outages, and one attack

Six more outages belong beside these, all between November 13 and 17. Three have a cause on the record, and three do not.

On Friday, November 13, Workday took its service offline for customers served from two of its data centers, and said it was restored about 18 hours later. Its notice to customers describes a check of its disaster recovery environment during a weekly maintenance window: services at the disaster recovery site had connected to production databases and left data inconsistent. Workday said it was not a cybersecurity incident. It began with a planned operation on the system meant to protect production.

On Monday afternoon, United Airlines reported what it called a “widespread system outage” affecting its departures. It said the problem was resolved at about 6 p.m. Eastern, less than two hours after pilots first reported it, and that it had no indication of a cybersecurity incident. United has not named a cause, so it stays out of the count. A pilots’ union memo said crews could not get the final aircraft weights they need before takeoff. Whatever the cause turns out to be, departures stopped because pilots depended on that data.

On Tuesday, November 17, sign-in to Apple accounts, iCloud, the App Store and Apple Music were slow or unavailable, and Apple’s own stores could not complete purchases. Apple said the cause was an internal update, not a cyberattack, and that most systems were back. It has not said what the update changed. That makes five outages with a cause on the record, and all five began inside the organization.

The same day, the National Security Agency said it had experienced “temporary degradation during scheduled routine maintenance”. News reports said the classified network its staff use to communicate internally was down, and that arrivals at its headquarters were staggered. The agency said mission-critical capabilities stayed fully operational, and the White House said it saw no evidence of a cyberattack. The agency has not said what in the maintenance went wrong. The Director of National Intelligence did, on television that afternoon: systems that were being patched had not come back up as expected. It is the sixth of these outages with a starting point on the record, and the sixth to begin inside, this time during the agency’s own planned maintenance.

Two more came with no cause at all. On Sunday, November 15, sign-in and several other services in the Northern Virginia region of Amazon Web Services were disrupted for about three hours, and AWS has not published a cause. On Tuesday, November 17, sign-in to Marriott’s website and app failed for several hours, and guests reported they could not check in or use a digital room key. Marriott called it a technical issue and said its review found no sign of a cybersecurity cause. Like United, both stay out of the count.

An attack did happen in the same days, and it was a different kind of event. On November 16, malicious code reached visitors to the websites of 178 organizations through Optimizely, a testing service those sites load. Optimizely confirmed it, fixed the flaw and removed the code. It was not an outage, so it is not counted here. Its reach was set the same way, by how many sites depended on one shared component.

What the government says happened

On Wednesday, November 18, the Director of National Intelligence, Jay Clayton, told the House Homeland Security Committee that these were “separate service outages associated with maintenance and technical changes”, and not a coordinated or state-sponsored attack. He said there was no indication the events were related. The deputy director of the FBI told the same hearing there was no sign of malicious cyber activity in any of them.

The day before, Clayton had said what kind of work he meant: systems being patched and upgraded, which means taking them down and bringing them back up. Senator Mark Warner, the vice chairman of the Senate intelligence committee, said the evidence so far points to efforts to secure systems, not to malicious activity, and that pressure to patch immediately is straining systems that were not built for that pace of change.

That is an account of the week, not a finding on each outage. Only the National Security Agency’s has been tied to a patch on the record, and three of the nine outages here still have no public cause. It does match what the owners have said themselves. Five of the six causes on the record began with planned work: two updates, a new router, a recovery test and a maintenance window.

Where REAP fits

This is change management. A proper change procedure, with checks before and after the work and a rollback trigger agreed in advance, is what shortens the distance between a change and the discovery that it hurt something. Raja Tadimeti, REAP’s CEO, made that case at ONUG Fall 2026 in The Change Succeeded. Did the Network?

REAP assesses a proposed change against topology, reachability, security policy, and known CVEs before the window. That is available today. After the window, it compares configuration and state with what was captured before and puts the evidence in the change ticket. We deliver that in pilots today, on one real change window with your team. See Change Assurance.

What a network team can take from this week

Assume the next bad week starts with your own change, not an attacker’s. Six of the nine outages above have a starting point on the record, and none of the six involved one, though the same days did see an attack of a different kind. The government’s account of the week is maintenance and technical changes. Capture the state of the network before a change and compare it afterward, including the neighbors and the services that ride over it, not only the device you touched. Agree on what would make you roll back before you start. Know what depends on the thing you are about to touch, because that, rather than the size of the change, decides how far the damage reaches. And when something breaks somewhere else, ask what changed first: it was a different site in Arkansas, and an update made months earlier in Microsoft’s case.

A note on the facts. Updated November 19, 2026: Amazon Web Services and Marriott are added, with what the Director of National Intelligence, the FBI and Senator Warner said on November 17 and 18. The account of the Microsoft outage comes from Microsoft’s preliminary post-incident report, as reported by BleepingComputer on November 16, 2026; a final report is expected. The account of the ground stop comes from the FAA’s questions and answers and the reporting cited in our earlier post. The account of the JPMorgan Chase incident comes from the bank’s statements, as reported by the New York Times and ABC News on November 16. The account of the Workday outage comes from its notice to customers and its statements, as reported by UC Today on November 14. The account of the United Airlines outage comes from the airline’s statements and a pilots’ union memo, as reported on November 16 by the New York Times, CBS News and AeroTime; United has not named a cause. The account of the Apple outage comes from Apple’s statement to Bloomberg and its system status page, as reported on November 17 by MacRumors and TechCrunch. The account of the National Security Agency outage comes from the agency’s statement and from reporting on November 17 by CNN and Reuters; what the Director of National Intelligence said about it is as reported by The Record on November 18. The account of the Amazon Web Services outage comes from its service health notices, as summarized by the Cloud Security Alliance on November 18; AWS has not published a cause. The account of the Marriott outage comes from the company’s statement, as reported by Skift on November 17. The statements made in government are as reported on November 18 by ABC News and CBS News. The account of the Optimizely attack comes from Help Net Security on November 17, reporting the research of Report URI. The FAA investigation is open, and these accounts may change. REAP was not on any of these networks and has no part in any of the investigations. We will update this post when final findings are published.

All postsTodd Yancey · Chief Commercial Officer