← Back to the dossier

12–17 November 2026 · Microsoft · FAA · JPMorgan Chase · United

Five days, four systems, no attack: how Microsoft, the FAA, Chase and United fell over

Between 12 and 16 November, four systems that millions of people use every day failed in five days. Microsoft 365 was down for almost 24 hours. The US aviation regulator stopped every departure for the first time since 2023. Chase's payment systems failed and dragged other banks with them. United could not weigh its planes for ninety minutes. Each time the company or the regulator said the same thing: not a cyberattack, not an AI agent. The shared cause was duller and harder to fix: one part everybody uses, too little spare capacity, and no way back.

By Mara Masaeva · Updated 17 November 2026

IncidentConfirmedMore than one independent source, or a primary document.

What happened

Microsoft's preliminary report puts the outage between 12 November at 16:20 UTC and 13 November at 16:10 UTC, just under 24 hours. Three things had to line up. A bug sat dormant in the cloud gateway since a deployment in July. Fifteen hours before the failure, a routine traffic shift moved more load onto that gateway. And the fleet was already about 14 percent below its planned capacity. At the peak, more than 80 percent of the gateway servers were frozen.

The design was supposed to contain that. It did not. Regional isolation failed, so the failure spread instead of staying in one region. Exchange Online could not fail over to a healthy copy. A rollback was not possible, because the change that mattered had been deployed four months earlier. Microsoft's report says in so many words that this was a technical issue, not related to cyber or AI or agents. A final report is due on 20 November.

The next day the Federal Aviation Administration stopped all departures in the United States, the first national ground stop since January 2023. A contractor, L3Harris, was replacing copper with fibre and swapping a router. The new router flooded the network: about a thousand logon attempts every four to five seconds. The routers themselves are Cisco 600-series machines built between 2001 and 2006. The instability reached 17 facilities. CBS reports 9,159 flights cancelled and about 3,500 aircraft in the air when the stop came. The White House has set up a "tiger team".

On 16 November it was the banks. Chase said a software bug in one transaction path spread to other banks and to Zelle, and that it was not a cyber issue. The same day United lost its weight-and-balance system for about ninety minutes: the software could not produce final weights, so aircraft could not leave. The pilots' union says crew communications and messaging were hit too. Fewer than 150 flights were involved. On 17 November the NSA's own classified internal network went down, and there the answer is still open: no evidence of an attack, six sources told CNN.

How it workedtechnical detail

All four are cascading failures, and cascading failures have a shape. A system is built to survive one thing going wrong: a disk, a region, a router. It is rarely built to survive three things going wrong in the same direction. Microsoft's three were a latent bug, a recent traffic shift and a fleet without headroom. Each was survivable alone. Together they removed the isolation that was supposed to stop the spread.

The second ingredient is correlation. When everyone runs on the same gateway, the same payment rail or the same router model, one mistake is not one failure; it is every customer's failure at once. That is why the Chase bug reached other banks: the path was shared, so the bug was shared. The third ingredient is the missing way back. A rollback needs a version to return to. If the bad change is four months old, or the hardware is twenty years old, there is nothing to return to.

This is also the part that makes the week interesting for AI. The systems that failed are exactly the systems AI agents are being wired into: email, ticketing, payments, flight operations. An agent does not make a correlated failure less likely. It adds a component that can issue a thousand actions a second, and it inherits every permission the integration gave it.

What it cost

Thousands of people could not fly, work or pay. CBS puts the flight cancellations at 9,159. In one reconstruction of the Microsoft outage, five people were injured at a hotel pool near Phantasialand because the chlorine alert went out by email through Outlook and did not arrive. The cost of the Microsoft outage alone is estimated at about 7 billion dollars by Anderson Economic Group; that is an estimate, not an audited figure. No report I read names anyone killed, and none establishes that any of the four failures were caused on purpose.

What came before

This is not the first month in which ordinary infrastructure failed in a way that looked like an attack. On 12 October, attackers heated the fridges of hundreds of French supermarkets to 50 degrees, and Ahold Delhaize took systems offline the same weekend: that weekend. In September, a campaign against South Korean banks showed what a real, deliberate intrusion looks like: the Korean case. The difference is not the damage. It is whether anyone chose it.

What may follow

The week produced two sentences that will be quoted for a while. Microsoft: not cyber, not AI. The NSA: no evidence of an attack, yet. Both are honest, and both are temporary. The question for the next twelve months is what happens when a failure is both: a bug that an agent found and widened. Google says that threshold was crossed on 16 November: the first fully automated attack. The defensive version is already being sold to governments at scale: OpenAI and NATO.

What I do not know

Microsoft's final report is due on 20 November. Until then the three causes are the company's own account of its own failure. I did not read the FAA's technical findings, only CBS's reporting. Chase gave a one-line explanation and no timeline. It is not established whether the FAA router was misconfigured, faulty, or both.

The 7 billion dollar figure is an outside estimate. The 9,159 cancellations cover more than one day of disruption and I have not seen the underlying count. Whether any of the four outages was a probe that happened to coincide is unknown; no source I read suggests it, and none rules it out.

“This was a technical issue, not related to cyber or AI/agents.”

Microsoft · preliminary incident report, via Ars Technica

Read next

Sources

  1. Ars Technica: Three small failures combined to take down Microsoft 365press · main source

    The reconstruction of the Microsoft outage: the July deployment, the traffic shift fifteen hours before, the fleet 14 percent below capacity, more than 80 percent of gateway servers frozen, the failed regional isolation and the impossible rollback. Also the company's line that this was not cyber or AI.

  2. CBS News: FAA ground stop cancelled flights — what we knowpress

    The L3Harris contractor, the router flood at about a thousand logon attempts every four to five seconds, the 2001–2006 Cisco routers, 17 facilities, 9,159 cancelled flights, 3,500 aircraft airborne, and the White House tiger team.

  3. CBS News: Banks online service outages at JPMorgan Chase, Bank of America, Wells Fargopress

    Chase's statement that a software bug in a shared transaction path reached other banks and Zelle, and that it was not a cyber issue.

  4. CBS News: United Airlines says it resolved a widespread system outagepress

    The ninety-minute weight-and-balance failure, the pilots' union on crew communications, and the fewer than 150 flights.

  5. NBC News: Microsoft, FAA, Chase — the tech glitches rattling everyday lifepress

    The overview that puts the four failures in one week, and the absence of any indication of an attack.

  6. Wikipedia: 2026 Microsoft outagepress

    Dates and the wider fallout, including the five injured at the Phantasialand hotel pool when the chlorine alert email did not arrive. A summary, not a primary source.

Comments