Five Days in November

Computed diagram: nine rows of dark dots scattered at random along thin lines; in the middle row four dots fall close together and are circled in magenta

In five days of November, Microsoft’s email, America’s flights, part of Amazon’s cloud and the largest US bank failed; on the sixth, Apple. Was it a pattern, or does it only look like one?

On Thursday 12 November 2026 Microsoft 365 stopped working around the world, and email did not fully return until the next day.[1] On Friday the United States halted every departing flight for about four hours; more than 9,000 flights were cancelled that day.[2] On Sunday Amazon’s cloud in northern Virginia had sign-in and connection failures in eleven services for about three hours.[3] On Monday a software bug at JPMorgan Chase, the largest American bank, delayed payments for its customers and for people at other banks dealing with them.[4] On Tuesday Apple’s iCloud sign-in, App Store, Apple Pay and support services failed for several hours; Apple blamed “an internal update” and said it was not a cyberattack.[5][6] It was the day a rumour said a 90-day nationwide blackout would begin, and the rumour had spread far enough for fact-checkers to answer it.[7]

Senator Josh Hawley said that if it was “all a coincidence, it’s like the greatest coincidence in the history of the world”.[8] Nothing to Spare, written a few days earlier, argued that large failures are not shown to be more frequent, only wider when they come. This essay asks whether these days change that: whether the failures were connected, whether outages are rising, or whether it only looks that way.

5large outages in six days, 12–17 November, each with its own cause
419network outage events worldwide that week, against a 2026 median of 387
34%chance that a year of random outages, at one a week, holds such a run somewhere
0attacks found behind any of them, so far
The count

Not more outages; bigger ones

The firm ThousandEyes, which watches networks from many points around the world, publishes a weekly count of “network outage events” across internet providers, cloud networks, collaboration apps and the services at the edge of the internet.[9] It is not a count of headlines: most of these events are small, and nobody outside a network operations room hears of them. That makes it a useful check on whether the week was unusual underneath.

Network outage events per week, worldwide, January 2025 to mid-November 2026 ThousandEyes via Network World. Weekly counts range from 74 to 687. Week of 9 November 2026: 419; median of 2026 weeks: 387. Week of 10 November 2025: 128. NETWORK OUTAGE EVENTS PER WEEK, WORLDWIDE 0 250 500 750 Jan 2025 Jul Jan 2026 Jul 2026 median: 387 9–15 Nov 2026: 419 10–16 Nov 2025: 128
Figure 1. Network outage events per week, worldwide, January 2025 to the week of 9 November 2026. The week of the Microsoft outage and the ground stop is marked; a gap in December 2025 is a week missing in the source. Source: ThousandEyes, as published weekly by Network World.[9][10]

It was not. The week of 9 to 15 November had 419 events worldwide and 282 in the United States, close to the medians for 2026 (387 and 234), and below the busy summer, when weekly counts passed 600.[9] Chase’s and Apple’s outages fell in the following week, whose count is not yet published. By this count 2026 as a whole runs higher than 2025, about 400 events a week on average against about 300, but the series swings so much from week to week, and is so little explained, that I cannot tell how much of that is more outages and how much is more watching.[10]

The cloud providers’ own failures tell a similar story. Control Plane, a cloud software firm, has collected the major outages that Amazon, Microsoft, Google and Cloudflare documented publicly since 2020: 16 in 2025 and 18 in 2026 up to late October, about 0.4 a week, more than in earlier years, though the authors warn that their yearly counts are “a floor, not a precise trend”.[11] What made the week in November stand out was not the number of failures, but that several of the few large ones fell together, and that they hit things everyone uses.

Chance

How often luck does this

Random events do not arrive evenly spaced. They bunch, and the bunches look meaningful. A simple way to test the week is to pretend that large outages arrive at random, at a steady average rate, and ask how often a year would contain four of them within five days, or five within six.

The answer depends almost entirely on the rate, which nobody measures. The four cloud providers alone manage about 0.4 a week; add the rest, from other software services to banks, airlines, air traffic control and telephone networks, and one a week is a plausible guess, though only a guess. At that rate a year of purely random outages holds a run of four in five days 73% of the time, and a run of five in six days 34% of the time; at 0.7 a week, those chances fall to a third and to under one in ten (Figure 2). Adding Apple makes the run rarer: no longer something most years contain, but still something chance produces every few years.

Chance that a year of random outages contains at least one run of three or four in five days, or five in six Simulated Poisson arrivals, 20,000 years per rate. 0.4 a week: 3 in 5 days 44%, 4 in 5 days 6%, 5 in 6 days 1%; 0.7 a week: 3 in 5 days 91%, 4 in 5 days 34%, 5 in 6 days 9%; 1.0 a week: 3 in 5 days 100%, 4 in 5 days 73%, 5 in 6 days 34%; 1.5 a week: 3 in 5 days 100%, 4 in 5 days 99%, 5 in 6 days 85%. CHANCE A YEAR HOLDS A RUN OF OUTAGES, BY AVERAGE RATE 3 in 5 days 4 in 5 days 5 in 6 days 0% 25% 50% 75% 100% 44 6 1 0.4 a week 91 34 9 0.7 a week 100 73 34 1.0 a week 100 99 85 1.5 a week
Figure 2. Chance that a year of outages arriving at random contains at least one run of three or four outages within five days, or five within six, for four average rates. 20,000 simulated years per rate. The rates are assumptions, not measurements. Computed by scripts/outage_cluster.py.

There is a second effect, and it is larger. Asked in advance about one particular five days, the same model gives a run of four less than one chance in a hundred, and about one particular six days, a run of five about one chance in five hundred. Asked afterwards about the days that happened to be bad, it gives the much larger chances above. The week was picked because it was striking; so was the list of five, which leaves out the failures that did not make the news and could have added others that nearly did. Nassim Taleb, reviewed on this blog in The Black Swan, spends a book on how readily we find a story in a run of chance. None of this proves the week was luck. It shows that luck is enough to produce weeks like it, every year or two for a run of four and every few years for a run of five.

One of the five is genuinely rare. The ground stop was only the second nationwide halt of US air traffic since September 2001; the first, in 2023, came when contract workers deleted critical files from a pilot-alert system.[2] A rare event can still fall into a busy week by chance, but it is the one failure here that would be news in any week.

Attention

Seeing a pattern

Attention does the rest. When email fails for millions of people at once, the next failure is reported faster, read more and joined to the first. Emily Harding of the Center for Strategic and International Studies put both sides in one breath: there “is absolutely a tendency to see a pattern where one may not exist”, but “I don’t really believe in coincidences. And this is a lot of things piled up at the same time.”[8]

The rumour of a blackout from 17 November shows how quickly the gap fills, and Apple’s failure on that very day could only feed it. It spread on social media with a manipulated video of President Trump, and grew from a real nationwide test of the Emergency Alert System on radio and television that day and a Pentagon project to let military bases run on their own power for 90 days. PolitiFact found “no evidence that is going to happen”.[7] The Homeland Security Secretary, trying to reassure people during the ground stop, said it was not another 9/11, which, as one analysis noted, left any number of frightening possibilities open.[8] No public evidence has emerged that any of the outages was an attack: Microsoft, the Transportation Department, Chase and Apple all said they were not, and Amazon has not suggested otherwise.[1][2][4][5]

On 18 November the Director of National Intelligence, Jay Clayton, said the same to the House Homeland Security Committee: “These are separate service outages associated with maintenance and technical changes. They are not a coordinated or state-sponsored attack.” The Homeland Security Secretary said he was “98% confident” that no foreign adversary was behind them, while admitting he had “absolutely not” seen so many outages in so short a time.[12] By then the list had grown again: a classified internal network at the National Security Agency had been degraded on Monday, which the agency put down to “scheduled routine maintenance”.[13] It is left out of the count here, because it is not a public service, but it points the same way.

The common thread

What they had in common

There is a pattern, but it is not a plot. Of the causes given so far, three are changes made on purpose and the fourth a software bug. Microsoft’s outage began with a fix deployed in July, met a routine switch of traffic and nearly 14% of capacity out of service after an unrelated problem, and then spread through the system built to route around failures.[1] The ground stop followed routine work by a contractor, L3Harris, replacing old copper telecom lines with fibre at a remote site; a router installed there began flooding other routers in the air traffic control system.[2] Chase’s was “a software bug in an on-premise hardware system”.[4] Apple said only that “an internal update caused some outages”, without saying what the update was.[5] Amazon has not given a cause.[3]

That is the ordinary pattern of large outages, not a new one. Asked about their major network outages of the past three years, 41% of data-centre operators surveyed by the Uptime Institute named a failure of configuration or change management, the most common answer; 9% named a cyberattack (Figure 3). For outages of systems and software the leading causes were software faults (53%) and, again, configuration and change (51%).[14]

Most common causes of major network outages, data-centre operators, past three years Uptime Institute survey 2026, up to three answers each (n=122): Configuration or change management failure 41%; Third-party network provider failure 34%; Hardware failure 34%; Firmware or software error 29%; Line breakages 25%; Network overload or congestion 14%; Malicious cyberattack 9%. SHARE OF OPERATORS NAMING EACH CAUSE, % Configuration or change management failure 41% Third-party network provider failure 34% Hardware failure 34% Firmware or software error 29% Line breakages 25% Network overload or congestion 14% Malicious cyberattack 9%
Figure 3. The most common causes of major network outages at data-centre operators in the past three years; each respondent could give up to three, so the shares add to more than 100%. Source: Uptime Institute, Data Center Resiliency Survey 2026 (122 respondents).[14]

Uptime adds a warning that fits Microsoft’s case exactly: automation improves recovery, but also increases “the risk of configuration errors and cascading failures”, so reliability now depends on “limiting how far and how quickly those failures can propagate”.[14] That is the argument of Nothing to Spare in an engineer’s words.

The pace of change

More change than ever

If change is what usually breaks things, the question becomes whether there is more change than before, and here something has moved. In November Microsoft fixed 1,093 security flaws in one monthly release, after more than 1,400 in October; it has shipped more than 5,100 security fixes in 2026, against about 100 a month a year earlier, and credits artificial intelligence for finding so many bugs.[15] Others are patching heavily too: Google’s Chrome release of 3 November alone fixed 213 flaws.[15] Every fix is a change that someone must test and deploy, and every deployment is a chance to break something. The Load-Bearing Volunteer describes the same pressure falling on the few people who maintain the software everyone uses.

On the evening of Apple’s outage this idea was reported to have reached Congress, in a stronger form. The Washington Sun, citing three unnamed people, reported that bank executives and the NSA had told members of Congress, in private, that the week’s outages came from rushing out fixes for flaws found by AI models. A congressional official it quoted put it this way: “They’re trying to push through a lot of patches all at once — more than the system was designed for.”[13] A spokesperson for Senator Mark Warner told Fortune that botched patches were linked to the NSA’s outage and to the banking outages.[16] No bank and no agency has said so in public, and the account names only the banks and the NSA: not Microsoft, the FAA, Amazon or Apple.[13]

Old systems are being replaced too. L3Harris says it has completed more than 70% of the FAA’s conversion from copper to fibre.[2] Replacing old equipment is the right thing to do, and the moments of switching over are when it is most likely to fail.

What this does not show is that the faster pace caused any of the five outages. Microsoft’s defect came from a change made in July, not from November’s fixes; the ground stop came from fibre work; Chase’s bug sat in its own hardware; Apple has not said what its update changed.[1][2][4][5] For Chase, rushed patching is now a reported cause, though only second-hand; for the rest, the pace of change is a plausible reason to expect more failures of this kind, and the thing to watch. It is not yet a finding.

Verdict

A run of luck, and a trend underneath

So was it a pattern? Mostly not, in the sense people feared, and partly yes, in a duller sense.

The count
Ordinary. The week’s 419 network outage events were near the 2026 median. Outages in 2026 run above 2025 by this count, but the series is too noisy to say why.
The cluster
Within what chance produces, if at the edge. At a plausible rate, four large outages in five days happen somewhere in most years and five in six days about one year in three; and the days were chosen after the fact.
The fear
Not supported. No evidence of an attack or of a link between the outages; the intelligence chief told Congress they were separate. A rumour filled the gap.
The thread
Real. The causes given were changes made on purpose and a software bug, as in most large outages, and the volume of change is rising fast. In private briefings, reported second-hand, the banks and the NSA blamed rushed fixes for flaws found by AI. That makes failures like these more likely; it has not been shown to have caused most of them.

Four things would settle more of it: Microsoft’s final report on its outage; an account from Apple of what its update was; the result of the FAA’s investigation into the router; and the weekly counts for the rest of November. If the counts stay where they are, the week was a run of luck in a year of heavy change. If they climb, it was the start of something.

On method and tools

This essay was written collaboratively with Claude Opus 5.5 (Anthropic) on 18 November 2026: Luis Matos Ferreira’s question and editorial direction; the machine’s research, analysis and drafting. Some sources were read second-hand: Microsoft’s preliminary report as published by the French site Next; Amazon’s status updates through GBHackers’ account of them; the Transportation Secretary’s and L3Harris’s statements through NPR; Chase’s statement through the Associated Press; Apple’s statement through 9to5Mac and Time; the congressional hearing of 18 November through CBS News; the private briefings to Congress through The Washington Sun, whose sources are unnamed, and Fortune; the rumour through PolitiFact’s fact-check as published by PBS; and the ThousandEyes counts through Network World, which does not describe how they are made. Claude and Anthropic were among the services hit by the Microsoft outage; that is not discussed here. The weekly counts are typed into docs/outage-cluster-data/; the figures, the ranks and the chance model are computed by scripts/outage_cluster.py (results in docs/outage-cluster-results.json). The chance model assumes outages arrive at random at a steady rate; the rates shown are assumptions, not measurements. The cover is computed by scripts/five_days_cover.py and is a diagram, not data.

Authored by: Luis Matos Ferreira — Physicist, Developer, Writer

Related essays on this blog
  1. Nothing to Spare — whether our systems have become too fragile, or we have forgotten how fragile they were.
  2. Rented Ground — Europe’s firms on three American clouds: failures that are rarer, but come all at once.
  3. The Load-Bearing Volunteer — the software the world runs on, and the few people who maintain it.
  4. Book Review: The Black Swan — Taleb on rare events with large effects.
Sources
  1. Microsoft, Preliminary Post Incident Report for Microsoft 365, 15 November 2026, as published by Next (incident from 16:20 UTC on 12 November to 16:10 UTC on 13 November).
  2. NPR, “FAA ground stop ends but passengers stuck at airport”, 14 November 2026, citing FlightAware, the US Transportation Secretary, the FAA Administrator and L3Harris.
  3. GBHackers, “AWS Confirms Major US-EAST-1 Outage Disrupted 11 Cloud Services Before Full Recovery”, 16 November 2026, from the updates on the AWS Health Dashboard.
  4. C. Bussewitz and K. Sweet, Associated Press, “Technical issue triggers online banking delays across major US banks”, 16 November 2026, as published by WBTV.
  5. Z. Hall, 9to5Mac, “iCloud, App Store, and other Apple services hit by ongoing outage”, 17 November 2026, updated with Apple’s statement.
  6. C. Greene, Time, “What to Know About the Apple Outage”, 17 November 2026.
  7. M. Briceño, PolitiFact, “Debunking social media posts that claim a 90-day blackout will start Nov. 17”, PBS News, 17 November 2026.
  8. CNN, “Analysis: What should really scare Americans amid blitz of tech outages”, 18 November 2026.
  9. D. Dubie, “2026 network outage report and internet health check”, Network World, weekly ThousandEyes data, read 18 November 2026.
  10. D. Dubie, “2025 global network outage report and internet health check”, Network World, weekly ThousandEyes data.
  11. Control Plane, “The 2026 Cloud Outage Report: What 62 Major Outages Reveal”, 26 October 2026.
  12. N. Sganga, CBS News, “‘No indication’ recent outages are related to coordinated cyberattacks, officials say”, 18 November 2026.
  13. J. Stein, The Washington Sun, “Bank CEOs, NSA Tell Congress That AI-Driven Patches Are Behind Outages”, 17 November 2026; based on three unnamed sources.
  14. Uptime Institute, Annual outage analysis 2026, May 2026, figures 11 and 12 and the text on networking outages.
  15. B. Krebs, “Microsoft Fixes Nearly 1,100 Security Vulnerabilities”, Krebs on Security, 10 November 2026.
  16. E. Forlini, Fortune, “‘A failure of implementation’: AI played a key role in some outages, but not in the way you might think”, 18 November 2026.

Comentários

Mensagens populares deste blogue

Le Grand Raid des Pyrénées

ITRA Performance Index - Everything You Always Wanted to Know But Were Afraid to Ask

Pequenos conselhos práticos - Solo Duro

How Trust Becomes Access

Provas Insanas - Westfield Sydney to Melbourne Ultramarathon 1983

UTAX - Ultra Trail Aldeias do Xisto - 2014