Forgot About 2009: Email, Flights and Banks Failed in Five Days. None Was an Attack. All Three Were Ordinary Changes With No Room and No Walls, and the FAA Was Warned in 2010
Before your next change to anything people depend on, answer five questions: can we undo it, how much spare capacity is left if it goes wrong, what stops it spreading, who is watching while it happens, and how do we talk if our usual tools are down. Between Thursday, November 12 and Monday, November 16, 2026, email, air travel and banking each failed in ways those five questions would have caught.
The week
When | What failed | What the operator says caused it | Attack? |
Thu Nov 12 | Microsoft 365: Outlook, Teams, OneDrive, SharePoint. Over 80% of gateway servers unresponsive at the worst point | A latent defect from a July gateway fix, set off by a traffic surge on a fleet with capacity reduced | No, per Microsoft |
Fri Nov 13 | FAA nationwide ground stop, about 3:30 to 8:15 p.m. Eastern; about 12,462 flights canceled | Data flooding the FTI network at 1,000 connection attempts every 4 to 5 seconds from two remote sites, one of them mid router replacement | No indication, per the FAA |
Mon Nov 16 | Chase transactions slowed; delays reached other banks through Zelle | A software problem at Chase, per Chase via the press | No, per Chase via the press |
One pattern, three times
1. A routine change. Microsoft shipped a gateway fix in July. L3Harris swapped a router at a remote FAA site at 3 p.m. on a Friday. Chase made a software change, by its own account. Nobody did anything exotic.
2. Not enough room. Microsoft had about 14 percent of the gateway fleet out of service when the surge hit; that alone puts 16 percent more load on every remaining server. The FAA's legacy equipment "became unstable" under the flood. Spare capacity is the margin that turns a bug into a slowdown instead of an outage.
3. No walls. Microsoft's separation between regions did not contain the failure. One FAA site's traffic reached 17 of 22 en route centers. One bank's delay reached other banks through a payment network that, as it happens, those same banks own. A wall is what keeps a local problem local, and all three walls leaked.
4. Recovery fought back. Microsoft's servers fell over again as they came back, swamped by waiting traffic. The FAA had to block two sites and then switch off a function on more than 60 routers before it trusted the network enough to let planes go.
The part where we were warned
The FAA had this exact failure on November 19, 2009. A Harris engineer misconfigured a temporary router on the same FTI network; when a replacement router went in at Los Angeles Center, its configuration was copied from the old one, error included. It reached 129 facilities instead of 21, blocked about 75 percent of the network's routes and delayed 820 flights. The DOT Inspector General told Congress in June 2010 that FAA "tends to have a reactive, rather than proactive approach," after recommending in 2008 better controls over the contractor's equipment configuration. Harris is now L3Harris, and it still runs the network.
In September 2026 the GAO found the FAA's modernization had 11,389 project schedules and no integrated master schedule. Two months later a router replacement at a remote site took down most of the system again.
That is what "forgot about 2009" means. The lesson was written down, delivered to Congress, and filed.
The five questions, in order
Can we undo it, and have we tried? Rollback that has never been tested is a hope. Microsoft's case adds a harder version: can you undo a change from four months ago, once it is part of the baseline?
How much room is left if this goes wrong? Count capacity with the maintenance you are doing right now subtracted, then add a surge.
What stops it spreading? Name the wall. Storm control on a router port, a cell boundary in a cloud service, a rate limit on a payment rail. If you cannot name it, there isn't one.
Who is watching, live? In 2009 the alarm meant to flag overloaded routers was itself misconfigured, which added hours. Watching means a person with a dashboard that is known to work, during the change.
How do we talk if our tools are down? On November 12 the outage took out the email and chat people would have used to coordinate the fix. Decide the backup channel before you need it: check where your alerts land.
None of this costs real money. It costs discipline, which is harder to buy.
From someone who has been on the other end of the phone
I spent the months after September 11, 2001 rebuilding infrastructure at JPMorgan Chase. What I wrote about it later still holds: "It wasn't about heroics. It was about uptime." Every outage looks like a technology failure from the outside and a process failure from the inside. The router, the bug and the payment message are just where it surfaced.
What is not known
Root causes are preliminary in all three cases. Microsoft has promised a full review; the FAA and L3Harris are still investigating; Chase's explanation so far comes through the press.
Whether the three failures share anything beyond the pattern. Every operator says they are unrelated, and we have no evidence otherwise.
Sources
FAA, "Nationwide Ground Stop: Questions and Answers," November 15, 2026. DOT Office of Inspector General, letter CC-2010-012, June 17, 2010. GAO-26-107992, September 14, 2026. ABC News and Fox Business on Microsoft's explanation, November 16, 2026. CBS News, AP via Al Jazeera, and ABC7 on the bank outage, November 16 and 17, 2026. Full citations are in each linked post.
How do AI models see YOUR brand?
AIPM has audited 250+ domains. 15 seconds. Free while still in beta.



Comments