Skip to content
A hooded wraith faces a neon radar scope drawn as a network of air traffic control centers; one router node is red and fractured with its links cut, planes hang frozen on the scope, and a data trace below sparks red and goes flat.

Transmission 011

One Network Update Grounded America. Should AI Rebuild the Sky?

Tech News11 min read

Signal
Channel
Tech News
Received
Length
11 min read

At 3:54 p.m. Eastern on Friday, November 13, the FAA did something it almost never does: it stopped every departing flight in the United States. Not a weather hold at a few hubs. Everything. Flights already in the air were landed, and the ones at the gate stayed there for more than four hours.

The official explanation was a “temporary telecommunications outage.” When a government says that on Friday the 13th, about an hour before every cable news channel starts saying “9/11,” a little skepticism is healthy. So I went looking for whether the story holds up. Short version: it does, mostly. And the longer version is a lot more interesting than a cyberattack would have been.

What happened

Here’s the timeline as it stands Friday night:

  • 3:54 p.m. ET: the Air Traffic Control System Command Center ordered a nationwide ground stop on departures, citing “extraordinary circumstances.” Shortly before that, Chicago O’Hare and Midway had both been ground-stopped for an equipment outage.
  • 4:16 p.m.: Transportation Secretary Sean Duffy said the stop followed a communication issue affecting all of the country’s Air Route Traffic Control Centers, the regional “Centers” (reports put the count at 22) that handle high-altitude traffic between airports.
  • Around 6:45 p.m.: CBS News, citing multiple sources, reported that an update pushed to the private communications network linking the whole air traffic control system caused it to reboot, and the reboot crashed it. CBS named L3Harris as the network’s operator.
  • Later: CNN reported the FAA told airlines a router problem, believed to have started at Chicago Midway, was behind it. Duffy said routers carrying air traffic control data became overloaded and started producing inconsistent information.
  • 8:15 p.m. ET: the ground stop was lifted.

The damage: Flightradar24 counted 4,042 flights inbound to US airports when the stop was issued. By Reuters’ count via Flightradar24, more than 4,700 flights were canceled and 10,000 delayed on Friday, with another 700 already canceled for Saturday. Europe held US-bound departures and Mexico briefly grounded flights to American airports. Reuters called it the longest nationwide grounding since September 11, 2001.

Is “telecom outage” the real story?

I went in expecting to find holes. Here’s what I found instead:

  • Everyone is telling the same story. The FAA, DOT and Homeland Security Secretary Markwayne Mullin all described a telecommunications failure. Mullin said there’s no indication of a cyberattack or deliberate interference, while carefully adding “as of now.”
  • The independent reporting points the same direction. CBS’s sources (a network update, a reboot, a crash) and CNN’s (a router problem at Midway) aren’t identical, but they’re compatible. An update that destabilizes a core router, which then floods the network with garbage, is a very believable failure.
  • The network in question is real, old, and known to be fragile. More on that below.

What I’d hold loosely: the specific trigger. CBS and CNN are both working from unnamed sources, L3Harris hadn’t commented by Friday night, and “update,” “router overload” and “radar data instability” are three different descriptions of one evening. Early root causes have a way of changing. I’ll update this post when the FAA publishes something official.

One more nuance worth catching: “telecom outage” doesn’t mean AT&T or Verizon went down. This was the FAA’s own private network, the one that carries radar feeds, flight data and controller-to-pilot radio between facilities. Cell phones were fine. The sky’s nervous system wasn’t.

The sky’s nervous system is a 40-year-old design, and we’re rebuilding it while it’s running.

The network underneath

The system that reportedly failed is almost certainly the FAA Telecommunications Infrastructure, or FTI. Harris Corp. (now L3Harris) won the contract to build and run it in 2002, and it has operated it ever since. L3Harris says FTI supports roughly 87,000 flights a day across 29,000+ services. It’s the backbone that ties towers, radars and control centers together.

And it’s been on its way out for years:

  • The replacement has been “coming” since 2019. The FAA planned a successor network, FENS, as far back as 2019. Verizon won it in 2023. Then in August 2026 the FAA handed the job to AT&T instead. Meanwhile FTI keeps running on extensions.
  • The FAA itself says FTI isn’t good enough. In the contract notice for the new network, the agency wrote that the existing infrastructure can no longer adequately support National Airspace System operations.
  • Some of it is literally copper and TDM. At an industry summit in April, L3Harris itself displayed copper wiring and time-division multiplexing gear from the legacy network. If you’ve ever pulled a T1 out of a closet, you know the vibe.
  • It’s mid-surgery. In May, L3Harris said it was past 50% on a nationwide modernization of FTI. That’s the uncomfortable detail: Mullin said Friday that introducing newer technology into the older system overwhelmed it. If that holds up, this wasn’t the old network simply dying. It was the old network choking on the upgrade.

We’ve seen this movie before

Friday was the biggest version of a story that keeps repeating:

  • January 2023: NOTAM. The last nationwide ground stop came when the Notice to Air Missions system went down overnight. The FAA later traced it to a contractor who accidentally deleted files while synchronizing databases. One human, one wrong command, every departure in the country.
  • Spring 2025: Newark. Controllers at the Philadelphia facility that handles Newark traffic lost radar and radio for up to 90 seconds on April 28, and again on May 9. Officials said the April outage happened when one of the facility’s old copper lines got fried. The radar data for Newark is processed in New York and shipped to Philadelphia over telecom lines, a single long thread. Several controllers went on trauma leave afterward.
  • September 21, 2026: Philadelphia again. A switch on the primary communications line failed, and the backup fiber turned out to have been cut by railroad construction. Primary and backup, gone at once. No commercial flights landed at Newark for most of the day. Duffy called the telecom system a 40-year-old design that needs a new architecture.
  • The big picture. A 2024 GAO report found that more than 100 of the FAA’s 138 air traffic control systems were unsustainable or nearing end of life.

FAA Administrator Bryan Bedford put it bluntly after the September outage: the FAA is swapping copper for fiber, but it’s still “the same data architecture that we’ve had for 50 years.” That’s the core problem. Faster wires don’t fix a design where one bad router or one cut fiber can take out a region, or apparently the whole country.

Enter AI

The modernization money is real. Congress approved $12.5 billion in 2025, which the FAA calls a down payment. Duffy has asked for $20 billion more, and Mullin floated about $50 billion for a full overhaul. A chunk of the plan is AI:

  • SMART (Strategic Management of Airspace, Routes and Trajectories) is an AI system built by Air Space Intelligence under an $875 million, 12-year contract awarded this summer. It analyzes airline schedules, weather, airport capacity and airspace constraints to predict traffic flows and spot conflicts before they happen. It started rolling out in September, and its debut was upstaged by, you guessed it, the Philadelphia outage.
  • It’s not flying the planes. The FAA and DOT say SMART is a planning tool for traffic managers, not a replacement for controllers, and it isn’t aimed at safety-critical separation. Humans still keep planes apart.
  • The broader plan also includes redesigning the computer systems, moving computing to the cloud, and AI-based problem prediction. Duffy says the money for those parts isn’t there yet.

So: good thing or bad thing?

The case for AI in the tower

There’s a real argument here, and it’s not just hype.

  • Prediction is the job AI is actually good at. Air traffic management today is largely reactive. Bedford has called the FAA’s traffic management tools “glorified calculators.” Forecasting demand, weather and capacity hours ahead is pattern-matching on huge amounts of data, which is exactly where machine learning earns its keep.
  • It could have helped Friday, but not where you’d think. The highest-value AI isn’t in the cockpit or the scope. It’s in the network operations center. Anomaly detection that notices a router behaving strangely after an update. Change-risk scoring before an update goes out. A digital twin of the network to test against first. That’s ordinary AIOps, the same thing big cloud providers run, and it targets exactly the failure we just had.
  • Controllers are stretched thin. The FAA is still thousands of certified controllers short. Anything that takes routine planning load off them buys back attention for the hard, human calls.
  • Recovery is a scheduling nightmare. After a four-hour stop, planes and crews are in the wrong places for days. Re-optimizing thousands of flights around those constraints is a problem computers solve better than spreadsheets and phone calls.

The case against (or at least, “not yet”)

Now the devil’s advocate, because I think this side gets waved away too easily.

  • AI doesn’t fix a cut fiber. Friday was a network problem. September was a switch plus a backhoe. Newark was copper. None of those are solved by smarter software sitting on top. If anything, AI adds one more dependency on the same fragile data pipes. A predictive model fed bad data during a meltdown makes confident bad predictions.
  • More complexity, more ways to fail. Mullin’s own explanation for Friday was that new technology overwhelmed old technology. Bolting AI onto a half-migrated network is more of exactly that. Every new system has to talk to every old one, and the seams are where things break.
  • Automation surprise is a real aviation hazard. Aviation safety research is full of cases where humans trusted an automated system right up until it did something unexpected. A tool that’s right 99% of the time trains people to stop checking. In a system where failures cluster (bad weather, outages, holiday traffic), that last 1% shows up at the worst moment.
  • The people who’ll use it weren’t asked. NATCA, the controllers’ union, said it hadn’t been involved in SMART’s design, testing or rollout. When the experts on the floor are left out, that’s a process smell, not a technicality.
  • Concentration risk. Today, one contractor’s network update reportedly stopped every departure in the country. The modernization adds more single-vendor platforms (a 12-year AI contract, a new prime integrator, a new telecom backbone). Each is reasonable on its own. Together, they’re a lot of eggs in a few baskets.

My take

Do I believe the “telecom outage” explanation? Yes. Everything credible points that way, and the theory that fits best is boring: an update hit a fragile, decades-old network mid-upgrade, and it fell over. It’s not a cover story. It’s an indictment.

On AI, I land in the middle, leaning skeptical about the order of operations. AI for forecasting and traffic flow is a good idea, and I’d love to see it in network operations too, catching bad updates before they spread. But it’s the paint, not the foundation. The foundation is boring stuff every homelabber already knows:

  • Truly independent paths. If your backup fiber runs next to the same railroad as your primary, it’s not a backup.
  • Staged rollouts. You don’t push an update to all 22 Centers at once. You push it to one, watch it, then the next.
  • Out-of-band management and fast rollback. When the network breaks, you need a separate way in to fix it.
  • No changes on Friday afternoon. I’m only half joking. Somebody pushed a network update at the start of a weekend, and it was Friday the 13th besides.

If my Proxmox cluster’s whole network can be taken down by one bad update, that’s a design flaw I’d fix before buying anything shiny. The FAA should hold itself to at least that standard. Fix the plumbing, then add the brain.

What do you think? Is AI the upgrade the system needs, or one more thing to break? And if you work in aviation or telecom, I’d genuinely like to hear where I’ve got this wrong.

Watch

Sources

Back to the archive
Scroll to Top