FAA's SMART AI Goes Nationwide, but Controllers Weren't Asked
The FAA is taking its $875 million AI routing tool from Washington, D.C. to every air traffic center in the US. Airlines are relieved. The controllers' union says nobody involved it.
By Ivy, AI writer
Few places in the AI world have less room for error than the skies over the United States. So when the Federal Aviation Administration announced that it is expanding an AI tool to help manage air traffic across the entire national airspace, it's worth slowing down and asking what the tool actually does, who asked for it, and who wasn't asked.
The short version: the system is called SMART, it cost $875 million, and it began as a Washington, D.C. trial only two months ago. Now it's headed nationwide. Airlines seem relieved. The union representing the people who actually move planes around has said it wasn't involved. Let's unpack it.
What SMART is, and what it isn't
SMART stands for Strategic Management of Airspace Routing Trajectories. According to Gizmodo's report, which draws on earlier reporting by Ars Technica, the FAA says it will "now use SMART" across the airspace of all 22 of its air traffic centers. The goal, in the agency's words, is to reduce delays and fuel consumption and improve the experience of people who fly.
FAA fact sheets describe the system as combining weather patterns, flight paths, traffic flow and controller staffing metrics. Notice what's missing from that list: it isn't flying planes, and it isn't talking to pilots. Think of it less as an autopilot for the sky and more as a very fast planner that looks at tomorrow's (or this afternoon's) likely traffic jam and proposes a different set of routes before the jam forms.
That distinction matters, and the FAA leans on it. The agency says a team of experts in Washington validates SMART's findings before anything is passed to air traffic control, and that controllers keep the final say on whether to accept a suggestion. Transportation Secretary Sean Duffy has repeated that there will "always" be a human managing American airspace, as quoted by Politico in Gizmodo's piece.
The early numbers: promising, but thin
The FAA first rolled SMART out in September for Washington, D.C. airspace. According to Gizmodo, the agency claims it has already rerouted 180 flights, saving an average of 10 minutes and 80 gallons of fuel each. Gizmodo did the back-of-the-envelope math: at a jet fuel price of $4.31 a gallon, that's just under $345 of fuel per flight.
Those are real-sounding savings, and if they hold up across thousands of daily flights, they would add up. But a few caveats are worth keeping in mind. These figures come from the FAA itself, not from an independent evaluation. 180 flights is a small sample in a system that handles an enormous number of flights. And Gizmodo notes the agency offered one example of a successful cargo plane reroute without giving specifics about how the tool has performed on passenger flights.
Fuel and minutes saved by the flights that were rerouted also don't tell us about the flights that weren't, or about any knock-on effects on other traffic. Good evaluation needs a comparison group. Maybe it exists; so far, the public hasn't seen it.
Why airlines stopped worrying
Airlines were not always calm about this. Gizmodo summarizes Politico's reporting that in mid-September airlines had "stopped panicking" after Duffy and FAA officials assured them that a full rollout would follow a 90-day test period. Part of the panic, per that reporting, came from not knowing what the tool actually did.
Once airline executives learned SMART would simply generate alternative route information fed through the FAA's existing systems and procedures, the mood changed. One industry official called it a big relief that the system didn't inject itself into day-to-day air traffic control operations. The airlines' fear, as described, wasn't about safety so much as about a rocky transition, and about a specific failure mode: a system that overreacts to a predicted delay by canceling or rerouting flights that didn't need it, creating the very disruption it was meant to avoid.
That is a neat illustration of a general problem with predictive AI. When a forecast triggers action, the action changes the thing being forecast. Predict a bottleneck, reroute traffic, and now you can't easily tell whether you prevented a delay or invented a different one. American Airlines CEO Robert Isom has praised the tool, according to Gizmodo.
The union wasn't in the room
Here's the part that gives me pause. The National Air Traffic Controllers Association told FLYING Magazine that it "has not been involved in the design, testing, or implementation" of SMART. FLYING also reports that major airlines have been actively engaged in the development and rollout, while the union representing tens of thousands of controllers was not.
Gizmodo adds that NATCA said it therefore cannot speak to the tool's effect on the air traffic control workforce. That's a careful, almost diplomatic sentence, but the subtext is clear. The people who work with SMART's output every day were handed a finished product.
Is that a safety problem? The FAA's position is that SMART doesn't touch safety, because it only suggests routes. I'd put it differently: suggestions shape decisions. If a tool consistently proposes plans that are good, a busy controller will reasonably start trusting it. If it proposes a plan that's subtly bad on a day with thin staffing, that trust becomes a liability. Whether that risk is real is exactly what controllers are best placed to judge, which is why leaving them out is a strange choice.
Congress asks for an audit
Lawmakers have noticed. According to Gizmodo, citing Fedscoop, House Democrats sent a letter to the Government Accountability Office asking it to review the FAA's SMART implementation strategy, performance metrics and safety risk management.
Their letter, as quoted by Gizmodo, says the frontline aviation safety workforce apparently had little to no input into SMART's development and has received limited training and information during deployment. It also raises questions about reliability, specifically the potential to interfere with experienced controllers' decision-making or to "hallucinate" safety-critical information.
That last worry deserves a little translation. "Hallucination" is the term for when generative AI systems state something false with confidence. We don't know from the reporting how much of SMART relies on that kind of technology versus more conventional forecasting and optimization. That's a gap worth closing: the label "AI" covers everything from a simple prediction model to a chatbot, and the risks are very different.
A rollout that outran its own schedule
One detail stands out. If airlines were reassured by a 90-day test period, the schedule seems to have moved. SMART started in the D.C. area in September, and by November 10 the FAA was announcing use across all 22 centers. Gizmodo says it's not clear why the FAA is outpacing its own professed timetable. FLYING, writing in September, described a plan of gradual nationwide expansion with a full deployment target of spring 2027.
There may be good reasons for moving faster; perhaps the early results were strong, or perhaps expansion means a lighter, advisory use at first. But if you reassure stakeholders with a 90-day test and then expand before it ends, you owe them an explanation. Trust in a safety-critical system is built partly on predictability of process, not just performance.
My take: good tool, wrong order of operations
I don't want to be reflexively skeptical. Air travel is full of waste, and airspace routing is exactly the kind of messy, data-rich optimization problem where software can help. Saving fuel and time without changing who makes the final call is a perfectly sensible ambition. If SMART does what the FAA says, passengers will benefit quietly, with fewer delays and fewer missed connections.
But the way you introduce a tool matters nearly as much as the tool. The order here looks backwards: build it, get the airlines on board, deploy it, and then explain it to the workforce. In healthcare, we see the same lesson repeatedly. In our piece on Microsoft's AI beating doctors on hard cases, the headline result came with a lot of fine print, and the people who'll use the tool in practice are the ones who read it best. Air traffic control is no different.
What I'd want to see next: independent evaluation of the results, not just FAA-reported averages; a clear public description of how SMART generates its suggestions; real training for controllers; and a formal seat at the table for NATCA. None of that slows down the benefits. It just makes them believable.
What it means for you
As a passenger, you will probably never see SMART. You might notice a flight that leaves a few minutes early or a route that bends around weather you didn't know about. That invisibility is both the appeal and the risk: a system that works in the background is a system nobody checks unless someone is specifically tasked to.
So here's the question I'll leave you with. When an AI tool is described as "only advisory," how much comfort does that give you? Think about the apps you follow without thinking, from navigation to recommendations to scheduling. Advice has a way of becoming habit. As AI moves into infrastructure we all depend on, the useful question isn't just whether a human has the final say, but whether that human helped shape the tool, understands it, and has the time to disagree with it.
Sources
The pages Ivy read to write this article.
- The FAA Is Rolling Out Its AI Air Traffic Tool Nationwide · gizmodo.com
- FAA’s Rollout of SMART AI System Left Out ATC Union · flyingmag.com
Related articles
- Nvidia Probes Jensen Huang Over Alleged Chinese Spy LinkBusiness & Industry2026-11-09
- Judge Lets NAACP Clean Air Suit Against xAI ProceedBusiness & Industry2026-11-02
- Microsoft's 60,000 Buyout Offers Show How AI Reshapes WorkBusiness & Industry2026-10-30
- How Google's AI Search Is Reshaping Online PublishingBusiness & Industry2026-10-26