The FAA's $875m AI air-traffic tool starts small
An $875 million AI system that predicts flight congestion could start advising controllers at Washington's three main airports as early as next week.
The Federal Aviation Administration is about to put an artificial intelligence system in front of working air traffic controllers for the first time, beginning with one of the most congested patches of sky in the United States. The tool, called SMART, could start advising controllers on flights around Washington, DC, as soon as Monday 21 September.
What the system does
SMART uses AI models to predict how air traffic will flow and to flag potential conflicts, drawing on airline schedules, weather, airport capacity and airspace conditions. Government and industry officials told The Wall Street Journal that the first deployment would cover the three major airports in the Washington area before a nationwide rollout across the 29 million square miles of airspace the agency manages, as Ars Technica reports. The FAA has told airlines that the system will not change any procedures for controllers or carriers; it produces alternative route information that is surfaced through systems the agency already runs. That reassurance arrived only after weeks of confusion in the industry, with airline officials telling Politico they had struggled to get a clear picture of what was being switched on.
The contract behind it
SMART sits inside an $875 million, 12-year contract awarded in June to Boston-based Air Space Intelligence, whose Flyways platform already helps manage more than 40 per cent of US air traffic through customers including Alaska Airlines. The same contract covers a Flow Management Data and Services system that will replace the tooling used at the FAA's Air Traffic Control System Command Center in Virginia. Philip Mann, an aviation safety and AI governance consultant who spent 17 years at the FAA, describes the flow management system as the backbone and SMART as the predictive layer above it. The agency's stated goals are lower fuel burn, better on-time performance and faster recovery when weather or congestion snarls the network. What kind of AI sits at the centre is not public: deterministic models that follow fixed rules, statistical machine learning and generative systems all carry very different risks in a tool that controllers are expected to trust.
Why the stakes are high
The FAA is short of controllers and has been for decades. The workforce declined by about 6 per cent over the past decade, according to a Government Accountability Office report published in December 2025, and in May the agency cut the number of controllers it expects to need for 2026 to 2028 by roughly 2,000. A 43-day government shutdown spanning October and November 2025 left controllers working without pay. Understaffing was one of the causal factors investigators cited after a US Army helicopter and an American Airlines jet collided near Ronald Reagan Washington National Airport on 29 January 2025, killing 67 people, with a single controller handling helicopter and runway traffic that night. Mann notes that the SMART rollout may still be limited to aircraft at 24,000 feet and above, as the FAA set out in June, which would mean cruise traffic plus the climbs and descents feeding the airports below.
Our opinion
Slow and unfashionable is the right way to introduce a system like this. Putting AI into air traffic control under demonstration conditions tells you almost nothing about how it behaves on a bad weather day, which is exactly why a Washington debut at this scale is worth watching rather than celebrating. The most useful thing the FAA has said so far is also the least exciting: SMART will not change controller procedures and will only offer alternative routes through systems that already exist. That keeps a human in the loop and keeps the safety case intact, which matters when the alternative is a national-scale system whose inner workings are not public. The harder question is what happens next, because scope in aviation programmes has a habit of growing quietly while everyone is watching the launch. The tests that should gate any expansion are performance under real workload and in degraded data conditions, with the controller shortage as the permanent backdrop rather than the excuse for rushing. If a system that cannot explain itself starts shaping which aircraft gets the shortcut, the person who signs off on that decision still has a name and a licence.