Federal aviation officials are hoping to sharply reduce close calls on airport runways by using artificial intelligence — an effort that comes after a deadly runway crash in New York this year refocused attention on such incidents.
The Federal Aviation Administration has partnered with technology company Palantir, which has developed a new AI tool that the agency is already deploying to distill a firehose of data and predict potential problems with takeoffs and landings. That work has accelerated in the past five months with funding from last year’s GOP-led megalaw which allowed the FAA to backfill other modernization costs and devote more of its core operations budget to expanding the project, according to a high-ranking FAA official who works in the aviation safety management office.
The official said the AI tool, known as Foundry, is pulling together information that had previously been scattered across the federal government and other sources. The FAA estimates it’ll spend almost $4 million for the project from it fiscal 2027 budget; the agency did not immediately have information on what has been spent thus far.
“This data has always been there,” said the official, who was granted anonymity to candidly discuss the agency’s ongoing project. “The problem is this data was always siloed,” the person added, noting that previously, multiple divisions across FAA were analyzing their own pieces of the puzzle.
Palantir’s tool — “the central integration platform” for multiple FAA datasets — helps pinpoint airports or operations experiencing unusual safety patterns, FAA said in a statement. The agency uploads incident reports, safety data and precursor events into the platform, where AI analyzes hundreds of thousands of records to identify trends, outliers and emerging risks.
Palantir did not provide comment when contacted by POLITICO.
Former National Transportation Safety Board chair Robert Sumwalt said while there’s “a great deal of hype” around AI, he thinks in this case it has the real potential to improve aviation safety when used appropriately. “It will enable the FAA to gain greater awareness of real and potential safety risks,” he said.
“Of course, one thing to guard against is over-reliance on AI,” he added. “At least for the intermediate term, human involvement with such data analysis will be essential.”
Humans still involved
Already, the FAA has acted on the tool’s recommendations. In April, the FAA announced it was barring what are known as parallel landings at San Francisco International Airport, in which two airplanes come in side by side simultaneously. The AI tool identified potential safety issues that led to the ban, the FAA official said.
For the aviation community, the incidents known officially as runway incursions are viewed as among the most consequential safety threats in aviation because of their potential to rapidly escalate into catastrophe. And while incursions have been low thus far this year, the deadly March crash between a fire truck and an Air Canada jet at New York’s LaGuardia Airport underscored the need for improvement.
Officials, however, acknowledge that the new tool would not have helped in that crash, which killed two pilots and injured dozens. Foundry is designed to identify recurring patterns and emerging risks over time and not necessarily foresee a single, complex event involving multiple contributing factors.
“Ideally, it will be able to,” the official said. “If something [were] consistently happening at LaGuardia, where we were having an increase in vehicle incursions or something like that, the AI tool [would have] identified [it], but there were so many factors in play in that LaGuardia” incident, the official said.
|
|
Each week, FAA analysts tweak or modify the model to include new data from multiple sources. Such data could include weather pattern or turbulence reports, live aircraft tracking data, traffic alert and collision alerts, news media articles, surface radar data, NTSB reports, laser-pointing incidents or even a drone sighting, the official said.
Analysts can then sort and query the data using virtually any variable, whether examining historical records or monitoring information as it arrives in real time.
FAA Administrator Bryan Bedford recently disclosed that the agency expanded its use of the platform from Palantir following the January 2025 Washington midair disaster that killed 67 people, tapping decades of airspace data to identify safety trends and emerging risks. Humans still remain at the center of the process: Transportation Secretary Sean Duffy has emphasized that AI is intended to support — not replace — human decision-making in air traffic management. The agency is expected to announce soon a new contract for that AI initiative, dubbed the “SMART” program.
Bedford said the FAA now feeds the company’s Foundry platform real-time information on aircraft operations each day.
Where it works, where it doesn’t
In the San Francisco case, the FAA official explained the agency noticed a spike in a type of audible alert for pilots when aircraft are too close to them in the air. Once FAA looked through the tool’s analysis of the alerts, known as the Traffic Collision Avoidance System, the agency realized landing procedures at the airport were being “inappropriately applied.”
“We immediately corrected that, and now we’re looking at establishing a long-term fix there, so they can get back to their normal arrival rate,” though, the official added, the FAA will still likely slow the rate down. (That may be unwelcome news to pilots and passengers, who have taken to social media to voice frustration over mounting arrival and departure delays stemming from the change).
According to a POLITICO review of runway incursion data — one of the databases the AI analyzes — there have been only three documented runway incursions so far this year, including the LaGuardia crash. Some incidents are still under investigation or have yet to be included in the FAA’s runway incident database, the agency said in its statement. Only two incursions happened by this point last year, and by the end of 2025, there had been only four total serious incursions — but not near-catastrophic — involving at least one passenger aircraft.
That’s a stark contrast from 2023, one of the worst years on record in the last decade. That year, near-runway strikes spiked to a whopping 11 incursions — more than double the previous year — amid a post-pandemic surge in air traffic (by this time in 2023, there had been five incursions, with another three occurring in early July). The number of these runway incidents fell to two total in 2024.
Notably, since last August, every one of the most serious runway surface incidents investigated by federal officials has involved an airport vehicle in some form. Those events ranged from the fatal collision at LaGuardia to incidents involving a vehicle towing an aircraft across the airfield to a maintenance facility, according to the review of the data — underscoring the growing focus on vehicle movements as a persistent runway-safety challenge.
In the case of LaGuardia, the NTSB’s preliminary findings showed that both visual and audio systems failed to alert controllers to a developing hazard because emergency-response vehicles were not equipped with transponders. The FAA official said the situation was further complicated by controllers managing multiple events at once, creating a convergence of factors rather than a recurring pattern that the Foundry tool could have detected in advance.
“We can’t guarantee everything because there’s so many variables and unknowns,” the official said. But, in the past, “we’ve been reactive to safety events, now we’re getting to proactive — and the long-term goal is to get to predictive” to the extent the agency can, the person said.
The nation’s top accident investigator recently got a chance to see the AI tool in action. “It was pretty impressive,” NTSB Chair Jennifer Homendy told reporters this week after observing an FAA demonstration. The system was “looking for hot spots across the national airspace — and it was live,” she added.
But, like former NTSB chair Sumwalt, she cautioned that AI is only as good as the parameters that humans set and how it analyzes the data. “It depends on programming,” she said.
Case by case
The focus on early risk detection is one reason the FAA is increasingly embracing artificial intelligence to enhance safety, said Hassan Shahidi, president and CEO of the nonprofit Flight Safety Foundation, which, because of its longstanding relationships with regulators, airlines and industry, conducts its own data analysis of safety events.
“That’s different than a real-time system,” which can stop a potential catastrophe in the moment, Shahidi added.
Shahidi gave the example of FAA’s Surface Awareness Initiative, which will eventually be active at 220 airports nationwide. The technology uses Automatic Dependent Surveillance — Broadcast, or ADS-B, to improve situational awareness for controllers on airport movement. It’s one of the efforts “now accelerating just over the past year,” Shahidi said.
While those systems are designed to prevent incidents in real time, FAA officials say AI-driven analysis can also help determine where safety investments and equipment upgrades are needed most.
The FAA official said knowing some of these risks like the LaGuardia incident are out there, the Foundry tool is in the process of identifying priority airports that may need more equipment upgrade sooner. Examples of such airports include highly congested locations such as LaGuardia, Chicago’s O’Hare and Ronald Reagan Washington National Airport, they said, but didn’t specify what kinds of changes are on the horizon.
The official said the AI system is expected to continue evolving as it is exposed to more data over time. When potential issues are identified, the FAA assesses and implements any corrective actions on a case-by-case basis.
Other aviation mishaps — from engine fires and cockpit smoke to aircraft striking one another on the tarmac — remain on the periphery of safety events, but still continue to loom large in the minds of travelers and capture public attention. The official said the Foundry tool is also mining for these types of reports to understand broader trends.
Overall, the endeavor has prompted the FAA to dig deeper: Even when an operation is technically compliant, within the parameters of safety, the tech will and can identify areas where risk may be increasing. The AI will also improve to the point of accounting for human error, too, the official said. For example, the vast majority — 61 percent — of runway incursions happen because of a pilot deviation, the FAA said, citing 2025 statistics.
“We continue to build on that tool, and we continue to get more refined on how we use that tool,” FAA Deputy Administrator Chris Rocheleau told POLITICO earlier this month. “Palantir has been a great partner for us.”