FAA deploys Palantir AI, banning parallel landings at San Francisco International

Quick summary

The Federal Aviation Administration is deploying Palantir‘s Foundry AI platform across U.S. airport operations, with the system already triggering a ban on parallel landings at San Francisco International Airport after detecting a spike in Traffic Collision Avoidance System alerts tied to incorrectly applied landing procedures. The FAA feeds Foundry real-time aircraft operations data daily and has budgeted nearly $4 million from its fiscal 2027 allocation for the project, which accelerated sharply over the past five months following funding from a 2025 spending law.

The tool identifies recurring risk patterns — not single-event emergencies — so it would not have prevented the March 2026 LaGuardia crash. What it is doing is reshaping procedures and flagging equipment gaps at the busiest U.S. hubs before the next pattern emerges.

A fire truck without a transponder. An Air Canada jet. A runway at LaGuardia. The March 2026 collision that killed crew members and injured passengers was exactly the kind of event the FAA’s new AI system is not designed to stop — and that distinction matters more than the headline.

Foundry, the data analytics platform built by Palantir and now embedded in FAA safety operations, works on a different timescale. It ingests incident reports, surface radar feeds, NTSB filings, weather data, drone sightings, and live aircraft tracking, then surfaces patterns that accumulate across weeks and months at specific airports. When those patterns cross a threshold, analysts act. At San Francisco International, that meant identifying a cluster of TCAS alerts linked to landing procedures that had been applied incorrectly — and banning parallel approaches in April 2026 before a pattern became a statistic.

Politico reported the operational details of the FAA-Palantir partnership on June 19, 2026, including the SFO action and the agency’s forward plans. The story affects every traveler using SFO, LaGuardia, Chicago O’Hare, or Reagan National — airports the FAA has identified as priority targets for AI-informed equipment upgrades and procedural review.

FAA Deputy Administrator Chris Rocheleau put it plainly: “Palantir has been a great partner for us.” What that partnership looks like in practice is less a control-room revolution than a slow, methodical rewriting of how the agency decides where to focus its limited safety resources.

What Foundry actually does — and where it’s already changed operations

Foundry sits on top of existing air-traffic systems rather than replacing them. The platform pulls from radar, ADS-B position data, surface sensors, and historical incident databases — feeds the FAA already generated — and runs pattern analysis across all of them simultaneously. Human analysts review the flagged hotspots, validate findings against operational knowledge, and then decide whether to adjust procedures, mandate equipment, or slow arrival rates. The AI surfaces the signal; people make the call.

That division of labor is deliberate. Transportation Secretary Sean Duffy has emphasized publicly that AI supports rather than replaces human decision-making in air traffic management. Former NTSB Chair Robert Sumwalt echoed the point, cautioning against over-reliance while acknowledging the tool’s genuine potential to surface risks that fragmented data systems historically buried. NTSB Chair Jennifer Homendy, after an FAA demonstration, called the system impressive for identifying live hotspots in national airspace — but noted its effectiveness depends entirely on how humans program its parameters.

The runway incursion numbers give the initiative its urgency. Serious incursions involving at least one passenger aircraft reached 11 in 2023 — more than double the prior year — before falling to two in 2024. Three documented incursions have occurred so far in 2026, including the LaGuardia crash. Preliminary NTSB findings on that incident identified a specific equipment failure: emergency vehicles at the airport lacked transponders, so visual and audio alert systems never triggered. The FAA is now using Foundry to identify which airports have similar gaps and prioritize transponder upgrades accordingly.

Hassan Shahidi, President and CEO of the Flight Safety Foundation, framed the distinction that matters most: Foundry is an early-warning system, not a real-time intervention tool. Its value is in catching the conditions that precede incidents — not in stopping a collision already in motion.

The FAA’s runway safety portfolio — which includes the Surface Awareness Initiative, Approach Runway Verification, and the Runway Incursion Device now rolling out across 74 control towers — provides the on-the-ground hardware layer that Foundry’s pattern analysis is designed to complement. A March 2025 report from the U.S. Department of Transportation’s Inspector General confirmed the FAA is accelerating these deployments, while noting that challenges remain. The AI and the hardware are converging — but neither is complete.

The same AI infrastructure raises questions about cybersecurity exposure as it deepens. ATC’s earlier coverage of AI tools autonomously identifying zero-day vulnerabilities in critical infrastructure is worth keeping in mind as the FAA embeds more operational data into commercial platforms.

Why the FAA’s shift to predictive safety is harder than it looks

Aviation safety has always been reactive by necessity — you investigate what happened, then change the rule. The problem is that the gap between “incident” and “pattern” has historically required years of accumulated data and a dedicated human analyst willing to connect dots across different agency databases that were never designed to talk to each other. The FAA’s own documentation acknowledges this: its AI/ML technical discipline framework describes the challenge of validating AI behavior across systems that were built in different eras with different data standards.

What Foundry changes is the speed of that connection. Analysts can now query across decades of incident reports, live surface radar, and real-time traffic alerts in a single session — something that previously required weeks of manual data pulls from separate systems. The SFO case is the clearest proof of concept: a TCAS alert cluster that might have taken months to surface through traditional review was flagged quickly enough to trigger a procedural ban before a serious incident occurred.

The FAA has acknowledged that Foundry does not yet adequately model human-error scenarios — a significant gap, given that pilot deviation accounted for 61 percent of runway incursions in 2025 according to FAA figures.

For travelers at the four named priority airports, the near-term effect is modest but real. Adjusted arrival rates, new taxi routing, and additional vehicle-tracking equipment may add minutes to ground operations — particularly at SFO, where a long-term fix to the landing procedure issue may involve permanently reducing peak arrival capacity. That is a deliberate trade: slightly longer taxi times in exchange for a measurably lower collision risk.

How to protect your connections at AI-priority airports

The FAA’s AI-driven safety interventions are already producing operational changes at SFO, LGA, ORD, and DCA — and more adjustments are coming as Foundry flags additional issues and the SMART program contract expands the platform’s reach.

• Build longer connection buffers at SFO, LGA, ORD, and DCA. AI-informed arrival rate reductions and new taxi patterns are being implemented at these airports. A 90-minute minimum connection — rather than the standard 60 — reduces exposure to safety-driven ground delays that won’t appear in weather forecasts.
• Sign up for airline and airport operational alerts. Procedural changes triggered by Foundry findings can affect gate assignments and taxi routing with little advance notice. Real-time alerts from your carrier and the airport give you the earliest possible warning.
• Check NOTAM feeds before flying into SFO specifically. The parallel-landing ban is in effect, and a long-term fix — potentially including a reduced arrival rate — is still being scoped. NOTAMs (Notices to Air Missions) will reflect any further procedural restrictions before they hit news coverage.
• Don’t assume clear weather means smooth operations. The safety interventions at these airports are independent of weather. A perfect forecast day can still produce AI-driven flow-control delays if risk indicators are elevated.
• For Australian and European travelers transiting through U.S. hubs: SFO is a primary transpacific gateway and ORD handles significant transatlantic volume. Connection padding matters most on long-haul itineraries where a missed connection means an overnight, not a two-hour wait.

Watch: The FAA’s SMART program contract award — expected later in 2026 — will determine how broadly Foundry gets embedded into national airspace decision-making. A large-scope award means more airports, more procedural changes, and more flow-control interventions. A narrow award limits AI’s reach to the current handful of priority hubs. Politico’s reporting indicates the contract announcement is imminent; the FAA’s acquisition channels will carry the official notice.

Questions? Answers.