AI Just Flew a Fighter Jet Intercept on Its Own — And the Air Force Has the Results

An X-62 VISTA AI intercept, 27 times over: a Skunk Works agent read live infrared feeds and flew a fighter into position against a T-38 — no pilot touching the stick.

An artificial intelligence agent aboard a modified F-16 tracked a target jet with an infrared sensor pod and intercepted it 27 times with no pilot at the controls, the U.S. Air Force and Lockheed Martin said Tuesday.

The intercepts were flown in April over eight sorties in a test series the Air Force designated HAVE HEAT, using the X-62 VISTA — the Variable In-Flight Simulation Test Aircraft operated by the U.S. Air Force Test Pilot School at Edwards. The target was a Northrop T-38 Talon trainer. Lockheed Martin Skunk Works and the Test Pilot School announced the results Aug. 4.

Why the X-62 VISTA AI Intercept Matters

The significance is not that a jet flew itself. It is what the software was reading.

Earlier autonomy milestones, including the widely publicized 2023 DARPA Air Combat Evolution dogfighting trials, leaned on simulated target data, according to Lockheed Martin, which says HAVE HEAT moves AI testing to real-time, onboard sensor streams. HAVE HEAT fed the agent real, unscripted and imperfect sensor data — the kind of messy picture pilots actually work from in combat.

That closes what engineers call the sensor-to-action loop: raw thermal returns in, aggressive maneuvering out, at machine speed, on an aircraft that flies today.

“Our ongoing partnership with TPS is driving important progress with this latest flight test series demonstrating that our AI can effectively and reliably close the sensor-to-action loop aboard an operational combat aircraft,” said Ron Fehlen, vice president and general manager of Lockheed Martin Skunk Works.

“Our autonomous agents consumed classified infrared search and track feeds and executed combat-critical maneuvers in real time,” Fehlen said. “This achievement marks a decisive advance toward delivering AI-augmented air dominance for the United States.”

Stacy Kubicek, vice president and general manager of Lockheed Martin Sensors and Global Sustainment, said reliable sensor data is only the foundation; the tactical payoff comes when that data connects to an agent empowered to act on it.

“This project demonstrates how sensing and autonomous AI can come together as a force multiplier to make faster, more informed action in complex environments,” Kubicek said.

Inside HAVE HEAT and HAVE HOLIDAYS

The X-62 carried a Lockheed Martin Legion Pod to watch the T-38. A parallel evaluation, HAVE HOLIDAYS, tested enhanced safety rules meant to keep an autonomous vehicle from violating user-established operational limits.

The agents themselves came out of a Skunk Works capability called Supermassive, which compressed integration and ground testing into three months.

“HAVE HEAT represents a meaningful expansion of avionics capability towards integrated, AI-driven control of multiple sensors and air vehicles for mission autonomy,” said Lt. Col. Joshua Strafaccia, dean of faculty for research at the U.S. Air Force Test Pilot School. “This bridges a capability gap and positions us to test advanced autonomy faster.”

HAVE HOLIDAYS also served as risk reduction, folding a non-prime, third-party autonomous agent and advanced microchips for sensor exploitation into the X-62’s Enterprise Open Mission System Architecture computer.

Col. Maryann Karlen, commandant of the Test Pilot School, framed the pace as the point.

“X-62 is continued proof that an agile, dedicated team can rapidly advance Department of War initiatives through intentional collaboration,” Karlen said. “Accomplishing these significant milestones is a direct reflection of the adaptability and warfighter focus personified by all members of the school.”

Manufacturers note the technology is also meant to assist pilots in crewed cockpits, not only replace them. Handing high-speed sensor management to software frees a pilot’s attention for tactics and battle management.

The Sensor Doing the Looking

The Legion Pod houses the IRST21 infrared search and track system — a passive, long-range sensor that detects engine exhaust and aerodynamic heating at long range while emitting nothing.

Radar shouts. IRST listens. Because it broadcasts no electromagnetic energy, Lockheed Martin says the sensor keeps tracking in electronic attack environments, and because it hunts heat rather than radar returns, it is not defeated by radar-evading stealth shaping — making it valuable against low-observable adversaries.

The catch is range. Infrared gives precise angles but poor distance; a small hot object nearby resembles a large one far away. Correlating angle-only data into a geometric intercept course, without a human doing the math, is the computational leap the test set out to prove.

The pod measures 16 inches in diameter and 98.5 inches long and weighs under 490 pounds. The podded configuration arrived in 2015, allowing plug-and-play fitment across fighter and non-fighter platforms. Its sensor lineage traces to legacy systems that logged more than 300,000 flight hours on F-14 Tomcats and international F-15s.

The T-38 made the problem harder, not easier. The supersonic trainer, powered by two afterburning General Electric J85-GE-5 turbojets, tops 812 mph, or Mach 1.08 at sea level, according to the Air Force, and climbs above 55,000 feet, presenting a slender airframe and modest thermal signature against the background clutter of the California High Desert.

A One-of-a-Kind Airplane

The X-62 is a heavily modified F-16D Block 30 with Block 40 avionics and a variable stability flight control system that lets its computers imitate the handling of nearly any other aircraft. Formerly the NF-16D, it was redesignated X-62A in June 2021. Its open architecture allows competing algorithms to be swapped in without hardware overhauls.

Human safety observers fly aboard and can disengage the AI instantly. An AI agent first flew a tactical aircraft in December 2022, when VISTA logged more than 17 hours of autonomous flight across a series of test sorties. Lockheed Martin announced the result the following February.

What Comes Next

The work feeds directly into the Air Force’s Collaborative Combat Aircraft program, which envisions uncrewed jets flying alongside crewed sixth-generation fighters. Initial planning assumptions have run to 1,000 aircraft, against roughly $6 billion in research spending requested across fiscal 2024 through 2028. Then-Air Force Secretary Frank Kendall set a goal in 2024 of about a third of an F-35’s roughly $80 million unit price, or near $27 million; Air Force officials said in March 2026 the program was beating that target.

The doctrine behind it is affordable mass: cheaper, attritable platforms to multiply a force built around expensive Next Generation Air Dominance fighters. Proving an agent can hunt and intercept using passive sensors alone matters because datalinks to human operators are among the first things severed in a high-end fight.

The Air Force expects an initial CCA capability around 2028 to 2029. Boeing’s F-47, the crewed Next Generation Air Dominance fighter awarded in March 2025, is targeted for first flight in 2028 and initial operational capability in 2029, with fielding in the early 2030s.

The next phase of the Test Resource Management Center’s Mission Systems Upgrade, which already drove the April campaign, will add the RTX PhantomStrike active electronically scanned array radar, expanded sensor and computing capacity, and modern network data communications — pushing the agent to juggle multiple sensor feeds at once. Stakeholders convene at Edwards for the inaugural X-62 Research Symposium Aug. 11-13.