The US Air Force puts AI in the cockpit for a real dogfight

Top Gun meets algorithmic warfare. The U.S. Air Force has officially put artificial intelligence inside the cockpit for a high-stakes, real-world dogfight.

On Thursday, DARPA revealed that an AI-controlled jet went head-to-head against a human pilot in an in-air combat test conducted back in September 2023 at Edwards Air Force Base in California. It marks the first confirmed in-flight dogfight between autonomous AI and a human in military aviation.

2,000 feet at 1,200 mph

The aerial face-off pitted an experimental X-62A jet, equipped with DARPA’s autonomous flight software, against a standard F-16 flown by a human pilot.

According to DARPA, the two aircraft engaged in high-aspect nose-to-nose maneuvers, closing the distance to within 2,000 feet while moving at speeds of 1,200 mph.

Safety pilots sat inside the X-62A cockpit ready to hit the kill switch if anything went sideways. But DARPA noted that the human pilots never had to intervene at any point during the engagement.

As for the million-dollar question—who actually won the dogfight—DARPA is keeping its cards close to its chest. The agency declined to disclose the victor.

From sim to sky

The milestone is part of DARPA’s Air Combat Evolution (ACE) program, which kicked off in December 2022.

The project aims to train AI algorithms to handle complex aerial combat while adhering strictly to military flight safety boundaries. After extensive simulation runs, engineers loaded the code onto the X-62A to see if digital reflexes held up under real G-forces.

Bill Gray, chief test pilot at the Air Force’s Test Pilot School, said solving dogfighting was essential because it represents the hardest real-time problem set in aviation. What they learn here, he noted, can transfer to practically any autonomous aerial task.

DARPA has carried out 21 test flights so far and plans to push forward throughout 2024. The test aligns with broader Pentagon ambitions reported by The Wall Street Journal to deploy AI across defense systems and build out swarms of autonomous drones.

Why it matters

We think the real story here is not whether AI scored a kill shot, but that human operators felt safe enough to keep their hands entirely off the controls at 1,200 mph.

For aerospace and defense, moving autonomous algorithms out of sterile simulators and into high-stress physical cockpits has long been the holy grail. Air combat requires split-second tactical decisions under extreme sensor noise and real-world turbulence. Proving that an AI agent can execute high-G dogfighting within safe parameters lays the groundwork for the military's loyal wingman drone programs.

DID DARPA play this smart? Yes. By treating the dogfight as a benchmark challenge rather than an immediate battlefield deployment, they established a high bar for measuring machine reliability.

Still, the broader implications are unsettling. As military autonomy accelerates, the line between machine assist and lethal autonomous decision-making will blur fast. Keeping a human in the loop is easy during an experimental drill; maintaining meaningful control during real geopolitical conflict will be an entirely different beast.