AI Beat a Fighter Pilot 5–0. Then They Put It in a Real F-16.

First it beat humans in simulations. Then they gave it a real fighter jet.

In August 2020, an experienced U.S. Air Force F-16 pilot climbed into a simulator to fight a very strange opponent. There was no pilot in the other cockpit. He was fighting AI.

DARPA had been running the AlphaDogfight Trials, a competition where different AI systems learned how to fly and fight simulated F-16s. One of them, built by a small company called Heron Systems, beat every other AI in the tournament. Then DARPA gave it a human opponent. The rules were simple: two F-16s, close-range combat, guns only.

The AI won the first fight, then the second, and kept going. By the end, the human pilot had fought the machine five times. The final score was 5–0.

The AI could constantly calculate distance, position and timing without getting tired or overloaded. It could react almost instantly and keep making tiny corrections while the fight was happening, while the human pilot was still dealing with the normal limits of being human. It was only a simulation, but he didn’t win once.

This Had Happened Before

The warning signs had actually appeared years earlier. In 2016, retired U.S. Air Force Colonel Gene Lee fought another AI called ALPHA in simulated air combat. Lee had decades of experience, yet the AI still gave him serious trouble.

At the time, it was easy to shrug off. Computers had already beaten humans at chess, so maybe this was just another computer beating somebody inside another computer. But AI kept moving into harder problems where things changed constantly, where it had to react quickly and guess what the human would do next.

And humans kept losing.

Then Another Human Lost 5–0

In 2019, DeepMind put an AI called AlphaStar against professional players in StarCraft II. That might sound like just another video game, but it isn’t a simple one. Players have to build armies, manage resources, attack, defend and change plans while the other side is doing the same thing. They never know exactly what the enemy is planning.

AlphaStar played professional player Grzegorz “MaNa” Komincz, and the score was strangely familiar: 5–0 for the AI. Later versions were deliberately limited so they couldn’t simply win by clicking faster than a human. They still became better than almost every human player.

That mattered because AI wasn’t just becoming fast. It was becoming good at making decisions while things were changing around it.

The Humans Were World Champions

In 2023, researchers at the University of Zurich built an AI drone pilot called Swift and raced it against some of the best human drone pilots in the world.

Drone racing is brutally fast. Pilots wear goggles that show a live camera feed from the drone while they fly through gates, around corners and past obstacles at high speed. One mistake can end the race instantly.

Swift had to do the same thing without a human holding the controls. Researchers put the AI against three champion racers, including world champions. Swift won 15 of 25 races and also recorded the fastest lap.

Now the old excuse was getting weaker. This wasn’t chess, StarCraft or a fighter-jet simulator. A real machine was moving through real air, dealing with real cameras, real motors and real mistakes, and the AI was still winning.

Then They Put AI in a Real F-16

After the 2020 dogfight, there was one huge question left: could AI do this in a real fighter jet?

Simulators are clean, but real aircraft aren’t. Wind changes, sensors aren’t perfect, hardware shakes and things go wrong. So DARPA and the U.S. Air Force put AI into a specially modified F-16 called the X-62A VISTA.

This time the AI wasn’t flying pixels. It was flying the airplane.

The AI-controlled jet flew against a human-piloted F-16 in real combat-style maneuvers over Edwards Air Force Base. The aircraft flew at high speed and came within a few thousand feet of each other. Two safety pilots sat inside the AI-controlled aircraft in case something went wrong, but they didn’t need to take over.

DARPA never released a simple score showing who won each fight. By then, the score almost didn’t matter because the important thing was that AI had crossed the line from fighting a pilot inside a computer to flying a real fighter jet against one.

Then It Stopped Being a One-Off Experiment

There was still one way to dismiss the story. The X-62A was special. It was basically a flying laboratory, one unusual airplane doing one unusual experiment.

Then came VENOM.

In 2026, the U.S. Air Force began testing more modified F-16s that could hand control over to AI. A human pilot could fly the aircraft normally, then let the AI take over. That matters because the idea is no longer tied to one special research jet. The military is building a larger test program around it, where different AI systems can be loaded into the aircraft, tested, changed and flown again.

The timeline moved fast. In 2016, AI was fighting experienced pilots in simulations. In 2020, one beat an F-16 pilot 5–0. Then AI beat professional gamers and world-champion drone racers. A few years later, AI was flying a real F-16 against a human, and now the Air Force is building more aircraft that can do it.

That is a lot of ground covered in about a decade.

It Isn’t Just Faster Reflexes

The easy explanation is that computers are simply faster than people. Sometimes they are, but that isn’t the whole story anymore.

AlphaStar had to make plans. Swift had to fly through the real world. The dogfighting AI had to predict where another aircraft was going and react before the human could gain an advantage. These systems were learning more than speed. They were learning timing, strategy and prediction.

They could also practice again and again without risking a real pilot. A human fighter pilot needs years of training, while an AI can run huge numbers of practice fights inside a computer. Then whatever it learns can potentially be copied into another aircraft.

That changes things very quickly.

The Pilot May Not Disappear

The goal isn’t necessarily to get rid of human pilots completely. It may be to change their job.

Instead of one person flying one aircraft and worrying about every turn, climb and attack, AI could handle more of the flying while the human focuses on the bigger picture. Eventually, one person could direct several AI-controlled aircraft at once. The human becomes less like the driver and more like the commander.

There are obvious advantages. AI doesn’t get tired, panic or need years of flight school, and an unmanned fighter can be sent somewhere commanders might hesitate to send a human. If the aircraft is destroyed, no pilot dies with it.

For more than a century, fighter aviation has been built around one basic idea: the airplane is important, but the pilot is more important. Then AI started beating humans in games, beat champion drone racers, beat an F-16 pilot five times in a row, and eventually started flying real fighter jets against human pilots.

At that point, the military stopped asking whether AI could handle the simulator.

They just handed it an airplane.