This is the moment everyone in defense tech says they’re ready for—right up until it actually happens. An AI-controlled fighter jet went up against a human pilot in a real dogfight, and the result is classified. That combination should make you uneasy. Not because AI can’t fly. Because once we start hiding the outcomes, we also hide the lessons, the mistakes, and the real reasons it worked or didn’t.
Based on what’s been shared publicly, the X-62 VISTA is an AI fighter jet built on an F-16 Viper airframe. In 2023 it engaged in close aerial combat against a manned fighter, monitored by the Air Force. That alone is a line-crossing milestone. It’s one thing to run simulations or fly predictable test routes. It’s another to put an AI in a chaotic, fast, close fight where the “right” move changes every second and a single bad read can cascade into disaster.
And now, as of late 2025, the aircraft is going through a Mission Systems Upgrade—better radar, sensors, and AI—meant to expand what it can do.
From our company’s perspective—building drone detection radar systems and AI that fuses different sensor feeds—this is both impressive and dangerously easy to misread.
The impressive part is obvious: if an AI can hold its own in a dogfight, that suggests a machine can perceive, decide, and act faster than a person under stress. It also suggests the control system can manage real physics, not just clean lab conditions. That’s not nothing. Anyone who’s watched real-world sensor data knows it’s messy. Targets blink in and out. Reflections trick you. A “track” looks solid until it doesn’t.
But I don’t love the story people will tell themselves next: “AI beat a pilot, therefore AI is ready.” The classified outcome is a giant red flag here. If the AI performed poorly, classification protects the program from embarrassment and slows learning across the broader ecosystem. If the AI performed extremely well, classification protects the advantage—but also prevents the public and even many inside industry from understanding what constraints kept it safe. Either way, secrecy invites overconfidence in the wrong places.
The part that matters most to us isn’t the jet’s airframe. It’s the sensors and the fusion between them. In a dogfight, you’re not dealing with a clean, cooperative target. You’re dealing with deception, weird angles, speed, background clutter, and constant change. That’s where radar drone detection and passive sensors, combined with smart fusion, can either produce clarity—or produce convincing nonsense.
Imagine a real base defense scenario, not a test range. You’ve got small drones near the perimeter, friendly aircraft overhead, birds, weather, and the usual radio noise. If your system fuses radar, electro-optical, and other sensor inputs the wrong way, you can create a “confident” track that is simply wrong. AI loves to be confident. Humans do too. The difference is speed: an AI can be confidently wrong at a pace you can’t interrupt in time.
Now scale that to air-to-air combat. A human pilot might notice a mismatch—“my eyes say one thing, my display says another”—and hesitate. An AI might treat that mismatch as just another data point and continue. Sometimes that will be better. Sometimes it will be catastrophic.
The upgrade piece is where I get torn. Advanced radar and sensors can absolutely make these systems safer and more capable. Better sensing gives the AI more honest inputs. Better fusion can reduce false targets and improve tracking. That’s what we work on, and we believe in it.
But upgrades also create a temptation: keep piling on capability before you’ve proven the discipline around it. If you add more sensors and more AI, you also add more ways to fail. Not dramatic “it falls out of the sky” failure—more subtle failure. The kind where it flies beautifully and makes one bad decision because it trusted the wrong cue at the wrong second.
There’s also a human consequence people dodge. If AI starts taking on the hardest, highest-skill parts of combat aviation, what happens to the pilot pipeline? Skills are not static. They’re habits built from reps. If the “top” of the mission shifts to automation, humans may get fewer chances to build judgment under pressure. Then, when the AI fails—or when rules require a human to take over—you’re asking a person with less lived experience to jump in at the worst moment. That’s not a fair trade.
Of course, the opposite argument is real too: maybe putting AI in the most dangerous roles saves pilots’ lives and reduces risk overall. Maybe the AI can handle the first contact, the first merge, the first chaotic seconds, and a human stays back to command. That’s a future many people want. I get it.
But the classified outcome still hangs over all of this. If we can’t talk about what went wrong, we can’t build better standards. We can’t compare systems. We can’t pressure-test claims. And in our world, claims are cheap. We’ve all seen demos that look perfect because the hardest cases were quietly excluded.
So yes, we’re watching this milestone closely, because it validates the direction: better sensing, better fusion, better autonomy. And we’re worried, because the narrative can run ahead of the reality, and secrecy makes that easier.
If an AI fighter can dogfight a human pilot and we keep the results hidden, how do we stop confidence from becoming policy before safety and accountability are actually earned?