SkyFall’s AI Interceptors Hunt Shaheds Without a Pilot’s Eyes, and Ukraine Knows the Risk It Is Taking

A Ukrainian drone manufacturer called SkyFall has flown artificial intelligence into the one job interceptor pilots could never do fast enough: spotting an incoming Shahed before the human eye registers it. At a launch site in a central Ukrainian pine forest, the company demonstrated its P1-SUN interceptor against a lightweight replica of the Iranian-designed attack drone, and the onboard system detected the decoy long before the pilot could have. The machine marked the target with a green square on the operator’s screen, then flew itself toward the kill once the pilot let go of the controls.

SkyFall says its interceptors have made dozens of AI-assisted strikes on Shahed-type drones since November, among thousands of interceptions overall. The system that guided this one was trained on more than 10,000 videos of actual Shahed interceptions. That detail is the whole story. Ukraine is not buying autonomy off a shelf. It is manufacturing the training data in real combat, every night, faster than any peacetime program could.

The reporting comes from a New York Times dispatch filed from a secret location in central Ukraine. For readers who have followed DroneXL’s interceptor coverage since the Pentagon started knocking on Kyiv’s door in March, the forest demonstration confirms a shift we have been tracking for months: the interceptor war is moving from human reflexes to machine perception, and SkyFall is at the front of it.

SkyFall’s P1-SUN turns the pilot into a supervisor

The P1-SUN is a vertically launched interceptor with a bullet-shaped nose, built on a partly 3D-printed modular airframe that costs Ukrainian units around 1,000 dollars per unit. In the demonstration, the pilot flew it toward the Shahed decoy until the shape was clear on screen, gave the auto-targeting system an order to follow, and released the sticks. The interceptor steered itself the rest of the way. The pilot’s last act was pressing the button to go in for the strike.

That division of labor matters more than it sounds. A human flying an interceptor spends most of the engagement just finding and tracking a small, fast target against open sky. The AI compresses that window. The stated goal of the system is to shorten the time needed to detect and track an enemy drone, which is the exact bottleneck that lets Shaheds slip through on mass-attack nights. We covered the P1-SUN’s place in Ukraine’s interceptor field earlier this year, when SkyFall said it could build up to 50,000 units a month and export a share of them without starving its own forces.

The pilot’s screen in the demonstration showed the system running in plane mode with a firmware build stamped in the corner, the kind of detail that tells you this is shipping software on real hardware, not a lab prototype. SkyFall also conducts exercises with replica targets, but the company admits it could never fly enough decoys to fully train the system. The 10,000 combat videos do what staged tests cannot.

The 95 percent automation claim already has a name, a price, and a combat date

Ukrainian Defense Minister Mykhailo Fedorov recently said a company in the government’s Brave1 defense-technology program had built a system that automated 95 percent of the interception process. It cannot launch on its own. It takes over after an operator selects a target and authorizes engagement, then guides the interceptor toward the Shahed, identifies it, and homes in. The Times identified the company as MaXon, a defense startup founded early last year.

DroneXL readers already know more about MaXon than the wire story tells. We reported on June 9 that MaXon Systems builds a fixed-wing interceptor that launches, transits, and strikes with one operator pressing two buttons, at roughly 3,500 dollars per unit, and that its first confirmed combat interceptions were carried out on June 8 by Ukraine’s 12th Separate Special Purpose Center in Kharkiv Oblast. The “95 percent” figure Fedorov cited is not an abstract milestone. It has a builder, a sticker price, and a battlefield record that is now a week old.

The direction of these systems points one way. With more autonomy, interceptors could stand ready to launch automatically after radar detects an attack, and SkyFall says it is already testing such a system. The payoff for an outmanned military is one pilot overseeing several engagements at once instead of one at a time, which is how it works today.

The same forest tests pointed the AI at people on the ground

The demonstration did not stop at aerial targets. SkyFall’s tests applied AI to help first-person-view drones find targets on the ground, both equipment and personnel. As a dark green minivan drove in and out of the trees, a pilot designated points near it and the system adjusted its aim onto the vehicle automatically. Then the exercise turned to human targets. Members of the team walked the clearing pretending to be Russian soldiers while a drone’s targeting system located a colleague and waited for the pilot’s order to strike or stand down.

This is the part of the story that should sit uncomfortably with anyone who cheers the interception numbers, and it should. Many AI systems under development can autonomously identify objects like vehicles. Some can single out people, and SkyFall is among the companies quietly testing people-targeting in Ukraine. Officials at firms like SkyFall and military commanders keep stressing that a human confirms the kill. The honest framing, by their own admission, is that human confirmation is now an ethical choice rather than a technical requirement. Telling friendly forces from enemy combatants by camera remains a real problem the machine has not solved.

Human rights groups argue that reducing a life-or-death decision to an algorithmic calculation is a threat to humanity. Ukrainian developers and leaders say they hear that, and keep building anyway, because the alternative is letting Shaheds through. Both things are true at once, and pretending otherwise is how you end up writing propaganda for one side or the other.

The interceptor edge is becoming an export business

Ukraine’s interest in autonomy goes beyond defending its own cities. It wants to become a major defense exporter, and the combat record is the sales pitch. After the United States and Israel opened their war against Iran this year, those countries and Persian Gulf states burned through hundreds of costly interceptor missiles to down cheap Iranian-made Shaheds. Ukraine offered its low-cost interceptor drones, without AI, as an alternative, and that pitch has helped President Volodymyr Zelensky cut security deals with Gulf states.

We have tracked that export turn closely, from the legal change that finally let Ukrainian firms sell surplus systems abroad to the Gulf monarchies and U.S. commanders now lining up to buy. Several Ukrainian companies have introduced remote-control technology that lets missions be flown from Kyiv, the kind of capability that turns a domestic defense program into a service a foreign buyer can rent. Germany has already funded thousands of Ukrainian interceptors for delivery, a sign that European buyers are past the demonstration stage.

SkyFall’s head of autonomy and computer vision, who asked not to be named for security reasons, said his company alone can make 50,000 interceptors a month. The harder constraint is people. Training enough skilled pilots is the bottleneck, and the ones Ukraine has stay on duty around the clock. That is the real argument for automation, stripped of the marketing: not that machines are better, but that there are not enough humans to go around.

DroneXL’s Take

The interceptor story has had three chapters, and DroneXL has covered each one. First came the cheap manual quadcopter, the Wild Hornets Sting, where a pilot in goggles flew the drone into the Shahed and pressed a detonation trigger by hand. We documented that fully manual workflow in May. Then came the price-and-scale race, the P1-SUN against the Sting against the rest, which we mapped in our interceptor scaling report in March. This forest demonstration is the third chapter: perception moving from the pilot’s eyes to the drone’s. The through-line is that every step has been pulled forward by Russian adaptation, not by a roadmap. Russia added rear-facing infrared spotlights to blind thermal cameras, flew Shaheds lower to defeat GPS, and strapped air-to-air missiles onto attack drones. Each move forced a Ukrainian counter, and AI perception is the latest one.

I have been writing about this specific company’s hardware since our March field guide, when the P1-SUN was a 1,000-dollar airframe with optional AI targeting and a kill count in the low thousands. Three months on, the optional part is becoming the point. That is a fast curve, and I have watched the price-per-kill argument harden from a talking point into the reason Gulf buyers are signing.

Here is what the demonstration did not resolve, and what the Times reporters did not get an answer to. SkyFall says it is testing a system that launches automatically on radar detection. Fedorov says MaXon already automates 95 percent of the cycle, holding the line only at target selection and the final strike order. Those two facts, put together, describe a system where the only remaining human steps are the first and the last. Nobody on the record said what happens to those two steps under a saturation attack, when hundreds of Shaheds arrive faster than operators can authorize them one by one. That is the question that decides whether “human in the loop” survives contact with a 2,000-drone night, and it is an open question, not a prediction. Watch Brave1’s next public capability disclosure for whether automatic launch-on-detection gets a combat date the way MaXon’s two-button strike just did. If it does, the loop will have quietly gotten shorter, and the ethical choice the commanders keep describing will be the only thing left holding it open.

Sources: The New York Times, Military Times.

DroneXL uses automated tools to support research and source retrieval. All reporting and editorial perspectives are by Haye Kesteloo.