Here’s what you’ll learn when you read this story:

• Ukraine’s Bumblebee and the Pentagon’s Bumblebee V2 are examples of how AI-assisted interceptor drones are becoming more common in warfare.
• However, the systems being used today still depend on humans for launch, guidance, and decision-making, and they’re not fully autonomous.
• Experts say that military will use AI drones as a helpful tool to improve outcomes, especially in tough combat environments.

A mysterious weapon hovering across the battlefields of Ukraine has torn back the curtain on the new era of modern warfare. Known as the Bumblebee drone, this relatively small quadcopter drone uses artificial intelligence to hunt and kill other drones, by tackling them head-on and ramming them out of the sky.

Ukraine’s Bumblebee is just one of many unmanned systems that can autonomously track and strike targets. Small and cheap to produce, the Bumblebee has forged a reputation as an effective killer of Shahed drones. What is the secret to its success—and does a growing military focus on AI-powered drones mean they are the ideal weapon of the future?

Ukrainian forces are not the only troops wielding interceptor drones. In fact, the Pentagon moved to acquire what can fairly be called a sister of the Bumblebee: the Bumblebee V2, produced by Perennial Autonomy, a U.S. defense technology company, which makes both of these drones.

“The Bumblebee V2 is a remarkable tool and, for the first time ever, we will be able to give our troops a dual purpose, low-cost … [Intelligence, Surveillance, and Reconnaissance] platform designed to be used at the tactical edge,” says U.S. Army Major Cole Price, Capability Manager for the Pentagon’s Joint Interagency Task Force (JIATF) 401 Rapid Acquisitions Division. The task force, whose mission is U.S. national counter-drone defense, announced an agreement in February to obtain the Bumblebee V2.

The drone, a first-person view system with four rotors, knocks out adversary drones with a direct physical collision. Within drone warfare, a direct physical hit on a drone is called a “hard kill” or kinetic strike. The Bumblebee V2 uses AI targeting recognition paired with human operators to lock onto and strike an adversary drone, Price explains.

But just how smart are these AI-powered drones? Their current capabilities are often overestimated because they continually require humans to refine them, according to U.S. Army Major Wolf Amacker, who manages adapting lessons learned from unmanned systems within the U.S. Army Aviation Center of Excellence.

Amacker says that the drones use AI to home in on targets after being initially guided by humans during launch. “So you get it up, you launch it immediately and then it gets to a point where it can either help the operator maneuver it to the target, or it will have a scalable control that takes it more on itself to strike,” Amacker explains regarding the Ukrainian Bumblebee drone.

He notes that many AI-equipped drones are replicating the military’s existing targeting capabilities in a condensed—and notably cheaper—form.

“[The Bumblebee] is just using some AI to help manage the visual environment to lock onto a target,” he says. While it’s similar to earlier target tracking software, AI improves continually, he adds, and the use of these drones is growing steadily. “The effectiveness of these is proven by the fact that a lot of different militaries are looking into how they can do this.”

“AI” is often used as an umbrella term for an array of different software, causing people to assume that drones have more esoteric powers than they actually do, U.S. Army Major Nick Lupo, Executive Officer of the 2nd Battalion, 13th Aviation Regiment, explains.

He says that there is “a wide variance in the capabilities of AI powered drones,” and that “many systems touted as AI are merely logic trees rather than reasoning and learning” systems. Some AI systems can be stunted in their performance ability if people don’t feed them refined information, he explains.

Regardless of the current capabilities of AI machine learning, it is undeniable that drones have upended the concept of a battlefield. Machines are steadily replacing humans in the front line, Lupo says.

“The limit of human elements, once considered the Forward Line of Troops (FLOT), is conceptually transitioning and projecting power further, to a Forward Line of Robots (FLOR) and Forward Line of Sensors (FLOS),” Lupo said. This means more unmanned systems will operate in battlespaces ahead of humans in multiple domains, rather than just the air, he said.

“Additionally, it may take the form of unmanned platforms being the preferred method to initially engage with and destroy the enemy,” he said.

While the idea of robots zeroing in on targets on the frontlines may seem like it belongs in a rather unpleasant sci-fi film, that future is already upon us. Drones have unleashed a Pandora’s box of military capabilities, and adversaries are using the technology for nefarious purposes. Organized crime groups around the world now use drones as weapons and smuggling tools, while hybrid drone attacks on infrastructure like airports cause fear.

For U.S. military forces, staying ahead of the game means not only having an array of drone and counter-drone technology, but learning how to use these tools quickly and effectively. “Drones are a defining threat for our time. Technology is evolving fast and our policies, technology, and training must adapt to meet this reality,” Price says, adding that the JIATF-401 task force is taking a government-wide approach to nationwide counter-drone defenses.

Although Unmanned Aircraft Systems have dramatically changed war overall, and particularly aerial warfare, both Amacker and Lupo maintain that AI-powered drones will not replace human aviators.

This is not due to sentimentality but practical warfighting reasons. For all the prowess of robots, we mortals have some innate flesh-and-blood superpowers—such as being impervious to jamming, becoming frozen due to battery failure, or otherwise being vulnerable to the effects of electromagnetic warfare.

“In a degraded environment, depending on how electromagnetic warfare is going, an unmanned system can be completely degraded, defeated and taken out of the fight versus a manned aircraft,” says Amacker. A compromised manned aircraft would still have a human on board “able to utilize their thinking power to get through those effects.”

And AI models lack innate human intuitions, the inherent “curiosity or even that gut feeling when something is going wrong,” he adds. “It’s cold. It’s [either] in parameters or not within parameters.”

“As we get better at identifying targets through AI, the enemy is going to find ways to fool AI.”

The real responsibility rests with humans to shape AI models and see them developed both effectively and responsibly, Amacker says. “The idea is that we have AI-enabled systems that are out there to help round out our understanding and actions in the environment.” Soldiers are currently “in the loop” when using autonomous systems, he says, making decisions based on machine recommendations.

Lupo expressed similar views. “This isn’t a question of AI versus humans,” he said. “It’s a question of how well they complement one another. Human aviators bring creativity and intuition, judgement and accountability, adaptability, and teaming—these are tenets critical to training AI behavior and models.”

True to the ironic nature of war, drones are not only a means of inflicting damage on adversaries, but also the keys to sparing humans from death and injury. “The intent of drone inclusion is to minimize loss to human life,” Lupo said.

Amacker believes that AI is vital to military progress. “I really think AI is meant to help us and enable us. I heard a leader say, ‘You will either be an AI-enabled leader, or you will be left behind.’” Hopefully, AI will “keep [our soldiers] out of dangerous positions,” he adds.

In any case, AI’s emerging capabilities have opened a new arena of competition—one that adversaries won’t hesitate to exploit. Forces grappling in an age of robotic surveillance will need to explore entirely new methods of camouflage.

“As we get better at identifying targets through AI, the enemy is going to find ways to fool AI,” Amacker says. “It’s going to go back and forth, which is going to require some human curiosity and some human intelligence to review and help refine that process.”