This is a piece originally written for a national security audience at Frontiers. Although I think the ceiling of war is much, much higher than autopilot quadcopters, it's also important to understand how much AI is already lifting the floor, and just how vulnerable the world is to accessible weapons of mass destruction.

Drones are cheap, disposable, and the future of war. Over the past four years, we have seen platforms, missiles, and heavy infantry become increasingly obsolete in the face of $500 drones carrying a pack of explosives—a cost advantage that has let Iranians and Ukrainians alike neuter the conventional capabilities of their great power rivals. Eighty percent of casualties in the bloodiest war since 1945 are from drone strikes, Russia has managed to lose one-third of its fleet to a country without a navy, and the US is spending millions of dollars to intercept five-figure Shaheds flying over the Strait of Hormuz.

All this is the result of a technology that is still immature. The violence inflicted by today’s drones is the handiwork of the scant few that manage to evade countermeasures (a mix of radio jamming, high-power microwave weapons, missiles, automatic cannons, interceptor drones, and nets) before making contact. These defenses exploit the inherent limitations of drones—human guidance, GPS feedback, flight exposure, radio links, range—to take them down en masse. And yet, even though 75% of drones manufactured today never reach their targets, they have nonetheless been strategically decisive in Ukraine and elsewhere.

These limitations will not hold for long. Just like bacteria being overexposed to antibiotics, overexposure to counterdrone tech has created an arms race for ever-more-autonomous drone technologies. In the process of facilitating this arms race, states are likely to incrementally create and deploy an entirely new class of WMD—one that could provide rogue states with the nonnuclear means to threaten superpowers, or hand terrorists the means to selectively assassinate their political targets or civilians en masse.

Unfortunately, drone weapons intended for mass destruction have few barriers remaining to mass deployment. Even well before they reach the level of autonomy needed to surgically take out hardened targets on the battlefield, drones will be capable of employing their existing ability to navigate interiors, find and track human targets, and deploy simple antipersonnel devices to indiscriminately threaten civilians. Below, we discuss the looming arrival of miniature autonomous weapons, the limits of counterdrone technology, and the applications of drones as weapons of mass destruction.

Breaking the Last Barriers to Autonomous Weapons

The ideal drone weapon is a slaughterbot: a small, fully autonomous weapon system that can independently select and hunt its targets. For the most part, the necessary technology for such weapons already exists: airframes the size of a fist and the capability to track human targets are already on the front lines in the form of reconnaissance drones and semiautonomous weapons like the Russian V2U. Even now, these micro drones are agile and autonomous enough to hunt down and kill small moving targets like mosquitos—to say nothing of the advances in drone technology expected in the coming years.

From here, the only barrier to weaponization is integration: improving navigation enough to make drone technology useful for mass homicide in an urban setting, as well as packing the necessary guidance, sensor, and payload technology onto a small and energy-efficient chassis. Regrettably, this seems like less of an engineering problem than one of mission design: so long as the attacker is willing to accept indiscriminate targeting and use simple payloads aimed at civilians, the underlying technology is already—or very nearly—ready for practical use.

To understand how close we are to these kinds of weapons, it helps to understand why we do not already employ fully autonomous drones. Right now, the most deadly drones are small first-person-view (FPV) units, with the majority of Russian and Ukrainian casualties alike stemming from direct drone strikes. For the most part, however, these small drones are being piloted directly by humans, either through a radio link or a spool of fiber-optic cable, with just the final leg of the attack being delegated to an AI targeting system.

So, ethics aside, why does drone warfare still depend on human pilots?

Distinguishing enemy targets from friendly assets on the battlefield still requires humans. The main problem is that battlefields are intrinsically adversarial environments: an autonomous drone needs to avoid friendly fire on its own infrastructure and troops, anticipate pre-positioned counterdrone defenses, deal with camouflage and decoys, and destroy hardened targets like vehicles and infrastructure. This is especially complicated if you need drones to autonomously work together to accomplish an objective, such as by having specialized units target defenses to allow others through. For the moment, only humans have the skills to distinguish a camouflaged ally from an enemy unit, or to exploit the underbelly of an armored vehicle.

Designing drones for indiscriminate mass destruction is easier than for precise battlefield use. Unfortunately, adversarial target selection is not a meaningful barrier for applying drones to mass terrorism. Destroying an armored vehicle needs the skill to reason about and single out its weak points; but, to kill an unarmored human, a drone need only make contact with them and deploy an explosive or poison needle. The targeting requirements and level of autonomy needed to indiscriminately massacre civilians, in other words, are much lower than what you’d need to selectively destroy hardened targets on a battlefield. In the words of Ukraine’s Azov brigade, “If you don’t care about civilians, you can simply hit any target that moves.”

Drones can already navigate indoor environments and track humans. Aside from requiring guidance for target selection, autonomous drones also need the ability to navigate. Urban environments are cluttered and leave room for targets to shelter indoors, so drones must infiltrate and sweep through them to be maximally lethal. Autonomous navigation of this caliber already exists: as far back as 2022, drones have been capable of mapping and tracking indoor environments to find humans, a skill used to locate hostages and scan through tunnels for enemy soldiers. These kinds of drones typically cost tens of thousands of dollars, but they have an expensive use case: infiltrating a GPS-denied location, then escaping to broadcast information. If you do not need the drone to survive and report back, and if you do not care if your drone can tell whether someone is surrendering or not, then you do not need expensive sensors and plenty of onboard compute for decision-making—just the bare minimum to identify a target as human and fly at them.

Disposable drones that can autonomously attack civilians may soon be relatively cheap. For comparison, a last-mile module, an upgrade chip that lets FPV drones visually hunt down targets when they lose connection, is about $500. One-way autonomous search, target selection, and mapping, at least for this anti-civilian use case, would likely be similarly inexpensive—already, visual and laser mapping systems have been demonstrated, in principle, that can work their way around a room and track humans on a few hundred dollars of hardware. If a military system that navigates to the entrance and then plans an indiscriminate suicide mission inside is not already achievable for just thousands of dollars, it will be in a matter of years.

Of course, most countries do not have the motivation to build these kinds of systems and drive down their unit economics. Discrimination, ethical or otherwise, is useful on the battlefield. If the costs of autonomous targeting keep falling or AI guidance improves, however, states might be tempted to start employing indiscriminate drones as a means of deterrence, or as a cheap way to enable terrorist proxies.

Weaponization

Given these capabilities, how could these drones be weaponized and delivered in practice? The main limiter on these kinds of small drones is energy: a modern 30g micro drone like a Black Hornet can fly for about half an hour before needing to recharge, while a bigger FPV carrying an explosive payload will usually last only 15 minutes. However, there are plenty of ways to stretch this energy budget further for the purpose of mass destruction, even without better battery technology.

Lethal payloads could be much smaller.  Today, FPV units usually carry about a kilo of explosives, because they need to be flexible enough to target vehicles and defensive infrastructure as well as enemy soldiers. If the goal is to break windows and kill humans, however, even just 20g of frag explosives is enough at several meters of distance, with much less needed for a lethal wound at near-contact. Alternatively, something as simple as a spring-loaded needle, coated with a microgram quantity of a poison like botulinum toxin or a nerve agent, would be immediately lethal on contact.

Energy expenditure on delivery and flight could be significantly reduced. Rather than have the drones travel constantly under their own power, it is much more efficient to carry them into position with a drone mothership or even the hull of a missile. Once the drones are released into the air, they then need to navigate to a building without expending much power. A simple way to do this is to give the drone glide wings, letting it drift forward for most of its flight rather than loiter directly. This technique is already used in Ukraine to stretch the range of basic FPVs over 40 miles beyond the front line.

Drones could perch and idle, rather than hovering, while waiting for targets. Finally, and most importantly, the drones would be ambush hunters. Rather than loiter in the air, it is much more efficient to perch and idle under cover, running a milliwatt acoustic and visual sensor every few seconds until a target is detected. Although drones would still need to expend the energy to infiltrate a building, once inside, they could afford to act like improvised landmines for days or weeks until their battery finally dies.

Urban Drone Attack Sketch

To appreciate the implications of these capabilities, it helps to outline what a mass urban drone attack would actually look like.

At the outset, drones are delivered near the city through a large mothership, which independently stores them in a cargo hold. Depending on whether this is itself a larger drone, a plane, or the warhead of a missile, each could feasibly deliver anywhere from hundreds to tens of thousands of drones at once. Alternatively, the drones could be smuggled in through a pre-positioned shipping container, which then launches its contents from a nearby port or logistics yard. Either way, a large number of drones are then scattered above the city at a low altitude or are dispersed near street level.

How large? Assuming that the drones are arranged like capsuled quadcopters, they can be packed in extremely efficiently. Using the Ninox-40 system as an example, storage counts would be in the range of:

Once released, the drones glide down haphazardly, aiming to land in regions not already populated by other units. Some drift directly onto public crowds and attack their targets right away. Others do not land near a direct target, switching their focus to look for entrances, such as doorframes, windows, and tunnels. If the attacker is particularly sophisticated, these drones might be accompanied by a handful of larger drones carrying breaching charges, designed to create more openings for the main force. These drones then begin drifting through the building interior for a few minutes, looking for victims. If they cannot find a new target or are unable to find a route inside, they move to a dark corner or roadside and begin passively idling.

Many civilians would die in the initial attack. But the aftermath would be much worse. With the drones saturating the city, there would be no opportunity for any survivor to leave their barricade and seek help, and no way for resources and relief units to flow back in. Anyone in need of food or basic medical attention would be unreachable for weeks, during which the death toll would continue to mount. In effect, the entire city would be transformed into something akin to the Ukrainian front line today, with the omnipresent threat of assassination forcing the surviving humans to slowly starve out in isolated foxholes, without any chance of easy respite or rescue. But unlike on the front line, these humans would not be soldiers: most people will not have stashed food ahead of time, or armed themselves with anti-drone weapons, or sealed off every corner of their apartment with netting.

Defense Against Autonomous Weapons

Any improvements in drone technology will still need to contend with counterdrone defenses. It is precisely because drones are so threatening that states have invested heavily in tools like directed energy weapons and interceptor drones to counter them. If drones become even more strategically dominant, then we will surely see correspondingly greater counterdrone efforts. So why should we expect this arms race to resolve in favor of the drones, rather than their countermeasures?

Future drones will be less vulnerable to radio-frequency jamming. To appreciate the limitations of counterdrone technology, we can look at the difficulty states are already experiencing in their efforts to counter fiber-optic FPV units. These drones work by unspooling a thin fiber-optic cable behind them, letting a human pilot them directly for up to 40 kilometers without having to worry about GPS or input jamming.

The reason these drones are so effective is that they’re naturally resilient against common counterdrone techniques. Throughout the Ukraine war, the most important anti-drone tool has been radio-frequency jamming. As long as a human pilot is selecting targets and telling the drone where to go, or as long as the drone depends on GPS coordinates to navigate, overwhelming or spoofing those broadcasts with a countersignal will cause the drones to fly harmlessly off course. With a fiber-optic drone, all the piloting happens through a direct data link, so this kind of countersignal is harmless. The same is true of any autonomous drone—as long as all the decision making is processed on board, there’s no human input to jam or spoof in the first place.

Interception is asymmetrically difficult against small, stealthy drones. In cases where it’s difficult to achieve an electronic soft kill on the drone, the backup option is to physically destroy them with a kinetic interceptor. The reason this is a backup is that it’s expensive and prone to blind spots. Small FPVs are cheap and agile enough that it’s easy to spend much more to down them than they’re worth, especially when relying on the expensive autocannon in the path of a flying explosive. This is why direct kinetic interception is usually reserved for expensive drones with fixed flight paths (like Shaheds) and kept as an option of last resort for small drones.

There is also the problem of terrain blindness: if you cannot see a drone, you cannot shoot it down. Modern LSS (low, small, slow) drones are already so tiny that they are difficult for radar to distinguish from birds, trees, and ground clutter, allowing operators to fly them near the treeline until they are too close to reliably intercept. This is especially problematic in an urban environment, where there are many places to hide and many opportunities for collateral damage.

Physical barriers and EMP weapons will struggle to reach sufficient coverage in cities. Instead, the most effective tools against autonomous drones will be structural barriers and directed energy weapons. One of the most visible effects of the war on Ukrainian infrastructure is the miles-long corridor of nets that cover roads and even cities, without which transports would be constantly exposed to drone strikes. But nets are not hard barriers: if the initial carrier punches through, if a few specialized units carry breaching tools, or if the drones are released from below, there is little to stop them from flying unimpeded and sieging the streets beneath. To actually destroy the drones efficiently, the best candidate is a directed energy weapon, particularly an electromagnetic pulse (EMP) device. It creates an electric field powerful enough to short-circuit any electronics in range, physically destroying the drone controllers and motors. Unfortunately, cities are far from ideal places to deploy such weapons; building material is reasonably effective at shielding against an electric pulse, which means that, if the EMP does not catch the drones in the initial sortie, lack of a clean sight line will significantly weaken this countermeasure against drones dispersed throughout a city.

In other words, these defenses can be locally effective but struggle to get sufficient coverage, especially over an active residential area. They are also, of course, proactive defenses: they have to be installed ahead of time in order to have any defensive effect, which gives the attacker time to assess and plan around them.

Strategic Applications for Rogue Actors

In summary, small lethal autonomous weapons will be very difficult to defend against, especially in the context of securing large urban environments. This leaves the question of which actors would want to use them, and to what end?

Advanced drones would be more useful for rogue states than for superpowers. The first countries to acquire fully autonomous drones will be those that have a precision manufacturing base and frontier AIs capable of helping with military R&D: namely, the US and China. However, autonomous drones are only marginally useful for the existing great powers: a new option to further deter a Taiwanese invasion, a cheaper way to conduct remote assassinations, and a way to maintain conventional parity with rivals’ own drone mass. Rather than meaningfully changing the balance of power between nuclear states, near-term autonomous drones will likely be most useful for rogue actors seeking new weapons of mass destruction to enhance deterrence. In this regard, indiscriminate autonomous drone swarms possess many advantages.

Drones are inherently simple, which makes it very difficult to control their proliferation. A central reason why modern military drones are so cheap and widespread is their simple design, combined with the commodification of key components like memory and compute. As we have seen with semiconductor sanctions on Russia, the underlying materials are too accessible in ordinary consumer supply chains to easily deny mass production. Moreover, even if a specific country could be cordoned off from general drone production, Russia and China have proved willing to help export their military designs to allies. Even terrorist groups might be able to build, or at least acquire, large quantities of drones for urban attacks, either through covert smuggling or state sponsorship.

Drones can be used conventionally, not just for WMDs, making restrictions hard to enforce. To the extent that states can impose restrictions on drone manufacturing or acquisition, they have largely failed to do so. This is partly because there is not a clear point of intervention in drone development: unlike other WMDs, whose primary purposes are for terror and leverage, any state building drones can claim that they are for civilian or conventional military use rather than for mass destruction. As a result, international organizations like the UN have done little more than to condemn the principle of autonomous weapons, without yet addressing basic questions like the definition of autonomy.

Drone swarms are a far more precise, controllable deterrent than other nonnuclear WMDs. Chemical weapons, although useful for terror, are difficult to widely disperse and threaten entire cities with. This combination of extreme fear and limited destruction thus invites the risk of extreme escalation in retaliation, making them poor deterrents. On the opposite end of the spectrum, bioweapons are simply too destructive, symmetrically threatening those who deploy them, as well as too invisible and delayed to create an immediate effect. Massive drone swarms, however, could be used to reliably siege an entire city while remaining contained within it.

Autonomous weapons are fundamentally hard to stop, particularly when aimed at civilians. Even if states are able to secure individual targets, such as military bases and political offices, securing the whole of society such that there are no soft targets for advanced drones would be enormously challenging. Even aside from their sheer size, cities are difficult to cover because they are both open enough to let people move around freely and dense enough that interceptors will usually lack a clear line of sight, risking collateral damage. If this technology were to proliferate widely, the future might be one of constant and extreme geopolitical tension, where even minor military powers are constantly tempted to assemble large swarms of murderous drones as deterrents.

Taking Future Technology Seriously

This piece is an exercise in taking the future of military technology seriously. That means being credulous that new offensive technologies are possible and imagining what a serious effort to abuse and counter them would look like in practice. It means asking ourselves what the logistics of putting nets and energy platforms over New York would be. It means looking at what it is like to live in a trench on the Ukrainian frontline, and asking if these kinds of experiences will always stay so far away.

Taking future technology seriously also means looking further ahead. By all appearances, the kind of indiscriminate weapon described above is not far off. Autonomous drones that blindly hunt down humans and siege cities, if they do not already exist, are held back more by ethics and military opportunity cost than any fundamental engineering problems. But there’s no reason to imagine that dumb, flying landmines are as far as drone technology will progress. Drones the size of mosquitoes, drones as cheap as bullets, drones so numerous their swarms blot out the sky, drones that sit on your power lines and in the sun and never leave. These are not only possible but inevitable: if all that happens is just the normal grinding of drone engineering and mass production, cheap swarms of thousands, or even millions, of killer drones will eventually be universally available, in the same way that trench warfare was inevitable the day the British fielded the maxim gun for the first time. Whether this happens in 5, 10, or 15 years is much less important than whether we are prepared to deal with it when it does. Some policies, like nonproliferation and defensive investment, can only work while the threat is still unrealized. Implementing these policies for drones, and any other future military, starts with taking what will be possible tomorrow seriously and planning for it today.