The Flock Goes to War: Drone Swarms and the Next Revolution in Warfare

By Matthew Parish

Saturday 5 September 2026

There is something faintly misleading about the expression “drone swarm”. It summons images of thousands of tiny machines pouring over the horizon like locusts, communicating instantaneously amongst themselves, selecting targets by artificial intelligence and reorganising their attack whenever one of their number is destroyed. The implication is that this technology already exists and that the principal question is when somebody will be sufficiently reckless to use it.

The reality is both less spectacular and more consequential. Genuine autonomous drone swarms remain at the frontier of military technology. What exists today is an assortment of technologies — automated navigation, machine vision, target recognition, electronic-warfare-resistant guidance, networking, collaborative sensing and software permitting one operator to supervise several aircraft — that are gradually being assembled into something resembling a flock.

And the distinction matters. A hundred drones flying towards the same city are not necessarily a swarm, any more than a hundred aircraft participating in an air raid constitute a flock of birds. A genuine swarm involves collaboration between the machines. That is where the revolution lies.

What is a swarm?

Nature solved the problem long before engineers encountered it. Watch starlings turning above a field. There is no commander issuing instructions to every bird. Each animal follows relatively simple rules concerning distance, direction, collision avoidance and the movements of its neighbours. Out of these elementary interactions emerges remarkably sophisticated collective behaviour.

The same principle explains schools of fish, colonies of ants and swarms of insects.

Computer scientists call this emergent behaviour. Complex collective actions can arise from relatively simple rules followed by individual agents. Drone flocking attempts to reproduce something similar electronically.

A primitive military “swarm” might therefore consist of six aircraft programmed to search different sections of a forest. If one drone disappears, the remaining five redistribute the search area. If one identifies something resembling an armoured vehicle, others move towards it to obtain different sensor perspectives. If the object is confirmed as hostile, one aircraft may attack while the others continue reconnaissance.

Experiments are already approaching this level. During the 2026 Army Warfighting Experiment on Salisbury Plain, British, American and Australian personnel operating under AUKUS tested groups of up to six drones that divided areas between themselves, searched for potential targets and returned imagery to operators. The exercise incorporated artificial intelligence, collaborative sensing and target recognition.

This is flocking in embryonic form. The important development is not six drones. Militaries have been able to fly six drones simultaneously for years. It is the software architecture that permits six eventually to become sixty, and sixty eventually to become six hundred.

Swarms that are not yet swarms

The Ukraine war has created considerable confusion about this terminology. Russia routinely launches enormous numbers of Shahed-type one-way attack drones against Ukrainian cities. Contemporary Russian attacks can involve hundreds of unmanned aircraft mixed with missiles and decoys. Russia is simultaneously transforming her defence-industrial system around mass production of FPV drones, autonomous systems and inexpensive electronic warfare.

These attacks may look like swarms. Generally they are not. Colonel Ivan Pavlenko, head of electronic and cyber warfare for the Armed Forces of Ukraine, made precisely this distinction in July 2026. Russia can launch hundreds of drones but, in his assessment, she has not yet demonstrated genuine autonomous swarming in combat. Russian aircraft arriving separately over long periods actually provide Ukrainian defenders with additional opportunities to intercept them.

This illustrates the difference between mass and swarm intelligence. Imagine 500 people individually ordered to run towards a railway station. That is a crowd. Now imagine those people continuously exchanging information, changing routes when streets become blocked, dividing themselves between entrances, allocating different tasks and automatically replacing members who fall behind. That is closer to a swarm. The distinction will become extraordinarily important.

Ukraine’s laboratory

As with so many aspects of contemporary drone warfare, Ukraine has become the world’s most important experimental laboratory. Yet even here genuine swarming remains limited. A detailed 2026 assessment of Ukrainian military technology concluded that Ukraine had experimented with elements of swarm-like coordination but battlefield employment remained predominantly confined to small groups rather than large autonomous swarms. The more immediate revolution is occurring one level beneath the swarm: individual autonomy.

Electronic warfare has created powerful incentives to remove the radio link between drone and pilot. If a Russian jammer interrupts communications, an ordinary FPV drone may become useless. A machine equipped with visual navigation and terminal guidance can continue towards its target after communications disappear. Ukraine is increasingly deploying precisely these technologies. Computer vision can identify and track objects, navigation algorithms can operate without continuous GPS and terminal guidance can carry an aircraft through the final stage of an attack even where communications are degraded.

This summer Brave1 announced that a Ukrainian interceptor system had automated approximately 95 per cent of the process of intercepting Shahed-type drones. Radar information directs the interceptor towards its target, artificial intelligence identifies and locks onto the incoming aircraft and the human operator retains the ability to abort the engagement. The system has already undergone combat testing.

That development may ultimately prove more important than spectacular demonstrations involving dozens of drones flying together. Before machines can cooperate intelligently, each machine must first become independently competent.

From pilot to shepherd

The most important conceptual change is therefore occurring in the relationship between human beings and machines. Contemporary drone warfare remains astonishingly labour intensive. An FPV drone costing perhaps several hundred dollars may require a pilot, a navigator, communications specialists, reconnaissance personnel and supporting soldiers. Scaling the number of drones consequently means scaling the number of people operating them.

Eventually this becomes impossible. A million drones are not especially useful if an army requires a million trained pilots. Hence the emerging concept sometimes described as single-operator, multi-platform control. RUSI has argued that much of what is presently described as swarming is better understood in these terms: one human supervising numerous autonomous or semi-autonomous machines rather than individually flying each one.

The pilot thereby becomes a shepherd. He does not tell every sheep where to put its feet. He tells the flock where to go. This seemingly modest distinction has enormous military implications because it removes one of the principal constraints upon drone warfare: human attention.

The mathematics of saturation

Consider an air-defence battery capable of engaging twenty targets simultaneously. An attacker sends thirty drones. The defender must prioritise.

Now suppose the drones themselves understand what is happening. They exchange information about which aircraft have been engaged. Some climb. Others descend. Some divert towards radar installations. Others deliberately present themselves as conspicuous targets. Reconnaissance drones identify active defensive systems and transmit their locations to strike drones. When one route becomes dangerous, the surviving aircraft redistribute themselves.

The defender is no longer fighting thirty individual weapons. He is fighting a distributed adaptive system. This represents a fundamental change because modern military forces are built around relatively small numbers of extremely expensive machines. A Patriot battery, fighter aircraft, radar installation, frigate or command post embodies enormous financial and technological investment.

A swarm reverses the economic equation. Its constituent aircraft may be expendable. Indeed losses may be incorporated into the algorithm. The swarm does not ask whether an individual drone survives. It asks whether the collective objective is achieved. That is how ants behave. It is also how future machines may fight.

Software rather than aircraft

Another consequence follows. The decisive military technology of the swarm may not be the drone at all. It may be the operating system. In July 2026 Ukrainian manufacturer SkyFall and autonomy company Auterion announced a programme involving 50,000 Shrike drones equipped with autonomous terminal-guidance technology. Particularly interesting is the architecture: according to the manufacturers, the same hardware is intended eventually to acquire swarming capability through software updates rather than replacement of the aircraft.

This suggests a future in which military capability evolves rather like smartphones. The physical machine remains substantially unchanged while its behaviour is transformed by software.

A reconnaissance drone might receive an update permitting collaborative mapping. Another update could permit several aircraft to relay communications between themselves. Another might allow automatic redistribution of targets. Another could introduce collective responses to electronic warfare. Eventually the difference between fifty independent drones and a fifty-drone swarm may be little more than code. That should profoundly alter military procurement.

Armies traditionally purchase platforms: tanks, aircraft, ships and missiles. Increasingly they will purchase ecosystems.

The communications problem

Nevertheless enormous technical obstacles remain. The first is communication. A flock needs information. Yet radio communications reveal positions, consume bandwidth and are vulnerable to jamming. Hundreds of drones transmitting continuously could produce an electronic cacophony.

One solution is a mesh network in which every aircraft communicates principally with nearby neighbours rather than with a central command station. Information then propagates across the swarm. Destroy one aircraft and the network routes around it. Another solution is greater independence. The less each drone needs to communicate, the harder the swarm becomes to disrupt.

This creates an intriguing paradox. The most resilient swarm may be one whose members communicate surprisingly little. Again nature provides the analogy. A starling does not need to know what every other bird is doing. It needs information principally about the handful flying around it. Military swarms may ultimately operate according to similar principles.

Swarms against swarms

The next development is likely to be defensive. Ukraine faces a brutal arithmetic problem. Shooting down inexpensive Russian drones with expensive missiles is economically unsustainable. Hence the extraordinary growth of interceptor drones.

Ukraine’s Ministry of Defence and Brave1 are already building AI training infrastructure specifically intended to accelerate autonomous detection and interception of aerial threats. The Brave1 Dataroom contains visual and thermal datasets derived from actual combat imagery, allowing Ukrainian companies to train systems to recognise and intercept hostile aircraft.

The logical destination is an autonomous defensive flock. Imagine several hundred inexpensive interceptors stationed around Kyiv. Radar detects an incoming attack.

The defensive swarm launches automatically. Software allocates targets. Some aircraft intercept Shaheds. Others investigate objects whose identities remain uncertain. Drones running short of battery return to charging stations. Replacements launch. The system continuously recalculates the geometry of the engagement.

Human beings supervise the battle rather than flying each interceptor. The consequences could be revolutionary. The future equivalent of an anti-aircraft battery might not resemble a battery at all. It might be a shipping container containing hundreds of inexpensive aircraft, automated launch equipment, charging systems and computers.

The land battlefield

The same principle will spread across the front. Reconnaissance drones will search ahead. Strike drones will attack detected positions. Electronic-warfare drones will interfere with enemy communications. Relay aircraft will maintain the network. Larger drones will carry smaller ones. Ground robots may transport replacement aircraft and batteries.

The swarm will therefore cease to mean identical quadcopters moving together.

It will become heterogeneous. A biological ecosystem rather than a formation of aircraft. Some machines will see. Some will communicate. Some will jam. Some will carry weapons. Some will sacrifice themselves. And because the system is distributed, destroying one component will achieve very little.

The sea

At sea the consequences may be still greater. Ukraine has already demonstrated how inexpensive unmanned surface vessels can threaten vastly more expensive warships. The next stage is coordination between domains. A surface drone might carry aerial drones towards a coastline. Shortly before approaching the target area it releases them. The aircraft search for radar installations while the surface vessels approach from different directions. Reconnaissance drones identify defensive positions. Electronic-warfare aircraft interfere with communications. Strike drones attack exposed sensors. Only then do the maritime drones approach their principal targets.

No individual machine understands the entire battle. The network does. This may make traditional surface fleets extraordinarily vulnerable in confined waters.

The disappearing headquarters

Perhaps the most radical implication concerns command itself. Armies traditionally function hierarchically. A commander issues orders to subordinate commanders who transmit instructions further down the chain. Swarms are inherently decentralised. If the system requires continuous instructions from headquarters, destroying headquarters destroys the swarm. A mature autonomous system must therefore continue functioning after communications disappear. Mission command becomes machine command. The human gives an objective: Observe this area. Protect this installation. Prevent hostile vehicles crossing this line. The machines determine how to accomplish it.

That raises obvious ethical and legal questions when weapons are involved, particularly where algorithms classify human beings or objects as targets. It also creates one of the central dilemmas of future warfare: the military advantage of autonomy increases precisely as human supervision diminishes.

The swarm of the 2030s

Predictions about military technology should always be treated cautiously. War has a habit of humiliating futurists. Nevertheless the direction of travel is becoming reasonably clear. By the early 2030s we should expect considerably larger groups of autonomous aircraft operating under supervision by dramatically fewer human operators. They will navigate without reliable GPS, communicate through adaptive networks, recognise objects visually and redistribute tasks when individual aircraft are destroyed.

Mixed swarms will probably become more important than homogeneous ones. Aircraft, ground vehicles and maritime drones will cooperate. Some will be reusable. Others will be expendable. Artificial intelligence will increasingly determine routes, formations, sensor allocation and responses to threats. The United States has herself assessed that autonomous drone swarms are highly likely to produce a revolution in military affairs before 2035. The precise date hardly matters. The underlying economics make the transition extremely difficult to avoid.

Industrial warfare returns

The swarm also represents the return of something the West had partly forgotten: mass.

For thirty years Western military doctrine emphasised exquisitely capable platforms. Aircraft became progressively more sophisticated, missiles more accurate and sensors more expensive. Ukraine has demonstrated another possibility. Quantity possesses a quality of its own when quantity becomes intelligent. One $100 million aircraft is an extraordinary machine. Ten thousand $10,000 autonomous machines constitute an entirely different kind of military problem. The purpose of the swarm is not necessarily to make every drone brilliant. It is to make the collective behaviour brilliant while individual machines remain cheap enough to lose. This changes the economics of war.

The final threshold

We have not yet reached the age of the autonomous robotic swarm. Despite spectacular headlines, nobody is presently releasing tens of thousands of independently cooperating armed machines onto the battlefield and allowing them collectively to decide how a battle should unfold. Contemporary swarming remains experimental, limited in scale and dependent upon substantial human supervision. Ukraine’s battlefield experience is demonstrating sophisticated autonomy faster than it is demonstrating genuine swarm intelligence.

But almost all the constituent technologies now exist. Machine vision exists. Autonomous navigation exists. Terminal guidance exists. Mesh networking exists. Collaborative sensing exists. Single-operator control of multiple platforms exists. Mass-produced inexpensive drones exist. The remaining task is integration.

And integration can proceed extraordinarily quickly because much of it is software. The first tanks of 1916 were slow, unreliable and tactically confused machines. Twenty-three years later armoured formations were determining the fate of European states. The first military aeroplanes were principally observation platforms. Within a generation they could destroy cities.

Drone swarming may presently occupy a comparable historical interval. The machines buzzing above the trenches of Ukraine remain mostly individuals. They are piloted, jammed, lost, replaced and launched again in enormous numbers. But increasingly they can see. Increasingly they can navigate. Increasingly they can decide how to reach their destination. And soon they will talk to one another. At that moment the drone ceases to be merely a weapon. The flock itself becomes the weapon.