The War That Won’t Wait for Us

By Carlin McKeahow

Picture a command post at two in the morning. The coffee has been sitting on the burner too long. Somebody has half a sandwich beside a keyboard, and nobody remembers whose it is. On a large screen, a cluster of enemy drones appears at the edge of defended airspace. The operators know they’re there because software found them first.

The defensive system compares radar returns, infrared signatures and previous attack patterns. It decides that some of the incoming objects are probably decoys. Others look dangerous. Electronic warfare equipment begins interfering with their navigation. Interceptor drones are assigned. The attacking drones change course, shift communications methods and continue toward their targets. The defensive network adjusts again. The people in the room are watching all of it happen, but much of the contest is moving faster than any of them could manage by hand.

That scene is hypothetical. The pieces of it increasingly aren’t.

The war in Ukraine has already pushed drones, electronic warfare, computer vision and varying degrees of autonomy into the same battlespace. Ukrainian developers have built systems that can recognize targets and continue portions of an attack after communications are disrupted. Russia has worked on the same problem from the other direction. Investigators examining Russian attack drones have found increasingly sophisticated cameras, computing hardware and anti-jamming technology. Both sides have learned the same ugly lesson: anything that depends completely on an uninterrupted radio link eventually meets somebody with a jammer.

That lesson points toward where warfare is going.

A military facing heavy electronic warfare has an incentive to build a drone that can keep flying after it loses contact with its operator. A military facing autonomous drones has an incentive to build defenses that recognize and attack them automatically. Once those defenses become faster, the attacker needs machines capable of adapting faster. Nobody has to believe in artificial general intelligence, conscious machines or science-fiction armies for this to happen. Ordinary military competition is enough.

We are heading toward wars where software will increasingly fight software.

The United States is preparing for that environment openly. DARPA’s Artificial Intelligence Reinforcements program is developing AI-driven autonomy for multi-aircraft combat beyond visual range. In July 2026, DARPA and the Air Force announced that a modified F-16 was conducting in-air testing with an AI agent controlling the aircraft, part of work intended to support future teams of crewed and uncrewed aircraft. NATO’s 2026 digital strategy calls for wider AI use, predictive analysis, tactical-edge computing and tighter integration between sensors and weapons. The current U.S. military AI strategy goes further still, directing the department to become an “AI-first” fighting force.

None of this means an algorithm is about to be promoted to general.

It means the distance between finding something and doing something about it is shrinking.

For most of military history, that distance contained people. A scout saw movement and reported it. Somebody interpreted the report. A commander decided what it meant. An order traveled back down. Artillery crews received coordinates. Aircraft were tasked. Units moved. Every step consumed time, and every person in the chain could delay, misunderstand, question or change what came next.

Modern sensors have already compressed much of that process. AI can compress it further. A surveillance system can scan more imagery than a room full of analysts. A defensive system can track more incoming objects than a person can follow. Software can compare thousands of possibilities, assign weapons, reroute autonomous vehicles and react to electronic interference while the human operator is still figuring out what changed on the screen.

In combat, that speed is enormously attractive. Sometimes speed is survival.

If ten incoming drones are three minutes from a ship, nobody wants a committee meeting. If a swarm contains hundreds of objects, manual control may become impossible. If an enemy radar appears for twenty seconds and disappears, a force that needs five minutes to process the information may never get another chance. Military commanders will reasonably ask machines to handle more of those decisions because machines can operate at the pace of the threat.

Then the enemy does the same thing.

That’s where the problem becomes larger than autonomous weapons themselves.

Imagine two forces whose defensive and offensive systems are constantly testing one another. One side changes frequencies. The other recognizes the change and retunes its jammer. The first system identifies the interference and alters its communications pattern. An autonomous reconnaissance aircraft notices a radar going silent and predicts where the operator may have relocated. The opposing network recognizes the search pattern and feeds it deceptive signatures. Somewhere else, cyber systems are trying to find vulnerabilities in the networks coordinating all of it.

Each adaptation creates pressure for another adaptation. Human reaction time starts looking less like a safeguard and more like a tactical disadvantage.

That pressure won’t remain confined to drones. Aircraft, ships, missile defense, electronic warfare, intelligence analysis and cyber operations all reward speed. Military planners have spent generations trying to shorten the time between observation and action. AI offers a way to cut that time brutally short.

There is an uncomfortable consequence buried inside that achievement. Eventually the person who is supposed to be making the decision may receive less time to make it because the opposing machine has already acted.

That changes command.

The language used around military autonomy often reassures us that a person will remain somewhere in the process. Current U.S. policy requires autonomous and semi-autonomous weapons to allow commanders and operators to exercise appropriate levels of human judgment over the use of force. A June 2026 presidential memorandum on national-security AI also insists that commanders remain responsible and that AI systems respect the constitutional chain of command. Those are sensible requirements.

The harder question is what meaningful control looks like when events are moving at machine speed.

A commander who approves a system before a mission may technically have authority over it. That doesn’t necessarily mean the commander understands every decision the system will make after contact with the enemy. An operator may have an abort button, yet the opportunity to use it could last only seconds. A person might be presented with a recommendation that took an AI half a second to generate from thousands of sensor inputs the person can’t independently review.

The signature on the authorization form can remain human while the substance of the decision moves elsewhere.

Military institutions need to take that distinction seriously because bureaucracies are very good at preserving the appearance of responsibility after responsibility has been diluted. If an autonomous system makes a disastrous identification, there will be plenty of places to spread the blame. The commander approved the mission. The operator followed procedure. The intelligence feed met standards. The software passed testing. The model performed within its documented parameters. The contractor built what the government ordered.

Everyone can be partly responsible until nobody seems fully responsible.

War doesn’t become morally clean because the decision traveled through a processor.

There is another problem that soldiers understand instinctively: the enemy gets a vote. AI systems trained to recognize vehicles, terrain, radio patterns or human behavior will be used against opponents who know they are being watched by machines. Those opponents will deliberately try to fool them.

Camouflage will be designed for computer vision. Electronic signatures will be manufactured to trigger false conclusions. Decoys will become increasingly sophisticated. Training data will be studied, poisoned or rendered obsolete. Commanders will learn which behaviors cause an opposing system to classify something as threatening and may attempt to manufacture those behaviors elsewhere.

Deception is ancient. The difference is that the target of the deception may increasingly be a model.

That creates a strange battlefield where one military’s AI isn’t merely searching for enemy soldiers and equipment. It is searching through an environment the enemy is deliberately constructing to manipulate the way the AI interprets reality.

The side with the largest model may not win that contest. The side that learns fastest might.

Ukraine has shown why battlefield data matters so much. Reuters reported in late 2024 that a Ukrainian system aggregating feeds from thousands of drone crews had accumulated roughly two million hours of battlefield video. That material can be used to train systems to recognize what war actually looks like through a drone camera, including damaged vehicles, camouflage, strange angles, bad weather and all the clutter that rarely appears in a clean laboratory dataset.

Future military power will therefore depend partly on an industrial cycle that looks unfamiliar beside traditional measures of combat strength. Armies will still need ammunition, fuel, aircraft, ships and trained people. They will also need enormous quantities of trustworthy data, secure computing, engineers who can update models quickly, communications that survive attack and manufacturing systems capable of replacing autonomous platforms by the thousands.

Software updates may become battlefield events.

That should change how we think about an arms race. During the Cold War, much of strategic competition could be counted. Bombers, missiles, submarines and warheads were visible enough that governments could at least attempt to measure the balance. An AI arms race is harder to see. Two countries might own nearly identical drones while one has dramatically better software, targeting data and electronic-warfare adaptation. A capability could improve substantially without a new aircraft rolling out of a factory.

The balance of power could change with code.

There is a darker issue beyond battlefield effectiveness. Machine-speed warfare can compress political decision time too.

Suppose an AI-enabled warning system identifies what appears to be a large incoming attack. Defensive systems begin reacting automatically. The adversary’s surveillance network sees those defensive preparations and interprets them as evidence of offensive intent. Its systems raise readiness. The first side detects that change. What began as a questionable sensor interpretation is now producing real military movement on both sides.

Machines don’t need hatred, fear or ambition for escalation to occur. They only need to react to each other according to rules that made sense when somebody wrote them.

Humans can behave irrationally in a crisis, but we also possess something machines don’t reliably provide: the ability to look at a technically correct response and decide that carrying it out would be insane.

That ability becomes more valuable as warfare accelerates.

There are areas where automation is clearly worth using. A defensive system intercepting incoming drones may save lives precisely because it reacts faster than a person. Autonomous navigation can keep an aircraft functioning after communications disappear. AI can help sift intelligence, recognize patterns and reduce the crushing amount of information pushed toward commanders. Rejecting those capabilities wholesale would leave militaries slower and more vulnerable.

The standard should be harder than simply asking whether a person remains somewhere on an organizational chart.

Commanders need to know what decisions a system is authorized to make, what evidence it uses, where it fails and how quickly it can be stopped when circumstances move outside its assumptions. Testing has to include deception and chaos, because the enemy won’t cooperate with laboratory conditions. Rules governing autonomous action have to survive contact with exhausted operators, damaged networks and the pressure to act before the other side does.

Most of all, someone must still own the consequences.

This is especially important as AI moves closer to strategic decision-making. The United States has maintained a human decision requirement around nuclear employment, and senior U.S. Strategic Command leadership has said AI may assist with processing information while people retain control. That boundary deserves protection because some decisions are too consequential to reduce to which system completed its calculations first.

Go back to that command post at two in the morning. The attacking drones are still coming. The defensive software is still working faster than the captain ever could. It may be doing exactly what we want a good tool to do: buying time, protecting people and handling a problem whose scale has exceeded human hands.

The captain’s job has changed, though. He isn’t steering every interceptor. He is responsible for understanding what authority has been handed away, where the boundaries sit, when the system should be stopped and what happens when its confidence turns out to be wrong. His commanders carry the same burden farther up the chain.

That is probably what AI warfare will look like for a long time. People will continue giving orders, setting objectives and accepting responsibility while machines handle more of the fighting underneath them. Eventually, two militaries may engage each other through layers of autonomous systems reacting to other autonomous systems so quickly that the human participants experience much of the battle as something they supervise rather than directly conduct.

We can live with machines moving faster than we do. We already do that in countless parts of modern life.

What we can’t afford is to confuse speed with command.

A military that forgets the difference may someday discover that its people still carry all the responsibility for a war they no longer have enough time to control.

Sources

Ukraine’s Future Vision and Current Capabilities for Waging AI-Enabled Autonomous Warfare, by Kateryna Bondar (Center for Strategic and International Studies).
https://www.csis.org/analysis/ukraines-future-vision-and-current-capabilities-waging-ai-enabled-autonomous-warfare

DARPA, U.S. Air Force Fly AI-Controlled F-16, by Defense Advanced Research Projects Agency (DARPA).
https://www.darpa.mil/news/2026/darpa-us-air-force-fly-ai-controlled-f-16

Alliance Digital Strategy, by NATO (North Atlantic Treaty Organization).
https://www.nato.int/en/about-us/official-texts-and-resources/official-texts/2026/01/13/alliance-digital-strategy

Artificial Intelligence Strategy for the Department of War, by U.S. Department of War.
https://media.defense.gov/2026/Jan/12/2003855671/-1/-1/0/ARTIFICIAL-INTELLIGENCE-STRATEGY-FOR-THE-DEPARTMENT-OF-WAR.PDF

National Security Presidential Memorandum/NSPM-11, by The White House.
https://www.whitehouse.gov/presidential-actions/2026/06/national-security-presidential-memorandum-nspm-11/

DoD Announces Update to DoD Directive 3000.09, “Autonomy in Weapon Systems,” by U.S. Department of Defense.
https://www.defense.gov/News/Releases/Release/Article/3278076/dod-announces-update-to-dod-directive-300009-autonomy-in-weapon-systems/

Drone Debris Found in Ukraine Indicates Russia Is Using New Technology From Iran, by Emma Burrows (The Associated Press).
https://www.ap.org/news-highlights/spotlights/2025/drone-debris-found-in-ukraine-indicates-russia-is-using-new-technology-from-iran/

Ukraine Collects Vast War Data Trove to Train AI Models, by Max Hunder (Reuters).
https://www.investing.com/news/world-news/ukraine-collects-vast-war-data-trove-to-train-ai-models-3783672

Stratcom Commander Discusses Nuclear System Modernization, by Army Maj. Wes Shinego (U.S. Department of Defense).
https://www.defense.gov/News/News-Stories/Article/Article/3973074/stratcom-commander-discusses-nuclear-system-modernization/

The State of AI in the Department of Defense, by Kathleen Hicks (U.S. Department of Defense).
https://www.war.gov/News/Speeches/Speech/Article/3578046/remarks-by-deputy-secretary-of-defense-kathleen-h-hicks-on-the-state-of-ai-in-t/

_____________________________

Carlin McKeahow is a contributor to The Havok Journal. His work examines artificial intelligence, emerging technology, institutions, and the human consequences of systems that move faster than public understanding.

As the Voice of the Veteran Community, The Havok Journal seeks to publish a variety of perspectives on a number of sensitive subjects. Unless specifically noted otherwise, nothing we publish is an official point of view of The Havok Journal or any part of the U.S. government.