On June 1, 2025, Ukraine launched roughly 100 small drones at Russian airbases spread across several time zones. The platforms had been smuggled close to their targets in wooden containers, then released to strike together. The operation, known as Operation Spider's Web, was reportedly flown by individual pilots, one per drone, making it a striking feat of coordination rather than of machine autonomy. But it points directly to a harder battlefield question. As targeting functions shift from the pilot to the platform and as drones begin to act together with less human steering, who is accountable for what a swarm does?
The question is not rhetorical, and it is not new. It sits at the center of more than a decade of debate over autonomous weapons. The war in Ukraine has moved that debate out of the conference room and toward the battlefield at an unplanned for scale and tempo. Ukrainian and Russian systems are taking on more of the targeting and terminal-guidance work themselves and analysts at IEEE Spectrum have documented the steady move toward onboard autonomy as electronic warfare severs the link between operator and drone. The law of armed conflict still applies in full; the harder problem is that the architecture through which these systems increasingly act does not line up neatly with the way the law assigns responsibility.
The Law Has Not Changed, the Architecture Has
International humanitarian law does not offer a loophole to machines. States remain responsible for internationally wrongful acts, and individuals can still be prosecuted for war crimes. As Vincent Boulanin of SIPRI has put it, autonomous systems do not sit outside the law. The principles of distinction, proportionality, and precaution bind a commander who employs a drone swarm exactly as they bind one who orders an artillery strike. The difficulty is practical rather than doctrinal.
In a 2022 study for SIPRI, Boulanin set out four reasons accountability becomes unusually difficult once weapons act autonomously. The law itself is unsettled on how its rules apply to these systems. The unpredictability of some software compounds older disputes about who is responsible when something goes wrong. The sheer number of actors involved, from coders to commanders, diffuses responsibility across so many hands that no single hand can be held to it. And the conduct of the system itself becomes hard to foresee at the point a commander commits to its use. Only the last of these is, in part, an engineering problem. The first three are problems of governance, and they cannot be solved with better hardware. Together, they describe a single underlying difficulty: responsibility under the law of armed conflict depends on a foreseeable human decision, and the way these systems are now built tends to diffuse that decision rather than to locate it.
Richard Moyes, who leads the disarmament group Article 36, frames the stakes more bluntly. If we do not know how an autonomous decision was reached, where its information came from, or how current that information was, then human decision-making stops being meaningful. The law of armed conflict rests on human moral engagement and on accountability for human choices. Strip out human judgment and you have not just a technical gap; you have removed the thing the law was built to hold onto.
Three Ways a Swarm Can Be Put Together
To see why architecture matters for accountability, it helps to look at how a swarm composes a decision. A swarm is not just one machine. It is many, and the rule by which their individual readings combine into a single action is a design choice. That choice is made by engineers and commanders long before the swarm flies, and it determines where responsibility can later be located.
Consider three patterns apparent in Ukraine. In the first, a small group of drones engages only when each member's own targeting threshold is met. Authority rests at the lowest common denominator, and a commander can point to a clear chain of individual confirmations. Ukrainian small-unit drone teams early in the war, described in the Geneva Centre for Security Policy's lessons-learned work, operated close to this model. Accountability here is relatively easy to trace, because the human confirmations are explicit and recorded.
In the second pattern, a large, distributed operation succeeds when enough of its members reach their targets, not when any one member does. The design accepts the loss of many platforms and treats success as a property of the group. Ukraine's growing use of swarming and coordinated autonomy, documented by the Center for Strategic and International Studies, is moving in this direction as software takes on more of the coordination. Here, the trouble for accountability is that responsibility diffuses across the whole package. No single drone's conduct tells you whether the operation respected the law, and the after-action record must capture a collective behavior rather than a clean sequence of individual human decisions.
In the third pattern, one lead element coordinates and the rest follow. Russian long-range strike packages have often been described in these terms by analysts at the Royal United Services Institute. This concentrates the decision, which makes responsibility easy to locate in principle. But it also concentrates risk. Compromise or mislead the leader, and every follower inherits the error. Here the accountability question becomes whether the human who set the leader's parameters understood what the followers would do once committed.
These are not exotic engineering categories. They are ordinary ways of organizing a group, and each carries a different answer to the question of who decided. The point for the law is that the operational methodology by which a swarm composes its decisions is fixed at design time, not in the heat of the engagement. Where the human decision sits in that methodology is fixed at the same moment. A commander cannot retroactively locate a human decision that the architecture never required.
Why the Audit Trail is Necessary but Not Sufficient
One response to the diffusion-and-foreseeability problem is to demand a complete and tamper-evident record of what the system did and why. This is sound reasoning, and it should be a baseline requirement for any autonomous capability fielded by a State that takes its legal obligations seriously. A record that cannot be altered after the fact is the precondition for any honest investigation.
A well-designed record does more than confirm internal consistency. It can show when the system's sensors agreed and when they did not, the probability the system attached to a given determination, the elements that drove that determination, the timing of each step, and whether the determination shifted back and forth before settling. Those are exactly the inputs an investigator and a legal adviser need to reconstruct how a particular action came about and to test it against the principles of distinction, proportionality, and precaution. A record that captures them is a serious instrument of accountability, not merely a defensive log.
But the record's reach is bounded by what is recorded. It cannot, on its own, tell us whether what the system observed was true. An adversary who patiently corrupts the shared picture, feeding a swarm a slowly distorted sense of its environment over weeks, defeats a system whose internal records remain consistent with one another and whose probability traces look entirely orderly. The war in Ukraine is precisely the kind of long, adaptive contest, in which each side spends months learning to spoof and degrade the other's sensors, where that kind of slow manipulation is plausible. The integrity of the inputs is therefore a second, separate obligation, alongside the consistency of the records, and it is the one more easily forgotten.
What This Means for Commanders and Their Legal Advisors
The 2019 Guiding Principles agreed upon at the Convention on Certain Conventional Weapons affirm that humans must retain responsibility for the use of these systems, because a machine cannot be held to account for a violation of the law. That principle is right, but it is a destination, not a route. Retaining responsibility in practice means building it in before deployment.
Three implications follow. First, the choice of how a swarm composes its decisions is a legal decision as much as an engineering one, and it deserves to be reviewed as part of the weapons review that Article 36 of Additional Protocol I already requires. The reviewer should be able to say, before the system flies, where a human judgment sits in each mode of operation. Second, the record-keeping obligation should extend upstream to the integrity of the data the system relies on, not merely the consistency of its internal logs. Third, a commander's legal adviser needs to understand the composition rule in operational terms, because the question a court or an investigation will eventually ask is not how the software worked but who decided, and based on what.
Conclusion
Ukraine has compressed years of theoretical argument into a live demonstration. The technology is not waiting for the law to catch up, and the law does not need to be rewritten to meet it. What it needs is to be applied earlier, at the point where the architecture of a swarm is chosen, rather than after a strike when the question of responsibility has already become difficult to answer. The accountability gap is not opening because the law is silent. It is opening because the human decision the law depends on is being designed out, quietly, one composition rule at a time.
***
Burak Oktenli holds an MBA and a Master of Professional Studies in Applied Intelligence from Georgetown University. His work focuses on the governance of autonomous systems, the law of armed conflict, and the institutional architecture of accountability for AI-enabled military capabilities.
The views expressed are those of the author, and do not necessarily reflect the official position of the United States Military Academy, Department of the Army, or Department of Defense.
Articles of War is a forum for professionals to share opinions and cultivate ideas. Articles of War does not screen articles to fit a particular editorial agenda, nor endorse or advocate material that is published. Authorship does not indicate affiliation with Articles of War, the Lieber Institute, or the United States Military Academy West Point.
Â
Â
Â
Â
Â
Â
Â
Â
Photo credit: U.S. Army, Army Sgt. Dakota Bradford
