From War Booty to War Learning: The Legal Status of Captured Military Artificial Intelligence
The law of armed conflict has long recognised that military victory carries with it certain legal consequences beyond the battlefield. Among them is the ability to capture and exploit enemy military property. Whether described as “war booty” or, as some modern scholars prefer, the lawful appropriation of military property under international humanitarian law (IHL), the practical result has remained remarkably stable. A captured tank may be repaired and redeployed. A seized radar system may be dismantled and reverse engineered. An intercepted missile may be studied to improve domestic capabilities.
For centuries, the law has assumed that military advantage continues after the fighting through the exploitation of captured matériel. Military artificial intelligence challenges that assumption in subtle but significant ways.
Capturing Artificial Intelligence
Recent scholarship has revisited the legal foundations of war booty, questioning whether the concept survives as an independent legal doctrine or whether modern IHL simply permits the seizure of enemy military property as a consequence of military necessity. Whatever view one adopts, the law appears relatively settled on one point: enemy military property that contributes to military operations may generally be captured and appropriated. The 1907 Hague Regulations distinguish military property from protected private property, while Article 53 expressly recognises the seizure of movable State property capable of supporting military operations. The debate is therefore not so much whether military property may be captured, but why the law permits its capture.
Unlike traditional military equipment, artificial intelligence is not simply an object whose value lies in its physical characteristics. An autonomous drone certainly contains hardware, but its military utility increasingly resides in the software that controls it (see Operation Spider’s Web). More importantly, that software is rarely static. Machine-learning models are designed to adapt, improve, and be retrained. The capability embodied within the model today need not resemble the capability it possesses after months of further development. Capturing military artificial intelligence assets is therefore not merely acquiring an object; it is acquiring the foundation upon which future military capability can be built.
At first glance, existing law appears capable of accommodating this development. If an autonomous weapon system is lawfully captured, there seems little reason to distinguish between its hardware and its software. Both constitute military property forming part of a military objective under Article 52 of Additional Protocol I. Just as a State may dismantle a captured radar system to understand its design, it would appear equally entitled to extract, analyse and reverse engineer the artificial intelligence model embedded within an autonomous platform. States have exploited captured technology throughout history. As a Classicist, I consider best example to be the Roman invention of the corvus, which turned the tide in the First Punic War. For modern history, the recovery of encryption devices during the Second World War and the study of captured aircraft and missile systems during the Cold War show there is precedent. Artificial intelligence appears, initially at least, to represent little more than the latest iteration of an established practice.
Exploitation
The more difficult issue is not capture itself, but what follows. Traditional military equipment possesses relatively fixed capabilities. A captured tank can be repaired or upgraded, but it remains recognisably the same platform. Artificial intelligence is different. A captured machine-learning model can be retrained using entirely new operational data, fine-tuned for different environments, or integrated into weapons systems wholly unrelated to the platform from which it was captured. Indeed, the military value of the model may increase exponentially after capture. What begins as software developed by one belligerent may, through further learning, become a substantially different capability serving another.
In this respect, the traditional concept of exploiting captured military property begins to look incomplete. The law has generally assumed that capture transfers ownership of military equipment. Military artificial intelligence raises the possibility that what is really being transferred is military knowledge. The learned parameters within a model represent thousands, perhaps millions, of operational decisions encoded mathematically. They reflect observations, training data, and optimisation accumulated over years of development. Capturing such a model therefore resembles acquiring institutional experience rather than simply acquiring property.
This distinction becomes even more pronounced when contemporary software architecture is considered. Many military artificial intelligence systems rely upon distributed computing environments, cloud-hosted models and continuously updated datasets. The physical platform captured on the battlefield may contain only a fraction of the capability that actually produces military effect. Conversely, the seizure of military servers or command infrastructure may yield access to numerous deployable artificial intelligence models without capturing a single weapon. The traditional law of booty evolved in an era when military capability was inseparable from physical objects. Artificial intelligence increasingly separates capability from matériel altogether.
Legal Review
These developments also expose an underappreciated connection with Article 36 of Additional Protocol I (yes, it rears its head yet again!). States party to the Protocol must review new weapons, means and methods of warfare to determine whether their employment would be prohibited under international law.
If a captured AI model is substantially retrained before entering service with the capturing State, difficult questions arise as to whether that modified model constitutes a new weapon requiring fresh legal review. The original assessment conducted by the adversary may provide little guidance once the model has learned from entirely different datasets or has been integrated into a new operational environment. Unlike conventional weapons, whose legal characteristics remain comparatively stable after manufacture, artificial intelligence systems may change throughout their operational life.
Concluding Thoughts
Perhaps the most interesting implication, however, concerns the future direction of the law itself. For centuries, the law governing captured military property has focused upon the transfer of ownership and the exploitation of physical capability. Artificial intelligence shifts attention from ownership to learning and places the strategic value of capture less in obtaining an adversary’s hardware than in inheriting years of computational experience and using that experience to accelerate one’s own military development. Existing legal rules appear broad enough to permit much of this activity, but they were never drafted with continuously evolving software in mind.
The consequence is that IHL may not require entirely new rules governing captured military artificial intelligence. Rather, it may require a different understanding of what capture actually means. The legal framework governing military property remains largely fit for purpose. What has changed is the nature of the property itself. Captured artificial intelligence assets are not merely another weapon to be turned against their former owner. They are a repository of military learning capable of evolving long after the battlefield has fallen silent. That possibility transforms the law of capture from a doctrine concerned with military equipment into one increasingly concerned with military knowledge. It is that shift—from war booty to war learning—that deserves far greater attention as artificial intelligence becomes an increasingly integral feature of armed conflict.
***
Dr Samuel White is the Scientia Senior Researcher in Military Law and War Studies at UNSW Canberra (based within the Australian Defence Force Academy).
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.
Photo credit: Unsplash
