Capturing military artificial intelligence assets is not merely acquiring an object, but acquiring foundation for future military capability. This article explores the legal status and implications of capturing AI in armed conflict, moving beyond traditional concepts of war booty to consider the transfer of military knowledge and the challenges posed by evolving AI software for international humanitarian law.
Capturing Artificial Intelligence
The law of armed conflict allows for the capture and appropriation of enemy military property, traditionally termed 'war booty'. With military artificial intelligence (AI), the value extends beyond hardware to the adaptable software, which represents a foundation for future military capability. Existing laws broadly accommodate the capture of AI systems, treating their hardware and software as military objectives, similar to dismantling a captured radar system or studying intercepted missiles. Historically, states have exploited captured technology to gain military advantages, and AI appears to be the latest iteration of this practice.
Exploitation
Exploiting captured AI presents unique challenges compared to traditional military equipment. Unlike a tank with relatively fixed capabilities, a captured machine-learning model can be extensively retrained, fine-tuned for new environments, or integrated into different weapon systems, significantly increasing its military value. This transforms the act of capture into an acquisition of 'military knowledge' or 'institutional experience' rather than just physical property. Modern AI systems, relying on distributed computing and cloud-hosted models, further complicate this, as the physical platform may only hold a fraction of the system's true capability, with much of the military effect residing in intangible, continuously updated datasets.
Legal Review
The evolving nature of AI systems also raises questions concerning Article 36 of Additional Protocol I, which requires states to review new weapons for legality under international law. If a captured AI model undergoes substantial retraining by the capturing state, it may become a 'new weapon' requiring a fresh legal review. This is distinct from conventional weapons, whose legal characteristics remain stable after manufacture, as AI systems can change significantly throughout their operational lifespan, making initial legal assessments potentially obsolete.
Concluding Thoughts
Artificial intelligence fundamentally shifts the focus of the law governing captured military property from the transfer of ownership of physical capability to the acquisition of military learning and computational experience. The strategic value of capturing AI lies less in obtaining an adversary's hardware and more in inheriting years of computational development to accelerate one's own military advancement. While existing legal frameworks are broad enough to permit many of these activities, they were not designed with continuously evolving software in mind. Therefore, rather than entirely new rules, a re-understanding of what 'capture' entails—from 'war booty' to 'war learning'—is needed as AI becomes central to armed conflict.
