Malicious actors could gain access to sensitive information, although purchasing controls already help reduce this risk. Intelligence teams assess potential customers’ digital infrastructure to identify how data might reach adversaries, hostile governments, or other bad actors.
However, using battlefield information as AI training data creates new tracking and accountability challenges. Commercial datasets can sometimes be traced when planted contact details appear shortly after a sale. By contrast, the origin of military and civilian data can become nearly impossible to identify once it has been absorbed into an artificial intelligence model. This creates another risk: an extractive data economy in which wealthy countries profit from threats faced by frontline states, potentially creating financial incentives for conflict to continue as an endless source of valuable digital information.
A New Frontier With Significant Challenges
Existing international law regulates how militaries conduct armed conflict. However, few rules address what happens when records created during combat are removed from their operational context, converted into datasets, and licensed to technology companies whose products can be distributed worldwide.
The responsibilities of companies developing AI systems trained on battlefield data also remain unclear. Ukraine is advancing access controls outlined in its newly signed UK-Ukraine AI Agreement, but no government is currently taking comprehensive action to regulate what happens when military data is absorbed into AI models and later released into civilian markets.
These datasets represent real human lives. Soldiers and civilians captured in battlefield footage did not consent to becoming training material for commercial products that could be sold years later. Yet sensor data, video recordings, geolocation information, and records of civilians fleeing drone attacks may influence how future autonomous systems identify objects, assess threats, and make decisions.
At the center of the issue is consent. Individuals represented in these datasets—including targets, operators, soldiers, and civilians—can become part of AI training materials without their knowledge or permission. The autonomous capabilities developed from this data can extend beyond the battlefield and be transferred to military and commercial technologies, including delivery vehicles, agricultural machinery, and other automated systems. Any errors, biases, or assumptions embedded in the original data may travel with the model long after it enters civilian life.
Battlefield data should not be treated as ordinary commercial information. Yet no single regulatory body currently has clear jurisdiction over its collection, sale, or reuse. Until stronger rules are established, governments providing access to defense data should manage it like a controlled arms transfer by documenting its origins, licensing approved users, and limiting future redistribution. Ukraine has begun exploring this approach. The Avengers Labs program allows companies to train AI models on battlefield data without receiving direct access to sensitive databases. However, this approach addresses only part of the broader privacy, consent, security, and data governance challenges.
Source: www.technologyreview.com