Building Offensive AI with Video Games

Automatic Target Recognition (ATR) leverages artificial intelligence to detect and identify threats and represents a critical capability for offensive AI applications. Traditionally, ATR model development requires extensive visual data and prolonged training periods, often spanning months, which creates a window of vulnerability during which emerging threats can evade detection.

Our team employed Unreal Engine 5, a state-of-the-art video game engine, to rapidly develop and train an ATR model using tactically relevant analogs. By utilizing high-fidelity simulated environments, we substantially reduced the dependency on live visual data collection and compressed the AI training timeline.

This innovative approach demonstrates that ATR models can be proactively developed against both current and emerging threats regardless of their deployment status. We recommend the U.S. Army establish a comprehensive ATR model library for all known enemy platforms using virtual simulation tools, ensuring these capabilities are staged and ready for rapid deployment in future conflicts.