The Department of War is leveraging mathematical proof to bridge the gap between decades-old simulation code and modern AI-driven decision-making tools.

The Pentagon awarded an $80 million Other Transaction Authority contract to Boston-based startup Code Metal on Thursday, signaling a strategic shift toward mathematically verified artificial intelligence in military planning. This investment aims to modernize the Department of War’s WarMatrix platform by translating legacy simulation code into modern, AI-compatible architectures without compromising the integrity of decades of validated data.

The procurement process concluded in under 12 months, moving from an initial $17 million operating capability grant to the current full-scale award. Peter Morales, CEO of Code Metal, noted the unusual speed of the acquisition, highlighting the urgency with which the Department of War seeks to integrate advanced computational tools into its existing infrastructure.

WarMatrix serves as an orchestration layer for the Advanced Framework for Simulation, Integration, and Modeling, or AFSIM, a complex C++ environment originally developed by Boeing. While AFSIM provides high-fidelity modeling for scenarios ranging from subsurface to space, its steep learning curve has historically limited its accessibility to specialized personnel.

The platform enables planners to execute complex queries regarding force structure and logistics without requiring deep expertise in simulation programming. By adjusting variables such as munitions depletion or base vulnerability, users can compare multiple courses of action in a dynamic, queryable environment that the Air Force aims to accelerate to 10,000 times faster than real time.

Code Metal utilizes a neuro-symbolic architecture to bridge the gap between legacy environments and modern hardware. The neural component handles the translation of code from languages like FORTRAN or MATLAB, while the symbolic component generates a formal mathematical proof to ensure the translated code maintains the behavioral intent of the original.

This verification process mirrors techniques used in high-stakes environments such as the seL4 microkernel and Airbus flight-control software. By proving the absence of runtime errors during translation, the Pentagon avoids the need for a costly and time-consuming re-validation cycle for its established modeling ecosystem.

The platform achieved its first operational milestone during the GE 26 Benchmark Wargame, which concluded on March 27, 2026, in Alexandria, Virginia. Over 150 participants utilized the system to conduct physics-based modeling, providing decision-informative insights to senior Air Force leadership regarding future force design and capability requirements.

Formal verification provides a guarantee that the translated code computes outcomes identical to the original models, yet it does not eliminate the risk of analytical inaccuracies. The system ensures structural fidelity, but it cannot inherently validate the underlying assumptions regarding real-world weapons performance or operational logistics.

Michael Horowitz, director of Perry World House at the University of Pennsylvania, emphasized that while these tools offer significant potential for modernization, they require rigorous oversight. He warned that users must remain vigilant against automation bias, where planners might accept AI-generated outputs without sufficient critical scrutiny.

Elke Schwarz, a professor of political theory at Queen Mary University London, underscored the persistent challenge of AI hallucinations in high-stakes environments. She noted that while verification can minimize technical errors, the risk of plausible-but-false outputs remains a fundamental concern for time-sensitive military decision-making.

The Pentagon now faces the challenge of balancing the speed of AI-enabled simulation with the necessity of human-in-the-loop validation. Future milestones will likely focus on how effectively these tools integrate into broader strategic planning cycles and whether the formal proof architecture can withstand the complexities of evolving operational requirements.

The long-term success of this initiative depends on the ability of the Department of War to maintain a rigorous verification standard as the underlying AI models grow in complexity. If the formal proof architecture proves successful, it could set a new benchmark for how the military adopts emerging technologies while preserving the reliability of its foundational data assets.

At Global Business & Economics Journal, we contribute to the  future of business and technology with news,  innovative research, expert insights, and thought-provoking analysis.