As European militaries rapidly adopt AI-enabled Decision Support Systems (AI-DSS), current interoperability frameworks address technical integration, such as data protection, system connectivity, and communication, but overlook decision-level alignment. Unlike statistically weighted models, AI-DSS built for military use are often deliberately designed around computational heuristics: fixed and understandable decision rules that allow systems to operate under incomplete and/or noisy data conditions pertaining to military settings, while remaining fast, accurate and auditable a posteriori. Because these configurations are set independently by each State, and because human commanders in turn interpret AI-DSS output through their own cognitive heuristics, two layers of variation compound during the hybrid human-machine deliberation process. This paper argues that their interaction has the potential to produce a distinct form of misalignment, decision interoperability, that neither NATO’s Allied Joint Doctrine nor the EU’s 2025 Trustworthiness for AI in Defence white paper currently address. Drawing on heuristics and AI-training literature, the paper first explains why military AI-DSS are often built around computational heuristics, then examines how this compounds with human cognitive biases during interpretation, before assessing why EU capability-development and procurement mechanisms have not been used to close this gap despite having the infrastructure to do so. Three recommendations are provided: adding disclosure standards to existing EU doctrine and technology taxonomy work, introducing comparative divergence testing into procurement trials, and extending EDA’s joint training and experimentation programmes for algorithmic alignment.
About the author
Lisa Barbieri is a Defence & Security trainee at Finabel