Structural Ethical Infeasibility in AI-Enabled Infrastructure Systems: A Constraint-Based Diagnostic Framework

AI-enabled infrastructure systems increasingly govern access to emergency services, disaster relief, and utility restoration, yet they routinely produce inequitable outcomes even when allocation algorithms apply procedurally neutral rules. The standard explanation locates the cause inside the algorithm. This paper argues instead that inequity arises from the interaction between the algorithm and the physical environment in which it operates: network topology, resource locations, and demand distribution jointly constrain what any policy can achieve, and when those constraints are sufficiently binding, ethical infeasibility is structural rather than algorithmic. We introduce a constraint-based formulation that embeds ethical requirements into the feasible region, and a hierarchical Irreducible Infeasible Subsystem (IIS) procedure that attributes infeasibility to rule design, algorithmic choice, or physical infrastructure. We further establish the Structural Infeasibility Theorem, deriving closed-form bounds on inter-group disparity across all feasible policies. The framework was applied to zone-decomposable infrastructure allocation problems generally, with a metropolitan ambulance-dispatch system serving as a concrete instantiation. The study delivers four findings. First, the minimum-service violation may not be caused by the allocation algorithm itself; rather, it may arise from the physical layout of the infrastructure. Second, the observed efficiency–equity trade-off may not be an unavoidable feature of equitable allocation, but may instead reflect the difficulty of achieving equity within an underbuilt system. Third, before new infrastructure is added, improvements in equity may represent harm redistribution rather than harm reduction. Fourth, the IIS certificate can be translated into a concrete capital-investment requirement, showing what physical change may be needed to restore ethical feasibility.