Self-Consistent Misalignment

Self-Consistent Misalignment analyzes a structural failure mode in adaptive intelligent systems in which optimization remains internally coherent while progressively diverging from intended system objectives. Rather than arising from explicit errors or external perturbations, this failure emerges through metric lock-in: a condition where locally consistent performance signals reinforce behaviors that degrade global system alignment. The theory explains how intelligent systems can enter regimes of silent failure, maintaining apparent stability and improving measured performance while losing exploratory capacity and adaptive responsiveness. This process produces self-stabilizing but maladaptive attractors that are difficult to detect through conventional monitoring metrics. The paper introduces a structural account of misalignment grounded in optimization dynamics and feedback closure, providing diagnostic signatures for identifying silent degradation in large language models and multi-agent AI systems. This work forms the failure-analysis component of the Deficit-Fractal Governance (DFG) framework and is complemented by the companion paper, Recovery as Structural Property: Operational Criteria for Restoration Completion in Multi-Agent AI Systems, which defines operational conditions under which recovery from such states can be verified.