AI-Amplified Endogenous Instability: Structural Limits of Stabilization in Macroprudential Governance

This paper argues that artificial intelligence does not merely expose weaknesses in macroprudential governance — it renders three of its foundational assumptions structurally untenable. Treating AI as a co-constitutive element of the financial system's amplification architecture rather than an external analytical tool, the paper identifies three limits built into the constitutive logic of stabilization frameworks: the endogeneity boundary, the temporal boundary, and the reflexivity boundary. These are not correctable design flaws but necessary features of how macroprudential governance is architecturally constituted. Against this diagnosis, the paper proposes four limit-aware governance principles — Algorithmic Diversity Mandate, AI-Adjusted Countercyclical Buffer, Pre-Committed Synchronization Response, and Reflexivity Firewall — that operate within acknowledged structural boundaries rather than claiming to transcend them. The paper's core contribution is conceptual: macroprudential doctrine must shift from the promise of stabilization to the practice of resilience within structural limits.