Governance of AI and by AI: Feedback Loops, Regime Variation and Reflexive Polycentric Control

This article develops a conceptual framework that links the governance of and by artificial intelligence (AI) into a single continuum that varies across political regime types. Although existing scholarship typically treats governance of AI—oversight, regulation and ethical alignment—and governance by AI—the embedding of algorithms into decision‐making—as separate domains, this article argues that such a dichotomy obscures their mutual dependence. It introduces the framework of Reflexive Polycentric Control (RPC), which views AI governance as two coupled feedback loops: an outer loop of societal oversight and an inner loop of operational deployment. Drawing on reflexive modernization, cybernetic control theory and polycentric governance, RPC specifies three design principles—distributed authority, requisite variety and reflexive checkpoints—that enable stability and adaptability. The framework further highlights how regime type conditions loop dynamics. Democracies exhibit resilience but fragmentation, autocracies achieve coherence at the cost of brittleness, and transitional regimes face volatility because of inconsistent feedback. RPC thereby provides a logically coherent, testable baseline for analysing AI governance across diverse political contexts and for guiding future empirical research.