Policy mix framework for AI regulation: A game-theoretic analysis of risk-based direct regulation and standardization approaches

Abstract

This study addresses the growing challenge of designing effective AI regulation policies amid rapidly evolving technologies and diverse stakeholder needs, conducting a theoretical analysis for improved design of artificial intelligence (AI) regulation policies. Through comparative analysis of the EU's risk-based direct regulation and the UK's self-standardization approach, this research develops a policy optimization model incorporating sector-specific characteristics (risk level α, cost structure β, innovation effect γ). The methodology employs a two-stage sequential game model to analyze interactions between regulatory policy variables (direct regulation intensity r, incentive level s) and AI firms’ strategic choices (compliance level c, innovation investment i). Simulation results suggest that incentive-centered policies (s* = 0.156) are more effective in high-risk/high-cost domains, while mixed regulation-incentive policies (r* = 0.539, s* = 0.413) are preferred in high-risk/low-cost areas. Notably, cost structure (β) emerged as a crucial determinant in policy selection, and the relationship between regulation and incentives shifts between substitutive and complementary depending on risk levels. The analysis demonstrates the necessity of differentiated approaches based on sector-specific characteristics and dynamic policy adjustment mechanisms in AI regulatory framework design. These findings are expected to serve as an empirical foundation for establishing global AI governance frameworks.