An evolutionary game model for strategic interactions in AI-driven crime systems

• A three-player evolutionary game models AI use in crime, enforcement, and regulatory authority. • Replicator dynamics reveal stable equilibria for AI adoption patterns. • MARL simulations confirm analytical equilibria and enhance prediction. • Findings guide the design of AI regulation and enforcement strategies. We develop a three-player evolutionary model to examine the strategic interactions of government, law-enforcement agencies, and criminal actors in artificial intelligence (AI) enhanced crime systems. Each group adapts its strategy over time: the government chooses between regulating AI or promoting technological innovation; investigators select AI-assisted or traditional detection methods; and criminals opt for AI-enabled or conventional approaches. Using replicator dynamics, we characterize equilibrium strategies and assess their stability. Numerical simulations illustrate the resulting evolutionary trajectories, identifying conditions under which AI-assisted enforcement is effective and those that enable the spread of AI-enhanced criminal activity. To strengthen predictive accuracy, we incorporate multi-agent reinforcement learning with a reward-penalty mechanism, showing that learned behavioral patterns closely align with the analytical stability results. The findings offer theoretical and practical insights into adaptive behavior in justice systems and support the design of effective AI crime prevention policies.