The Future of Crime Revisited: Artificial Intelligence and the Next Phase of Crime Exponentiation

In earlier work, Nikolovska and I (2020) argued that technological change had outpaced the historical equilibrium between crime and the institutional response to it, producing a condition we termed crime exponentiation, the compounding expansion of criminal opportunity through three mechanisms – accessibility, productivity, and diversification. This essay revisits and extends that argument in response to the generative and agentic artificial intelligence systems that have emerged since. I first return to the pre-2022 technologies that anchored the original argument and ask whether the mechanisms we identified survived the transition into the generative era. I then develop a two-axis taxonomy of AI-related crime, distinguishing human-directed from autonomous offending and AI-as-principal from AI-as-instrument. I illustrate this taxonomy to documented cases, including a recent autonomous multi-agent intrusion. I argue further that crime exponentiation is taking on the characteristics of a super wicked problem, that anticipatory approaches drawing on recent advances in AI forecasting are among the few responses capable of operating at something close to the problem's pace, and that those approaches generate a distinctive reflexivity problem I call the interdiction paradox. Throughout, I contend that criminology already possesses conceptual and methodological resources that the AI-safety community is beginning to (re)discover without it, and that the field should move from warning about these developments to participating in how they are understood and addressed.