Artificial intelligence for intelligence analysis

Preventing and countering violent extremism (P/CVE) research has long relied on profiling approaches that draw on demographic variables, ideological labels, and observable late-stage behaviours. These strategies have consistently performed poorly, and their limitations are further exposed in contemporary digital environments, where ideological identities are fluid, most individuals who engage with extremist content never mobilise, and a highly fragmented ideological landscape frustrates stable categorisation. In parallel, psychological research has identified mechanisms that help explain why only a minority escalate to violence, yet these mechanisms remain difficult to operationalise and test at scale. This Perspective argues that recent advances in artificial intelligence—broadly defined to include statistical learning, generative modelling, and decision-oriented optimisation—provide tools to close this operational gap when explicitly aligned with psychological theory. At the individual level, machine learning, natural language processing, and large language models enable measurement and forecasting from heterogeneous digital traces. At the interpersonal level, graph-based approaches may capture influence dynamics, exposure pathways, and the evolution of extremist social milieus. At the collective level, agent-based simulations and field experiments support explanatory and counterfactual analysis of mobilisation processes. We advance a hybrid research agenda that prioritises theory testing, mechanism evaluation, and carefully bounded intervention analysis over automated individualised profiling, advancing a more mechanistic, empirically grounded, and scalable science of P/CVE.