Artificial Intelligence Propaganda Factories with Language Models

Artificial intelligence (AI)-powered influence operations can now be executed end-to-end on commodity hardware. We show that small language models produce coherent, persona-driven political messaging and can be evaluated automatically without human raters. We evaluate the building blocks of an influence architecture by deploying personas across small language models to generate responses to real-world discussion threads, with a locally run language model serving as evaluating judge to assess persona fidelity and ideological adherence. Two behavioural findings emerge. First, persona-over-model: persona design explains behaviour more than model identity. Second, engagement as a stressor: when replies must provide counter-arguments, ideological adherence strengthens, and the prevalence of extreme content increases. We demonstratethat fully automated influence-content production is within reach of both large and small actors. Consequently, defence should shift from restricting model access towards conversation-centric detection and disruption of campaigns and coordination infrastructure. Paradoxically, the very consistency that enables these operations also provides a detection signature.