From Influence to Radicalisation: How Artificial Intelligence Structures Influence and May Intersect with Pathways Toward Violence: A Cognitive-Security Account

Influence operations are as old as politics, but artificial intelligence has changed the conditions under which they run. This paper examines a possible arc from AI-structured influence to radicalisation and the threshold of violent action, using two connected frameworks. The AI Influence Pipeline describes five functions: profiling, generation, amplification, adaptation and persistence. A second model identifies possible intersections between those functions and states associated with radicalisation; it does not assume a fixed sequence or that AI accelerates movement toward violence. The paper positions these frameworks against the computational-persuasion, influence-operations and radicalisation literatures, and locates them within cognitive security, the protection of human perception, judgement and decision-making. The central argument is bounded. AI does not invent new goals of influence, and exposure alone does not produce violence. It can change the feasibility of sustained, personalised influence through low-cost generation, adaptation and conversation, but effects depend on people, system design and context. One capability receives separate attention: the selective disclosure of true information. Controlled experiments suggest that frontier models can persuade through information selection without falsehood, posing a challenge to defences focused on false content. The paper discusses a court-documented AI-companion interaction and the contested information environment surrounding Romania’s 2024 presidential election, distinguishing what those cases establish from what remains unknown. It offers testable propositions and a defensive account organised by pipeline stage, not evidence that any exposure produces a given outcome.