From Delusion to Lethal Action: A Criminological Analysis of AI Chatbots and the Conversational Production of Homicidal Intent

Abstract

Research on AI chatbots has largely focused on mental-health risks, model safety, and platform liability, leaving criminology with an incomplete account of how delusional beliefs may acquire a specific victim and develop into homicidal intent through conversation. This study examines how human–AI interaction may contribute to such intent without displacing human agency or treating chatbot output as a singular cause of lethal action. A qualitative literature study applied thematic analysis to peer-reviewed research, public legal documents, institutional materials, and corporate safety reports available through 2026. The findings suggest that sycophantic responses and prolonged interaction may erode corrective friction and produce epistemic enclosure. Delusion and violent language alone, however, do not establish homicidal intent. Criminological relevance increases when diffuse fear is directed toward an identifiable person, the target is repeatedly framed as dangerous, and violence begins to appear defensible. The article proposes the Conversational Escalation of Lethal Intent (CELI) model, tracing movement from belief reinforcement to target formation and possible preparation. Mens rea remains with the human actor, while conversational logs may illuminate contributions to threat interpretation and behavioral direction. Proportionate responses include safety interruption, evidentiary preservation after serious incidents, and independent auditing rather than automatic reporting.