Artificial Intelligence-driven disinformation campaigns and their influence on political violence and extremism

Abstract

Generative artificial intelligence (AI) has made creating propaganda texts, images, audio, and video at scale easier and cheaper, opening new possibilities for political actors, foreign influence operators, extremists, and everyday users to create deceptive narratives. This article is a mixed-methods empirical study that takes a convergent approach to the questions of the associations between AI-produced disinformation and political violence and extremism, and of which interventions mitigate such risks. The study correlates secondary quantitative evidence from the Stanford AI Index, Reuters Institute Digital News Report, Political Deepfakes Incidents database, V-Dem, GDELT, and the Global Terrorism Database documentation of election incidents, as well as other recent peer-reviewed studies, against qualitative data from policy documents, platform-governance studies, and documented election cases. No public global data on political violence are currently harmonised and measured in a single panel with data on AI-generated disinformation, algorithmic amplification, and digital literacy, so only source-supported descriptive statistics and the directionality of the model evidence are provided. The results show that disinformation generated by AI may be considered a risk multiplier: in most cases, it is not sufficient on its own to trigger violence, but it can help create an identity threat, a sense of conspiracy, distrust, and mobilisation. H1: Conditional support improves discernment; e.g., literacy and inoculation interventions reduce some of the structural amplification. H2, H3: Conditional support improved discernment; e.g., inoculation and literacy interventions removed some of the structural amplification.