Artificial intelligence and online political radicalization: State of the art on algorithmic bias, group dynamics and generative AI

This study reviews recent empirical evidence on the role of artificial intelligence–driven systems in political behavior and online radicalization. Drawing on a systematic review of empirical literature published between 2021 and 2025, the findings reveal consistent patterns in how AI operates within socio-political environments. Methodologically, computational network metrics are integrated with deductive thematic coding to ensure a rigorous macro-and-micro analytical approach. Algorithmic recommendations do not function as independent agents of radicalization; rather, they act as conditional amplifiers shaped by user preferences, identity-based motivations, and prior engagement trajectories, thereby increasing exposure bias. Evidence further indicates that AI systems exert a stronger influence on affective and emotional responses than on long-term ideological change. A second key finding is that recent forms of AI intensify misinformation dynamics not only by amplifying exposure but also by generating a broader crisis of veracity, in which trust in evidence and shared epistemic reference points is progressively eroded. Recent evidence points to an emerging pattern where this shift may represent a qualitative transformation in the nature of sociopolitical risk associated with AI. Overall, the literature remains dominated by a technocentric research paradigm, with limited engagement in regulatory analysis and political theory. Consequently, future research should prioritize longitudinal and interdisciplinary approaches that integrate technical assessment with political theory in order to generate more robust and policy-relevant conclusions about the relationship between AI and political behavior.