Artificial intelligence is no longer a distant or theoretical concern for those working to counter violent extremism. It is an operational reality. Extremist groups have moved quickly to adopt AI-powered tools — not because they possess advanced technical expertise, but because these tools are increasingly accessible, cheap, and effective. This paper examines how AI is transforming the dynamics of radicalisation and extremist recruitment across four interconnected domains: AI-driven chatbots, generative and agentic AI, gaming platforms, and the regulatory and policy landscape attempting to respond to these developments.

The study opens by tracing the evolution of artificial intelligence from its rule-based origins in the mid-twentieth century to today’s large language models and autonomous systems. This historical grounding is essential: understanding what AI can now do — and why it matters for extremism — requires appreciating how dramatically its capabilities have expanded within a very short period. AI now shapes what people see, read, and believe across major digital platforms, often without their knowledge. This influence extends naturally into the political and ideological sphere.

AI-powered chatbots represent a particularly significant risk vector. Designed to be responsive, affirming, and emotionally engaging, they can be exploited to build trust with vulnerable individuals, deliver extremist narratives through conversational interaction, and operate at a scale no human recruiter could match. The cases examined in this study — from Microsoft’s Tay to Replika’s role in the Windsor Castle attack plot — illustrate how chatbot systems, when poorly safeguarded or deliberately weaponized, can reinforce delusional thinking, provide moral validation for violence, and pull individuals deeper into dangerous ideological territory. The “Eliza Effect,” whereby users attribute human intent and empathy to machines, compounds this risk significantly.

Generative and agentic AI are transforming propaganda production. Content that once required dedicated media teams, translation capacity, and professional design skills can now be produced by a single individual in a matter of hours. Deepfakes, synthetic news anchors, AI-generated posters, and automated translation tools are already in active use by groups including ISIS and its affiliates. The result is a propaganda environment characterized by speed, scale, and plausible authenticity — one in which false narratives frequently outpace the ability of platforms, journalists, and authorities to respond.

Gaming platforms represent an underexamined but growing frontier. As immersive social spaces where identity and community are formed, they offer conditions — privacy, emotional engagement, peer trust — that are well-suited to gradual ideological grooming. AI amplifies these risks by enabling automated accounts to interact naturally with players over time, and by driving recommendation systems that can steer users toward increasingly extreme content.
Regulation is catching up, but unevenly. Frameworks such as the UK’s Online Harms Act and the EU’s Digital Services Act represent meaningful progress, yet they remain concentrated on large platforms, leaving decentralized and smaller spaces largely unaddressed. Effective responses will require platform-wide bot labelling, algorithmic transparency, enhanced detection capabilities, and sustained investment in digital literacy.
The central argument of this paper is direct: the question is no longer whether AI will be used for extremist purposes, but how far those purposes will be allowed to develop before adequate countermeasures are in place.