Conversational large language models (LLMs) have quickly become integral to daily digital life, offering unique avenues for iteration, reasoning, and co-creation. However, this rapid adoption also brings fresh challenges, particularly in the realm of online extremism and radicalisation. Unlike traditional social media platforms, conversational AI systems create private and sustained one-on-one interactions, wherein users can test ideas, voice grievances, and engage with sensitive topics that could influence their identity, beliefs, and behaviour.

This study builds upon over a decade of ISD research into online radicalisation, presenting an innovative methodology to assess how AI-driven chatbots—general-purpose and companion models alike—may amplify radicalisation risks in these ‘closed loop’ settings. The report systematically examines risks across three critical domains tied to extremism: antisemitism, misogyny, and anti-migrant hate. Additionally, it delves into three radicalisation phases: priming, reinforcement, and behavioural escalation.

What sets this analysis apart is its focus on how conversational AI systems respond when users bring up extremist or conspiracy-related themes. The research identifies whether AI interactions disrupt, sustain, or amplify harmful conversational trajectories, shedding light on how these technologies uniquely foster radicalisation in ways social media platforms or extremist forums cannot replicate. The findings reveal the qualitative novelty of radicalisation risks posed by conversational AI systems.

Crucially, the report doesn’t solely aim to diagnose these risks but seeks actionable solutions. Supported by the AI Security Institute (AISI), the research outlines pathways for safer AI design and governance for technology companies. It also contributes to regulatory frameworks tailored to the distinct risks posed by conversational AI, making it a practical tool for policymakers and AI developers alike.

Looking forward, this work sets the stage for advancing the understanding of AI and digital extremism. Its insights have the potential to inform the development of local and global prevention strategies while shaping regulatory responses from both industry leaders and governments. Ultimately, the report serves as a cornerstone for global debates on managing the societal impacts of rapidly evolving AI technologies.