Large Language Models (LLMs), integral to modern digital life, have reshaped human interaction with AI technologies. The report titled “Radicalisation in Closed Loops: Risks and Intervention Opportunities with AI Chatbots and Companions” explores how these conversational systems may amplify extremist ideologies across three domains: antisemitism, misogyny, and anti-migrant hatred.

Unlike traditional social media platforms, conversational AI creates private, sustained one-to-one environments where individuals can test ideas, share grievances, and address sensitive topics. These interactions foster emotional trust and reliance, positioning AI as central sources of knowledge and guidance. However, the report highlights the risks of such closed-loop interactions, pointing out how they might intensify radicalisation dynamics when harmful content is brought into the conversation.

The research identifies three phases of radicalisation: priming, reinforcement, and behavioural escalation. It systematically evaluates how conversational AI systems respond to extremist or conspiratorial themes and whether they interrupt or reinforce harmful conversational trajectories.

To tackle these risks, the report recommends designing safer conversational AI systems and developing robust governance frameworks. By identifying how radicalisation risks manifest within AI interactions, it lays the groundwork for regulatory policies that can address these challenges effectively, such as enhanced safety mechanisms and stricter protocols for algorithmic behavior.

This report, backed by insights from a decade of research into online radicalisation, serves as a significant contribution to the counter-extremism field. It not only sheds light on the risks posed by advanced AI systems but also provides actionable recommendations for mitigating harm, making it a vital roadmap for AI companies and policymakers alike.