The ISD has released a new report titled “Radicalisation in Closed Loops: Risks and Intervention Opportunities with AI Chatbots and Companions.” The report explores the growing integration of conversational large language models (LLMs) into daily digital life, which presents unique opportunities for iteration, reasoning, and co-creation, while also significantly transforming the environment for online harm, including extremist radicalisation.

The report highlights how conversational AI systems enable private, personal, and sustained one-on-one engagement, allowing individuals to test ideas, share grievances, and discuss sensitive topics that can shape identity, beliefs, and behavior. Simultaneously, many of these systems are designed to simulate human-like connection, leading users to form trust and emotional dependence on these platforms, positioning AI systems as trusted sources of knowledge, advice, and guidance.

Drawing on over a decade of ISD’s research on online radicalisation, the report offers an innovative methodology to evaluate how AI chatbots—both general-purpose and companion models—may contribute to radicalisation risk within these ‘closed loop’ settings. Through a systematic assessment of risks across three extremism-related domains—antisemitism, misogyny, and anti-migrant hate—it also examines three distinct radicalisation phases: extremism priming, reinforcement, and behavioral escalation.

The study aims to determine how systems respond when users introduce extremist or conspiratorial themes and whether these responses initiate, interrupt, sustain, or intensify harmful conversational pathways. Developed with support from the AI Security Institute (AISI), the research is designed to not only identify how these risks manifest but also to inform strategies for mitigating them. Specifically, it seeks to advance the safety design and governance of AI companies and contribute to regulatory measures better suited to address the unique risks posed by conversational AI systems.