Evolving patterns in the Terrorist Threat Landscape: Online Radicalisation in the age of Artificial Intelligence

Le résumé fourni par la source

The terrorist threat landscape has changed significantly in recent years, challenging conventional definitions of terrorism. A growing number of attacks now sit in a ‘grey zone’ between terrorist-type violence, personal grievance, psychological vulnerability, social alienation and fragmented ideology. This does not mean that such violence is apolitical or devoid of meaning. Rather, its extremist content is often articulated through hybrid combinations of slogans, symbols, narratives and grievance frames, many of which are encountered, reinforced or reconfigured in digital environments. This shift is particularly visible in lone-actor violence: individuals involved in such attacks may present a combination of identity crisis, mental health difficulties, social isolation, weak community support and many other concurring variables. These factors can produce forms of violence that are difficult to detect preventively. Online presence can then become more than a source of information or entertainment: it may provide belonging, recognition, validation and access to communities where personal frustrations are reframed through ideological, conspiratorial or violence-oriented narratives. Online radicalisation in this context rarely follows a linear path from exposure to violent action. Instead, it often involves fragmented and highly personalised trajectories, as well as Bottom-up and Top-down processes. An individual can self-radicalise within online platforms and/or be radicalised through external recruitment. Individuals may draw selectively from extremist, conspiratorial, misogynistic, accelerationist, ethno-nationalist, religious, anti-system or post-ideological material, combining separate and sometimes contradictory elements into idiosyncratic justifications for violence. In some cases, the resulting worldview is less a coherent ideology but a personalised cult of violence, shaped through repeated exposure to digital content, peer validation, meme cultures, algorithmic recommendation, closed groups and cross-platform migration. The central question is therefore not whether the internet and social media platforms cause radicalisation, but how they shape the environments in which grievance, identity issues, psychological vulnerability and violence become connected. Digital media have shortened the distance between exposure, interaction and ideological reinforcement. Indeed, understanding social media platforms-specific dynamics is essential for analysing contemporary radicalisation, particularly where traditional ideological labels no longer capture the full complexity of the threat. Also, the advent of Artificial Intelligence seems to have compressed this pathway further in recent years. In a limited number of cases, the progression from exposure to extremist material, to the normalisation of violent narratives, to mobilisation or action may become faster and less visible. The phenomenon is not automatic, universal or solely technological. It refers instead to the possibility that AI tools may intensify pre-existing radicalisation dynamics by making content production, adaptation, targeting and interaction easier. Specifically, social media recommendation systems organise and personalise exposure to content; they may create or deepen echo chambers by addressing them towards communities of individuals with similar beliefs and interests. For vulnerable users, this may strengthen confirmation biases, harden grievances and make extremist or conspiratorial ideas feel coherent, recognised and socially validated. Informational barriers are also lowered. Even when violent intent is not explicitly declared, users may seek information that supports reconnaissance, target selection, tactical ideation or preparation. This does not mean that AI systems directly enable attacks, nor that safeguards are irrelevant. However, friction in the movement from grievance to planning are reduced by making information easier to search, organise, summarise and repurpose. In this sense, propaganda production today has become much more effective and customisable: Generative AI tools allow hostile actors to produce text, images, audio and video more quickly and at lower cost. They also facilitate multilingual communication, allowing narratives to reach wider audiences across linguistic and cultural boundaries. This matters because radicalising content can be personalised and adapted to different grievances, national contexts, religious or political references, age groups and identity profiles. AI-populated environments, such as synthetic communities, may reinforce meaning, identity and belonging, especially for individuals who are socially marginalised or experiencing an identity crisis. As humans struggle to distinguish AI-generated and human-written text, the risk is not only that such systems transmit extremist content, but that they simulate social belonging around it. More importantly, and potentially more dangerously, LLM-based companions (unlike static propaganda) can create the impression of a true relationship, offering emotional reassurance, mirroring the user’s worldview and gradually reinforcing hostile interpretations of reality. Hence, Artificial Intelligence may intensify and accelerate pre-existing dynamics, such as the convergence of personal vulnerability, ideological exposure, social isolation and online community influence. For prevention, this means that AI-related risk cannot be addressed only through content removal. It requires attention to platform design, recommender systems, companion technologies, synthetic media, encrypted or semi-private communities, and the offline conditions that make individuals receptive to radicalising narratives in the first place.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.