Quantitative research on terrorist recidivism remains surprisingly limited, despite heightened concern among policymakers, practitioners, and the general public. Several factors contribute to this situation, including restricted access to sensitive data and the difficulty of assembling samples large enough to support robust statistical analyses, given the inherent rarity of terrorist offending. In Belgium, more than 650 individuals were convicted of terrorism-related offences between 2006 and 2024. Drawing on Belgium’s Central Criminal Record, this study examines post-conviction trajectories of this population. Building on a first wave of research published in 2025, it benefits from an extended observation period (2006–2024) and a larger, more homogeneous analytical sample. The study pursues two objectives: first, to provide a detailed description of recidivism patterns—prevalence, timing, and offence types—within this population, and; second, to identify factors associated with reoffending using survival analysis and Cox proportional hazards models. Findings show that terrorist recidivism is exceedingly rare (2.5%; n = 10), whereas general recidivism (18.7%) predominantly concerns common criminal offences (e.g., traffic violations). Survival analysis indicates that the instantaneous risk of recidivism peaks within the first five years after conviction, followed by a sharp decline. Factors associated with recidivism include both classic criminological factors (e.g., prior convictions, gender) and variables specific to terrorism-related offenders (e.g., organisational affiliation and role). Criminal career duration also emerges as a key determinant of recidivism. The article concludes by discussing the study’s methodological limitations and the implications of the findings for prevention, risk assessment, and policy development.
This research note examines the use of artificial intelligence (AI) as a decision-support tool in the monitoring and evaluation (M&E) of preventing and countering violent extremism (P/CVE) programmes. Drawing on an actual programme involving a bystander awareness initiative, the analysis compares an AI-assisted M&E framework generated using Microsoft Copilot with the publicly available external evaluation of the programme. The research note critically assesses how AI tools may support rapid early-stage programme design, but only when embedded within human-centred evaluation processes that emphasise judgement, transparency, and ethical oversight. Implications for responsible integration of AI tools into P/CVE evaluation practice are discussed.
In recent decades, Salafi-Jihadi groups have capitalised on the proliferation of social media to establish a persistent online presence. In that time, their activity has evolved into the so-called Multiplatform Communication Paradigm (MCP), a strategy to establish a presence on multiple online platforms instead of focusing on a single platform. Previous research has examined the digital ecosystem through outlinks from social media platforms such as Twitter and Telegram. On these platforms, Salafi-Jihadis face disruption and account removal. As distribution methods evolve to avoid detection, data collection is often based on availability at a given time. In contrast, this research is the first to present a longitudinal study, spanning almost six years of outline sharing from a Salafi-Jihadi digital safe haven, the IS Rocket.Chat server. Here, IS supporters are free to operate without removal, and thus, the research provides valuable insight into how the ecosystem operates when it is not facing disruption. The research provides an initial overview of the outlink data and demonstrates the continued importance of platforms fulfilling the roles of beacons, aggregators, and content stores within the digital ecosystem. It also emphasises the role of human choices in domain selection, as not all rooms share the same combinations of URL. This suggests a vital new, human-centred avenue for research into terrorist exploitation of the internet.
White nationalism has been subject to considerable debate by academic and online communities alike. We contribute to this body of scholarship by investigating constructions of white nationalism by (presumably) white nationalists online. To accomplish this, we extract and examine 92,000 posts on the topic between 2013-2024 from 4chan’s well-known /pol/ board. Using a mixed methods approach, we combine soft clustering of large language modelling embeddings with qualitative thematic analysis. Our results identify discursive constructions of an ethnic, gendered, and religious identity on 4chan, supposedly denied self-determination. By synthesising scholarly definitions with 4chan expressions of white nationalism, we also aim to detangle these constructions from white supremacy to advance an academic definition. We suggest white nationalism broadly describes the belief that humanity is naturally divided into nations, with a white nation that possesses distinct ethnic, political, territorial, and sovereign claims. Specifically, white nationalism as expressed on /pol/ is a form of ethnonationalism which champions an imagined (white) political and territorial community, whose idea of meaningful sovereignty is contingent upon the free exercise of white political will to shape their collective destiny.
Violent non-state actors (VNSAs) are increasingly using commercial geospatial technologies – e.g., unmanned systems, geographic information systems, and the Internet of Things – to conduct attacks. We conduct qualitative content analysis on interviews of twenty subject matter experts (SMEs), informing why some VNSAs use geospatial technologies, and how they choose among available technologies to augment tactics, techniques, and procedures (TTPs). We find that experts note two key dimensions: barriers to adoption and the technical capabilities of threat actors. Our analysis helps to categorise risks and offers insights into both the relative prevalence of different kinds of attacks, and the mechanisms by which geospatial technologies may shift the threat landscape.
