Executive Summary - Blurred Boundaries: Legal, Ethical, and Practical Limits in Detecting and Moderating Terrorist, Illegal and Implicit Extremist Content Online while Respecting Freedom of Expression

In June 2024, the Dutch Terrorist Threat Assessment (DTN) issued a stark warning: terrorist and extremist actors continue to exploit online platforms to disseminate propaganda, coordinate activities, and incite violence. This content ranges from overtly terrorist material to so-called ‘borderline’ content - material that does not clearly fall within the legal definitions of terrorist or illegal content but nonetheless exerts corrosive effects on democratic values and social cohesion, therefore also referred to as ‘awful but lawful’. The proliferation of harmful content online is not new, yet its scope, speed, and sophistication have expanded dramatically with technological innovation. The emergence of generative artificial intelligence (AI) and large language models (LLMs) has enabled extremist actors to create and disseminate content at unprecedented volume, speed, and precision, often in multiple languages simultaneously. Jihadist groups such as ISIS and Al-Qaeda, as well as rightwing extremist movements in Europe and North America, already deploy these technologies in their propaganda campaigns. The stakes are high. Such content can incite hatred, normalise terrorist violence and deepen societal polarisation. Young people, who primarily access information through social media, are particularly vulnerable to online radicalisation. Extremist actors exploit not only mainstream social media platforms, but also gaming environments, streaming sites, and file-sharing networks. The shift from obscure, difficult-to-access corners of the internet a decade ago to openly accessible platforms today underscore the urgency of the problem. At the same time, the issue raises complex dilemmas at the intersection of security, technology, and fundamental rights. Harmful content often masquerades as humour, irony, or satire, blurring the line between protected freedom of expression and incitement to violence. Overly broad content moderation risks stifling legitimate debate, while insufficient action leaves societies exposed to manipulation and radicalisation. The role of the technology sector is pivotal in addressing the spread of harmful content online, yet its response has been uneven and increasingly subject to criticism. While platforms bear significant responsibility for detection and moderation, their cooperation with independent research and public institutions remains limited. This reluctance reflects broader concerns that major technology companies prioritise profit-driven strategies over societal responsibility, investing heavily in automated detection systems while simultaneously downsizing teams of human moderators. Such trends risk undermining both the quality and legitimacy of moderation, given that algorithmic tools alone are ill-suited to capture the nuance and context of implicit extremist content. Against this backdrop, the Research and Data Centre (WODC), at the request of the Netherlands’ National Coordinator for Counterterrorism and Security (NCTV), commissioned a study - conducted by ICCT - examining the feasibility of developing a reliable framework for detecting and moderating extremist and terrorist content online, without limiting the freedom of expression. Included in this category of content is the so-