Click here to read our latest report “Beyond Extremism: Platform Responses to Online Subcultures of Nihilistic Violence”

When Benign Signals Add Up: The Cross-Platform Blind Spot in Indonesia’s Violent Extremist Ecosystem

When Benign Signals Add Up: The Cross-Platform Blind Spot in Indonesia’s Violent Extremist Ecosystem
18th November 2026 Jose Akmal
In Insights

Researchers often examine online violent extremism through the content, accounts, and interactions visible on individual digital platforms. This approach helps identify extremist material and enforce platform rules, but offers only a partial view of how online activity develops over time. Past research has already documented movement between mainstream platforms and more private or encrypted spaces. The question is therefore not simply whether extremist actors migrate between services, but what happens when different activities are distributed across several services with different functions, levels of visibility and governance arrangements.

Cases investigated by Indonesian authorities offer an opportunity to examine a more specific question: how activities across different digital platforms connect within particular cases of extremist activity or serious violence, and what these connections reveal about an individual’s wider online trajectory. The inclusion of online marketplaces is particularly significant here, as commercial activity may appear routine when examined in isolation, while its relevance to a wider pattern of activity depends on the circumstances of the case and its relationship to other forms of online engagement. 

This Insight examines recent Indonesian cases involving the Islamic State (IS, commonly referred to as ISIS and described as Daulah Islamiyah in some Indonesian law-enforcement reporting), including activity by Indonesian individuals and networks that have pledged allegiance to, are affiliated with, or are inspired by IS, alongside a separate case situated in the broader violent online ecosystem. The Insight argues that the key problem is not platform migration itself, which the literature already establishes, but the fragmentation of behavioural context across categories of digital services. This fragmentation creates a governance challenge: each platform may observe only part of an activity while applying a safety framework designed primarily around its own service. 

Beyond Platform Migration 

Existing research has established that extremist activity can span multiple digital platforms, including movement between mainstream services and private or encrypted environments. Studies have examined how content moderation and platform removals shape this activity, while cross-platform research has increasingly explored how different services interact within wider extremist ecosystems. For example, a 2025 study of German anti-government extremist networks analysed 785,865 Telegram messages from 86 public channels and extracted 332,506 URLs to examine links to other platforms. YouTube was the second most-shared platform for both networks, illustrating how Telegram activity connected to a wider digital ecosystem. Research on Indonesia has also examined the online activities of 55 convicted terrorists between 2017 and 2022. Their research documented multiple categories of cyber activity and identified a specific link between Telegram and the AMN file-sharing application used by the Mujahidin Indonesia Timur (MIT) group, documenting the use of messaging platforms, websites, social media, file-sharing services and e-commerce, and highlighting the limitations of approaches focused primarily on content removal and censorship. 

From Migration to Functional Fragmentation 

The following cases illustrate different ways online activity can be distributed across digital services. While the HOK case demonstrates movement between platforms serving distinct functions, the SMAN 72 case provides a boundary example of engagement with violent online communities without an established terrorist affiliation; together, they show why understanding online activity requires attention not only to platform migration but also to the functions of different services and the limits of what such activity can establish. 

The 2025 investigations into the online recruitment of children offer a clear example of changing visibility. Detachment 88 (Densus 88), the Indonesian National Police’s counter-terrorism unit, reported that around 110 children aged 10-18 across 23 provinces had been identified as having been recruited through online activity. Police described early recruitment and propaganda as taking place on more open platforms, including social media and online games, before interactions moved into more specific and encrypted groups. This pattern is consistent with existing research on the exploitation of social-media and gaming environments by extremist actors, particularly through online communities, in-game communication and gaming-related social networks, as well as the UN Office of Counter-Terrorism. For example, Nuraniyah’s study of pro-ISIS Telegram Groups in Indonesia documents how restrictions on public-facing platforms contributed to the increased use of encrypted messaging applications, examining how private groups strengthened social bonds, group cohesion and ideological commitment. The cases discussed therefore serve a more specific purpose: to examine how different forms of online activity intersect within individual investigations, and what the reported connections between services can reveal about the wider context of that activity. 

The 2024 case of HOK, a 19-year-old suspect arrested in Batu City, East Java, illustrates a different form of fragmentation. Densus 88 identified HOK as a supporter of Daulah Islamiyah/ISIS and described a sequence beginning with interaction in a social-media group (whose platform was not publicly identified), followed by participation in a paid group and later Telegram channels containing ISIS propaganda. Social media provided an initial point of contact; the more restricted group offered a different setting for interaction; Telegram then provided access to explicitly ideological material. 

The 2023 case of DE in Bekasi City, West Java, presents a further variation. Densus 88 reported that DE, whom police identified as an ISIS supporter, used Facebook to disseminate pro-ISIS material and administered Telegram channels containing terrorism-related material. Police also reported that he operated marketplace accounts using false identities to sell military-themed toy weapons, describing his marketplace activity as serving financial and concealment purposes. Instead, it shows functional diversification across service categories: social media for public dissemination, Telegram for more focused communication, and an e-commerce environment that could appear commercially ordinary in isolation. 

The 2025 bombing at State Senior High School 72 (SMAN 72 Jakarta) in Jakarta provides a useful boundary case. Indonesian authorities did not classify the incident as terrorism, despite evidence of engagement with violent online communities and mainstream digital services. The case therefore differs from the ISIS-linked investigations discussed above: the reported online activity involved communities such as the true crime community (TCC), and mainstream digital services on TikTok and a Telegram group reportedly glorifying violent attacks. At the same time, the publicly reported evidence does not establish a clear sequence of movement between named platforms. The case thus demonstrates why online activity must be examined in context, rather than treating exposure to violent content or participation in an online community as sufficient evidence of terrorist affiliation.  

Taken together, these cases do not establish a single pathway from online activity to violent extremism. They instead reveal three related forms of fragmentation. The key governance challenge is that activity across multiple digital services may be difficult to understand when each platform is examined in isolation, due to differences in visibility, functionality and safety frameworks that can limit the ability to identify broader patterns. This highlights the need for a more contextual and coordinated approach to cross-platform governance, without assuming that activity across services or in less visible spaces is, in itself, evidence of harmful or extremist intent. 

From Cross-Platform Activity to the Limits of Platform-Specific Governance

These cases raise questions about how platform-specific safety policies interact with activity that spans different digital environments. Companies do not necessarily approach this problem in the same way. Meta, for example, applies its community standards to its public-facing social-media platforms, particularly Facebook and Instagram, governing content shared through public posts, groups, and other platforms; it distinguishes its approach to WhatsApp’s encrypted private messaging, which is also owned by Meta, which operates under a different model centred on private messaging and end-to-end encryption, limiting the company’s ability to access the content of private communications, while Telegram’s terms specifically address the promotion of violence in publicly viewable channels and bots. Shopee Indonesia, by contrast, explicitly states that it may consider information from other platforms and offline sources when identifying violent or extremist actors on its service. The DE investigation offers a concrete Indonesian example of the practical significance of these boundaries: Densus 88 reported coordinating with PPATK (Indonesia’s Financial Intelligence Unit (FIU)), online platforms, and marketplaces to trace transactions linked to accounts operating under false identities. However, the available reporting does not establish that the companies involved made conflicting moderation decisions or that any particular company failed to recognise a risk. The key governance question is how to interpret information shared across different services. This requires considering the context in which the activity occurred and each platform’s specific responsibilities. Platforms should not be assumed to have identical responsibilities, and ordinary activity should not, on its own, be treated as evidence of extremist intent. 

An Australian eSafety Commissioner review of platform reporting, for example, identified differences in detecting terrorist and violent extremist content: Telegram’s use of hash-matching was limited to previously known content in certain private environments, while WhatsApp reported differences in detecting new video and images. Mechanisms for cross-platform cooperation already exist. The Global Internet Forum to Counter Terrorism (GIFCT), for instance, facilitates the sharing of digital fingerprints of known extremist content and coordinates responses to incidents with an online dimension. Its revised 2025 framework also provides for information-sharing when an attacker has a significant online presence but has not produced media as part of the attack. However, these mechanisms do not amount to a universal system for connecting individual activity across social media, gaming services, messaging platforms and online marketplaces. The Indonesian cases therefore raise a more specific question: to what extent can existing detection and cooperation mechanisms account for relationships between different forms of online activity, and where might contextual gaps remain? 

In the HOK investigation, Densus 88 described a sequence of engagement across social-media groups and Telegram channels, while the DE investigation involved activity across Facebook, Telegram, and online marketplaces, with police later coordinating with financial investigators and online services to trace transactions. These cases show that understanding a broader pattern of activity may require connecting different forms of digital evidence. However, the publicly reported investigations do not establish that individual platforms failed to detect relevant activity or that information-sharing between companies was inadequate. The SMAN 72 Jakarta case further demonstrates the importance of contextual interpretation: despite reported engagement with online communities featuring graphic violence, Densus 88 found no evidence linking the incident to a terrorist network.  This distinction is important because the online communities involved, including TCC, are not themselves established terrorist organisations. Engagement with communities that glorify violence should therefore be distinguished from affiliation with a terrorist network. The significance of cross-platform analysis, therefore, lies not in assuming that activity across services indicates a common trajectory towards extremism, but in examining when connections between activities are substantiated and what those connections can legitimately establish. 

What Can Technology Companies Do?

If the problem is fragmented context, the response should not simply be to collect or share more user data. A more proportionate approach would be to improve how platforms recognise, assess and communicate limited contextual signals while preserving legitimate privacy and avoiding assumptions about intent. 

First, strengthen contextual escalation through existing moderation and review processes. Platforms could build on mechanisms such as GIFCT’s Hash-Sharing Database and Tech Against Terrorism’s Terrorist Content Analytics Platform, which help identify known extremist content and alert services to relevant URLs. These mechanisms provide an initial basis for identifying verifiable signals, but they do not establish whether otherwise ambiguous activity across several services forms part of a wider pattern. A further step would be to introduce clearly defined escalation criteria under which multiple independently substantiated indicators trigger review by a trained specialist rather than automatic enforcement. For example, an interaction in a social-media group should not trigger escalation solely because it precedes activity elsewhere. Greater scrutiny would be justified only where additional evidence meets established thresholds and the combined context warrants examination. This approach would extend existing content-focused processes towards contextual assessment without treating cross-platform activity itself as evidence of extremist intent. 

Second, develop narrowly scoped, privacy-preserving cross-service referrals. GIFCT’s Hash-Sharing Database and the Internet Watch Foundation’s Hash List demonstrate that information can be shared through verified digital fingerprints without routinely exchanging the underlying content or user histories. These models address known content, however, rather than relationships between individuals’ activities across services. A potential extension would be a trusted referral mechanism through which a platform facing a sufficiently serious and documented concern could submit a minimal case record for specialist assessment. The record might contain a case reference, the relevant type and time of activity, links or hashes for verified content where applicable, and a documented explanation of the concern. Participating services would not receive unrestricted access to one another’s user databases or communications. Where a more specific cross-service match is necessary, any identity-related processing would require a defined legal basis, strict eligibility criteria and additional privacy safeguards. A referral would prompt contextual review, not constitute proof of wrongdoing or automatically trigger account restrictions. Such a person-level referral mechanism would be a proposal to develop and evaluate, rather than a capability demonstrated by existing content-hashing systems.

Third, extend cross-industry coordination beyond individual content categories. GIFCT’s Incident Response Framework coordinates incidents with an online dimension, while Tech Against Terrorism’s TCAP maintains a database of verified terrorist content and provides advisory alerts to participating platforms when it identifies matching content on their services. already provides structured coordination during incidents with an online dimension, while Tech Against Terrorism’s TCAP helps alert services to terrorist content identified across the internet. Participating companies could develop a common referral taxonomy distinguishing verified content, reported interactions, account-level concerns, transaction-related evidence and contextual links between activities. A designated coordination channel could then route qualifying referrals to the relevant services and specialist reviewers, with each company retaining responsibility for decisions under its own policies. Shared procedures should also specify evidence thresholds, permissible purposes, retention periods, audit requirements and mechanisms to challenge erroneous referrals. The objective would not be to create a universal monitoring system or impose identical moderation rules across industries, but to make existing cooperation mechanisms better able to handle carefully evidenced concerns that span different service categories. 

–

Mochammad Jose Akmal is a policy and research professional with a Bachelor’s degree in Political Science and Government from Universitas Diponegoro. He currently serves as a Program and Research Associate at the Foreign Policy Community of Indonesia (FPCI) Climate Unit, contributing to climate policy research, climate diplomacy, and stakeholder engagement. His professional experience spans international institutions, strategic research, and economic journalism, including work with the European Union Delegation to Indonesia, Semar Sentinel, and Kontan Indonesia. His analysis and commentary have been published in international outlets such as International Policy Digest, The Global Network on Extremism and Technology (GNET), and IR UI Commentaries. With expertise in geopolitical analysis, strategic communication, governance, and foreign policy, Jose focuses on translating complex political, security, economic, and climate-related developments into actionable insights for policy and decision-making.

The views and opinions expressed in this Insight are the authors’ own and do not necessarily reflect those of GNET or any of its partners.