Terrorists have always adopted and employed emerging technologies to organise acts of terror, radicalise, and produce propaganda to mobilise recruits. Among them, the Salafi-Jihadi groups and, in particular, the Islamic State (IS), have proved to be consistently ahead of the curve in the adoption of new, cutting-edge technology, creating a consistent and persistent online presence. Despite theological quibbles, Artificial Intelligence (AI) is among the technologies that IS supporters are employing to produce and disseminate propaganda. By analysing previous uses of AI within the IS online ecosystem, this Insight will discuss a case in which a pro-IS user adopted an open-source AI tool to spread IS-English translated propaganda. The Insight will also explore how these models can facilitate and promote the dissemination of propaganda materials.
IS sympathisers and supporters’ interest in AI systems became apparent in 2023, when a pro-IS non-institutional media house released a guide on using these tools while protecting users’ privacy. In the same year, Islamic State Khorasan Province (ISKP) issued AI-powered courses to train propagandists. Then, in March 2024, the first widespread use of AI was detected on the pro-IS server, TechHaven, on Rocket.Chat. Indeed, following the Crocus City Hall attack in Moscow, a video news bulletin with AI-generated characters was published on the platform. Subsequently, on 17 May 2024, ISKP produced a propaganda bulletin generated with AI tools to claim responsibility for the public market attack in the Afghan Bamiyan Province. Later that month, another AI video announcing a bombing in the city of Kandahar was released. Finally, in April 2025, the non-institutional media house Qimam Electronic Foundation (QEF) shared a guide to AI tools and the risks related to them, particularly in terms of security and privacy, also on Rocket.Chat.
The use of Whisper OpenAI within the pro-IS ecosystem on Rocket.Chat
The key case of interest, propaganda translation using AI, was observed within TechHaven, the main chat room of the pro-IS server hosted on Rocket.Chat, which is considered a digital safe-heaven for Salafi-Jihadi extremists. Platforms like Telegram, Element, WhatsApp, and, Rocket.Chat forms part of the so-called “beacons,” the core platforms sustaining IS’s online ecosystem.
On 6 May 2025, a user shared a video made by the Al-Battar Foundation (a media house part of the Ansar Production) on TechHaven. The clip is part of a series of more than seventy videos produced by Al-Battar Foundation since February 2024. The videos consist of Arabic-language vocal recitations of editorials from An-Naba, a weekly institutional publication of IS. These videos are well produced, including imagery of the lightly edited cover of the editorial in question, and a male narrator’s voice in Arabic reading the content in full. It’s possible to identify the logo of Al-Battar Foundation in the top-right corner, under which one can see the logo of An-Naba and, following the icon of a microphone, the sentence in Arabic “Reading the editorial of An-Naba issue [in this case] 493.” The peculiar aspect of the video shared by the user is that the producer, as expressed in the comment accompanying the content, used Whisper OpenAI to translate and transcribe the Arabic audio into English. Later, the producer used the text created by Whisper OpenAI to add English subtitles to the video, too. Moreover, the video does not show any logo or affiliation to any non-institutional media house. Possibly, since this is the first video, the producer did not insert any logo. Yet, as has already been documented in another case, the producer could insert its own logo secondarily to be identified later within the pro-IS online ecosystem. One of the most notable features is that the video ends with a name in the subtitle, namely “Abu Abdullah.” Since this name has often been used as a kunya (honorific title) to address AQ and IS leaders, it is not possible to assert that it is the name or nickname of the producer. What can be said definitively is that this name is not that of the user who shared the video on TechHaven.
Whisper is an OpenAI automatic speech recognition (ASR) model that converts spoken words into text. It is an ASR system trained on 680,000 hours of audio data collected from the web. It was announced by OpenAI at the end of 2022 with its mission to help identify and transcribe human speech. This tool can be heralded as an important development in open-source AI models for a number of reasons, such as the multilingual support, the high level of accuracy, and the ability to adapt to diverse accents. Additionally, it allows for complete translations into English and is available for free. However, the platform has a few limitations. Namely, there is scarce support for less common languages, and inaccuracies when dealing with very specialised topics. Nevertheless, said constraints can be overcome, since the model can be fine-tuned in order to conduct specific activities. The system supports nearly 100 languages, but training in less-spoken ones is needed. In view of the above considerations, Whisper OpenAI is thus remarkable as it improves communication and automation. However, in the context of the IS digital ecosystem, the model surely facilitates the dissemination of the group’s propaganda materials online.
Importantly, the user who shared the video under analysis chose TechHaven to spread the propaganda product. This choice is significant in several respects. First, as discussed above, Rocket.Chat constitutes one of IS’s core online platforms, and it represents IS’s propaganda launchpad. Moreover, the user aimed to reach a substantial segment of the group’s supporters and sympathisers since, on TechHaven, there are more than 10,800 users, as identified through the authors’ monitoring. Had the video been published on a different platform, the content may have gone unnoticed or had less resonance within the IS online ecosystem. Instead, the user received an almost immediate reaction from the IS community on Rocket.Chat. On the same day the user published the video, another pro-IS user reacted to it, praising the quality of the translation and welcoming the spread of AI tools to distribute the group’s content online (Figure 1).
Figure 1: User reacts with praise and positivity to the AI-generated video posted by another IS supporter
Furthermore, from an operational security perspective, Rocket.Chat is the most secure alternative among the platforms that are typically used by IS to disseminate propaganda materials. On Rocket.Chat, IS can thus self-manage its server, namely TechHaven. Put differently, it is a communication platform which embodies several key principles of the Web 3.0 era, including decentralisation, independence, and cryptography. Since TechHaven is self-hosted by IS, the content is subject to complete control by the group. Therefore, the user was sure that the video would have been neither reported nor deleted, since Rocket.Chat could be seen as IS’s “citadel”.
AI Translation: Possibilities and Countermeasures
The voice recitations produced by Al-Battar are an advanced step in the propaganda of the Ansar Production media houses. On the one hand, they provide a vocal version of An-Naba in the most widely used languages within IS’s online ecosystem. On the other hand, they make An-Naba’s texts much more accessible for the thousands of IS supporters and sympathisers.
Whisper OpenAI’s potential can continue to be exploited by IS if proper countermeasures are not implemented. As noted above, Whisper could be highly useful in translating Arabic propaganda content into English if adopted more widely, thereby enabling IS to reach a wider range of users on the web for radicalisation purposes. Furthermore, materials such as the video under analysis could be spread with ease on other social media, e.g., Facebook, Instagram, and TikTok, wherein IS propaganda material is openly accessible. This kind of propaganda content might allow the group to avoid content moderation. As the authors observed, on several platforms, IS supporters are able to overcome moderation procedures by simply spreading propaganda material in Arabic. Through direct monitoring, it is clear that IS Arabic propaganda material is less likely to be removed compared to English material. Moreover, it would be much more difficult for the platforms themselves to identify this type of vocal content rather than a post containing a visual excerpt from An-Naba’s editorial. As noted above, IS supporters frequently share audio materials, which further complicates the moderation process. As observed on TikTok, pro-IS users often insert extracts of IS leaders speeches in videos usually depicting nature or symbolic animals in Islamic and jihadi culture (such as lions and cats) to avoid content moderation.
Overall, AI tools can enhance the distribution of IS propaganda by increasing the speed of the Multiplatform Communication Paradigm (MCP). The latter refers to the issue of IS, and Salafi-Jihadi groups in general, operating on multiple interconnected platforms in order to communicate with supporters, advocate da’wa (namely, proselytism), and spread propaganda. Consequently, the ‘media mujahideen’, such as pro-IS online supporters, have created “fluid, dispersed networks” to disseminate their media content. In this context, the growing use of open-source AI models would increase IS’s ability to reach different communications platforms, thereby improving the resilience, agility, and speed of the group. How IS operates can be understood as a “Swarmcast”. Following this theoretical framework, the media mujahideen can constantly reconfigure their online presence much like “a swarm of bees or flock of birds”.
Conclusions
AI is profoundly re-shaping social interactions. Terrorist organisations have been exploiting the proliferation of these technologies as well, fulfilling their subversive purposes, especially as far as the diffusion of propaganda is concerned. The Salafi-Jihadi movements have exploited said tools in order to counter online disruption and content removal. IS continues to demonstrate an adept engagement and proficiency with emerging technology, increasing its broadcasting capabilities in accordance with the medium utilised. The video created with Whisper OpenAI clearly shows how the group’s supporters have been keeping up with the rise of AI-driven tools. IS’ employment of AI models appears to be highly beneficial for the group, allowing it to improve its ability to spread official propaganda and increase its da’wa efforts.
It is therefore essential that security agencies develop counter-terrorism responses that incorporate effective tools to limit the creation of AI-generated content and the diffusion of this type of propaganda. However, it is also imperative that AI providers undertake proactive monitoring activities and limit access to these technologies by users who intend to create terrorist content, striking a balance between innovation and regulation. Finally, as previously expressed, security and law enforcement agencies should draw on decades of research and studies on Salafi-Jihadi movements to train AI to recognise propaganda content, categorise it, and then eliminate it. Moreover, they should train AI tools with Arabic textual and visual content in order to improve their capacity to detect and remove terrorist propaganda from social networks. The lack of Arabic-speaking moderators and inadequate training data for the various Arabic dialects thus hinders effective moderation. The enormous amount of propaganda material produced by Salafi-jihadi movements over the past thirty years or so would, in fact, provide AI that can recognise visual and textual elements typical of Salafi-Jihadi propaganda.
—
Alessandro Bolpagni is a senior research analyst at the Italian Team for Security, Terroristic Issues, and Managing Emergencies – ITSTIME. He also specialised in digital HUMINT (Human Intelligence) and OSINT/SOCMINT (Open Source and Social Media Intelligence), oriented particularly on Islamic terrorism and Russian Private Military Companies (PMC). He focused on monitoring terrorist networks and modelling recruitment tactics in the digital environment, particularly on new communication technologies implemented by terrorist organisations. He has produced operational reports and lectures for training and education purposes for companies, research centres, and law enforcement agencies.
Eleonora Ristuccia is a junior research analyst at the Italian Team for Security, Terroristic Issues, and Managing Emergencies – ITSTIME. She specialises in the application of Open Source Intelligence (OSINT) and Social Media Intelligence (SOCMINT) techniques. Specifically, she focuses on the study of jihad communication strategies and propaganda through various online platforms.
—
Are you a tech company interested in strengthening your capacity to counter terrorist and violent extremist activity online? Apply for GIFCT membership to join over 30 other tech platforms working together to prevent terrorists and violent extremists from exploiting online platforms by leveraging technology, expertise, and cross-sector partnerships.