Content Warning: this Insight contains antisemitic, racist, and hateful imagery and language.
Introduction
On Telegram, there is a right-wing extremist (RWE) accelerationist collective that disseminates ideologically extremist materials, encourages violence, glorifies terrorism, and demonises minority populations. The collective functions as a loose network with no formal affiliation to any group but is closely associated with several extremist organisations, including Russian mercenaries, Ukrainian volunteer battalions, Ouest Casual (a French extreme-right pro-violence group), and The American Futurist, which is closely associated with the neo-Nazi James Mason and former members of Atomwaffen Division.
Most of those channels have a neo-Nazi ideological position and distribute guides and instructions on how to commit racially motivated acts of terrorism against the government and authorities. Their propaganda frequently invokes visual themes of militants, terrorists, troops, and scenes from ongoing disputes in the Middle East, Chechnya, the Balkans, and Northern Ireland.
The collective is highly decentralised. It is, therefore, the actions of individuals that determine the group’s online activities, making them highly unpredictable. At present, one of the most popular methods of RWE propaganda production is generative artificial intelligence (AI).
Through digital ethnographic data collection, this Insight delves into how accelerationists on Telegram use AI to create several types of images to spread propaganda. Furthermore, it considers their exploitation of large language models (LLMs) to obtain information to conduct attacks or interpret manifestos, providing an overview of how violent extremist actors exploit AI for their ideological purposes.
How RWEs on Telegram Use AI
Accelerationist manifestos call for the use of all technologies that will ultimately lead to societal collapse and a race war. Members of the online collective refer to the Unabomber/Ted Kaczynski’s manifesto to justify using every available tool to take down the system. AI has the potential to create massive disinformation campaigns, feed radicalising pieces of propaganda to unsuspecting online users, gather information on potential targets, or even find instructions to create explosive devices. It can also be exploited to write malware, enabling extremists to attack online infrastructures.
Certain far-right accelerationist Telegram channels are dedicated to creating and disseminating AI-generated memes and propaganda (Fig. 1). These channels have several thousand subscribers and contain thousands of images representing all ideological aspects of the extreme right. Based on our analysis, this type of content can be classified into three main categories.
Fig. 1: Two RWE Telegram channels focused on AI-created imagery content. Together, they have posted nearly 8,000 photos.
Exaltation of Nazi Imagery and Military Figures
Images depicting German WWII soldiers are aimed at reinforcing the archetype of the strong, white, militant man. The exaltation of the militant man is also effective in radicalising online users and convincing them not only of the need but also of the beauty inherent in violence. According to accelerationists, violence is the primary means of hastening the process of systemic collapse because there is no chance for a political solution; the system must fall to begin afresh.
Fig. 2, posted on an eco-fascist accelerationist channel, contains a military figure in tactical gear and a skullmask and is captioned “The Industrial Revolution and its consequences have been a disaster for the human race”. This is the famous incipit of Ted Kaczynski or the Unabomber’s manifesto. Kaczynski is a key figure in eco-fascist, accelerationist online subcultures, revered as a saint on many RWE Telegram channels, and his manifesto has become a fundamental cornerstone of their ideology. From an aesthetic perspective, the font in which the incipit is written is used by the neo-Nazi propagandist Dark Foreigner and is commonly found in the propaganda of terror groups such as Atomwaffen Division. This image could be appealing to the average accelerationist user; the font, skull mask, and tactical gear depict what is perceived as the archetypical man. Fig. 3 shows three WWII Nazi soldiers depicted in using the vaporwave/fashwave aesthetic to convey far-right extremist affiliations. The psychedelic aesthetic may be a personal preference of the content creator, who often shares visually similar content. This image relies heavily on its visual impact, glorifying the Nazis.
Fig. 2: A man in tactical gear, wearing a skull mask. The beginning of Ted Kaczynski’s Manifesto was written with a font that can be easily associated with the Terrorwave aesthetic.
Fig. 3: A psychedelic representation of German soldiers during the Second World War.
Racist and Antisemitic Imagery
Racism and antisemitism are fundamental components of far-right ideology. Themes of the perceived superiority of white people, conspiracy theories about an ongoing ‘white genocide’ brought on by racial mixing and the alleged invasion of migrants in Europe and North America are frequently observed in online propaganda. Jews are targeted in AI-generated propaganda, painted as a threat to Western civilisation, and responsible for issues of perceived moral decline from the pornography industry, the LGBTQ+ community, mass migration, and the COVID-19 pandemic. AI-generated images often contain harmful stereotypes, allusions to conspiracy theories, or explicit calls to violence (Fig. 4). Usually, the latter is accompanied by acronyms like TND (Total N***er Death) or TKD (Total K**e Death, an antisemitic slur).
Fig. 4: (Left) Jews are implied to be the ones truly behind the COVID-19 pandemic. This can be collocated on a wide range of conspiracy theories regarding the pandemic and the vaccine campaigns, which by many RWE online are believed to be a way to control the population. (Middle) Jews are implied to be behind the arrival of migrants from the sea to Europe. In this case, the reference is to the Great Replacement conspiracy and a supposed ‘White genocide’. (Right) Reference to the RWE internet trope TND (Total N***er Death), which is used to indicate the need to exterminate black people.
Memes
The third type of content typology relates to memes – an effective and simple means of disseminating RWE propaganda. The most popular formats in our dataset are those of Pepe the Frog and Moon Man. These are the most used memes in online alt-right and far-right communities.
Pepe the Frog originated in 2005 as part of an innocuous comic series ‘Boy’s Club’ and rapidly became a popular meme on 4chan by 2008. In 2014, the meme was coopted by the alt-right and the far-right to advance white supremacist narratives online. It even became a potent force in the 2016 US Presidential elections after Trump retweeted a version of himself as the character.
Moonman originated in 1986 as Mac Tonight, a McDonald’s mascot. The first appearance as a meme can be dated back to 2006, when Moonman appeared on YTMND (You’re the man now, dog!) as an animated gif. In 2015, the meme was coopted by the far-right and has since been frequently associated with white supremacist jargon. Examples of those two memes depicted by an AI model can be seen in Fig. 7, in which Pepe the Frog has been made to resemble Adolf Hitler, and in Fig. 8, where Moon Man is shown suffocating an African-American man.
Fig. 5: (Left) Pepe resembling Adolf Hitler. (Right) Moon Man suffocates a Black man with a rope.
The Exploitation of LLMs
The creation of images is not the only application of AI exploited by extremists. RWE channels have also exploited LLM, even developing their own or partially modifying existing ones to bypass built-in safety features designed to avoid users producing and disseminating dangerous or xenophobic content.
LLMs not only have the potential to forge large-scale disinformation campaigns but can also be forced to provide information that could help a violent extremist prepare for an attack. One example can be seen in Fig. 6, where an eco-fascist channel requested instructions to build a trigger mechanism – the firing device needed to initiate the explosion – and further explanation of the functioning of Kaczynski’s bombs. Another example of extremist use of LLMs is the development of ‘unbiased’ AI models (Fig. 7). A far-right user posted both instructions for running the model without censorship and a review of an accelerationist manifesto requested by one of the AI administrators. A manifesto review, given its brevity, can be disseminated more easily than the manifesto itself, increasing its radicalising potential.
Fig. 6: A user asked for information on the production of a triggering mechanism for fireworks and about Kaczynski’s bombs and then posted them on its eco-fascist channel [text redacted]
Fig. 7: A channel explaining how to unlock an unbiased AI and asking for a review of an accelerationist manifesto.
Conclusion
According to recent reports detailing extremist networks on Telegram, online RWEs exploit existing generative AI models for the production of visual propaganda and even the development of explosives used in kinetic attacks. However, there is a looming concern that with the increasing IT capabilities among far-right groups, a scenario could emerge where AI is harnessed to generate more sophisticated and targeted propaganda, as well as to carry out cybercrime campaigns targeting online infrastructures.
For this reason, it is imperative to intensify research efforts within these ecosystems to prevent and counter the use of AI by extremists and to adopt a proactive approach to prevent future threats. Tech companies must remain vigilant in monitoring the development of novel, extremist-owned models which may be misused for nefarious purposes. Implementing internal threat assessment teams and devising terrorism-focused procedures are crucial steps to identifying and addressing potential threats posed by RWEs using AI technologies.
Federico Borgonovo is a research analyst at the Italian Team for Security Terroristic Issues and Managing Emergencies – ITSTIME. He specialized in digital HUMINT, OSINT/SOCMINT, and Social Network Analysis oriented on Islamic terrorism and RWE. He focuses on monitoring terrorist networks and modelling recruitment tactics in the digital environment, with particular attention to new communication technologies implemented by terrorist organizations.
Silvano Rizieri Lucini is a research analyst at the Italian Team for Security Terroristic Issues and Managing Emergencies – ITSTIME. He specialised in digital HUMINT and OSINT/ SOCMINT oriented on Islamic terrorism, Whitejihadism, and RWE. He focuses on monitoring terrorist networks, with particular attention to new communication strategies implemented by terrorist organisations.
Giulia Porrino is a research analyst at the Italian Team for Security Terroristic Issues and Managing Emergencies – ITSTIME. She specialised in digital HUMINT, social media intelligence, Social Network Analysis, and Socio-Semantic Network Analysis. Her research activities are oriented on RWE, with a focus on PMC Wagner and Russian STRATCOM.