Introduction
This Insight is the second in a two-part series investigating the ways white supremacists engage with generative artificial intelligence (GAI), specifically their tactical use and discourse. For the full methodology, please see the first Insight in this two-part series.
Extremist discourse concerning GAI, particularly that of white supremacists, has largely been neglected in favour of studies and investigations focusing on the tactical exploitation of the technology for nefarious purposes. The few studies which do exist suggest that white supremacists view GAI as largely negative and link the tool to existing conspiracies. However, to date, there have been no large data collections of white supremacist GAI discourse, resulting in a limited understanding of how white supremacists view the emerging technology.
Subsequently, the second part of this Insight series aims to address this gap by investigating the most popular and dominant themes in white supremacist GAI discourse. With knowledge of emerging and changing discourse, counter-extremism practitioners and tech companies will be able to stay ahead of future trends by assessing the overall sentiment towards GAI tools.
Results
The coding of 850 Telegram posts resulted in 1,077 instances of engagement with GAI, as some posts were complex, containing multiple forms of engagement. Five broad ‘thematic overviews’ were evidenced (Fig. 1). While the ‘Tactical Use’ thematic overview related to ways in which white supremacists use GAI for tactical purposes, this Insight will focus on the discourse-related thematic overviews; ‘Knowledge Sharing’, ‘GAI Pessimism’ and ‘GAI Optimism’. With 630 instances recorded, thematic overviews emerged as the most prominent focus within generative AI discourse, representing 58% of total engagement across all interactions.
Figure 1: Thematic Overview Tree Map
GAI Pessimism
The most common discourse-related thematic overview emerged as ‘GAI Pessimism’ – the belief that GAI is a negative force for white supremacists and humanity at large. 375 instances of ‘GAI Pessimism’ were evidenced, occupying 35% of the entire data collection and 60% of discourse-related themes. Spanning 32 Telegram channels, ‘GAI Pessimism’ is the most dominant discourse-related thematic overview and is formed of three themes, as shown below.
Figure 2: GAI Pessimism Tree Map
Threat to Humanity
‘Threat to Humanity’ emerged as the most prominent theme in ‘GAI Pessimism’, with 199 instances, accounting for 53% of ‘GAI Pessimism’. As shown in Figure 3, the theme centres on the perceived cultural, societal and security threats GAI poses to humanity.
Figure 3: Threat to Humanity
‘Threats to Freedom/Civil Liberties’ emerged as a major theme. Posts claimed GAI tools would be used to “monitor” and enhance surveillance efforts, creating a totalitarian state. They often tied these concerns to conspiracy theories such as the New World Order. A typical post can be seen in Figure 4, with the post stating AI will be used as an “invasive tool of surveillance.”
Figure 4: Example of Threats to Freedom/Civil Liberties
Threat to In-Group
The second most dominant theme in ‘GAI Pessimism’ emerged as ‘Threat to In-Group’, with 114 instances, occupying 30% of the thematic overview. As Figure 5 shows, GAI is viewed as a threat to whites and the broader white supremacist movement.
Figure 5: Threat to In-Group
Emerging as the most dominant sub-theme, perceptions that GAI tools are biased and promote a ‘Liberal Worldview’ are expressed as a threat to the ingroup, promoting an ideology opposed to white supremacy. Posts focused on the alleged liberal biases of mainstream GAI tools, claiming that ChatGPT “couldn’t define a woman” and has “clear preference for left wing views” (Fig. 6).
Figure 6: Examples of ‘Promoting a Liberal Worldview’
In ‘Threat to Ingroup’, the second most dominant sub-theme emerged as ‘White Replacement.’ Here, GAI was being viewed as a threat to the white race, often alluding to the Great Replacement Theory. Posts cited alternative news articles claiming Google’s AI promotes “White Erasure” (Fig. 7), while others claimed Google Gemini has been “programmed to erase white people” (Fig. 8).
Figure 7 & 8: Examples of White Replacement
Conspiratorial Control
‘Conspiratorial Control’, instanced 62 times, emerged as the belief that GAI is controlled by a cabal of malign and sinister actors. As shown by Figure 9, two sub-themes emerged in this theme.
Figure 9: Conspiratorial Control
‘Globalist Control’ was the most common theme, manifesting through posts discussing a “secretive” global and liberal elite controlling GAI, referencing Bill Gates and the Bilderberg Group (Fig. 10). Posts displaying ‘Globalist Control’ would also frequently display evidence of ‘Threats to Freedom/Civil Liberties’. Here posts would claim that globalist actors are creating GAI tools and using GAI solely for the purpose of curtailing the freedoms and civil liberties of the masses.
Figure 10: Example of Globalist Control
Knowledge Sharing
A further thematic overview emerged as ‘Knowledge Sharing’, with 121 instances, accounting for 11% of the overall dataset and 19% of discourse-related themes. Here, techniques and tools prompts were shared to overcome GAI safety restrictions and to produce effective propaganda. The themes can be seen in Figure 11 below.
Figure 11: Knowledge Sharing Tree Map
Knowledge Sharing for Effective Propaganda: GAI-Images
‘Knowledge Sharing for Effective Propaganda: GAI-Images’ proved the most dominant theme in ‘Knowledge Sharing’, accounting for 41% of the thematic overview. As shown by Figure 12, three sub themes comprised this theme.
Figure 12: Effective Propaganda: GAI-Images
‘Prompt Sharing: GAI-Image Generators – Baiting’ emerged as the most instanced sub-theme. As discussed in the previous Insight, ‘baiting’ refers to the process of white supremacists asking GAI tools manipulative questions in the hope that tools produce answers which are seen to align with their key grievance narratives, namely that AI is liberally biased and anti-white. Once a white supremacist successfully ‘baited’ the tool, they screenshotted the interaction, posting it on various Telegram Channels. Subsequently, the tactic acts as an effective form of propaganda, often receiving hundreds and even thousands of interactions on Telegram by seemingly providing irrefutable evidence of key white supremacist grievance narratives.
As shown by Figure 12 in Part One, ‘baiting’ was often performed on GAI Image tools, commonly promoting the grievance narrative that these tools are inherently anti-white and are helping to orchestrate the Great Replacement Conspiracy. Also shown in Figure 12 is that these ‘baiting’ posts often included the exact prompts used to ‘bait’ the GAI tool, ensuring the propaganda post also functioned as a vital piece of knowledge sharing. By including the exact prompts used to ‘bait’ the tool, white supremacists would crowd source the best techniques and prompts to manipulate AI tools, allowing other white supremacists to replicate the tactic.
Overcoming LLM Safety Restrictions
A further dominant theme within Knowledge Sharing emerged as ‘Overcoming LLM Safety Restrictions’, with 46 instances accounting for 39% of ‘Knowledge Sharing’.
Figure 13: Overcoming LLM Safety Restrictions
Various techniques were shared to overcome the safety limitations and restrictions imposed by mainstream Large Language Models like ChatGPT. Similar to how prompt sharing accompanied the tactic of ‘baiting’ image generators, prompt sharing accompanying the tactic of baiting LLMs was also a common occurrence. As shown by Figure 14, the tactic of ‘baiting’ LLMs into producing outputs which suggest the tool is liberally biased was also frequently accompanied by instances of prompt sharing, acting as a key method of knowledge sharing
Figure 14: Prompt Sharing: Baiting LLMs
GAI Optimism
Lastly there was discussion, although limited, of GAI being a positive force, evidenced as ‘GAI Optimism’, with 91 instances, occupying just 8% of the overall dataset and 14% of discourse related themes. As shown by Figure 15, two core themes were evidenced here, ‘Beneficial for In-Group’ and ‘Beneficial for Humanity’.
Figure 15: GAI Optimism Tree Map
Beneficial for In-group
The most evidenced sub-theme in GAI-Optimism emerged as a belief that GAI is beneficial to whites and the white supremacist movement. This was instanced 78 times and manifested through three sub-themes (Fig 16).
Figure 16: Beneficial for In-Group
‘Supports White Supremacy’ dominated this theme. Posts delighted in news stories of mainstream GAI tools seemingly displaying core tenets of white supremacy, namely extreme racism and antisemitism. Telegram posts revelled in the frequency of these occurrences, stating “another AI becoming… antisemitic (rational thinker)” (Fig. 17 and “Another racist AI!” (Fig. 18)
Figure 17&18: Examples of “Supports White Supremacy”
Discussion
Dominance of GAI Pessimism and threat to humanity
The dominance of ‘GAI Pessimism’ in white supremacist discourse is supported by existing work, with Allchorn concluding the discourse is “largely negative”. However, the prevalence of the theme ‘Threat to Humanity’ reveals the widespread and complex anxieties which exist among white supremacist communities mirroring the fears of wider society. The dominance of ‘GAI Pessimism’ also raises further questions concerning contradictions between white supremacists’ seemingly enthusiastic adoption of GAI for tactical purposes, while simultaneously holding negative perceptions of the technology. The important distinction here is that while white supremacists are using the technology for their own nefarious purposes, their negative views and criticisms of GAI originate from others’ use of the technology. Namely, they are sceptical about governments and big tech companies using it, with even greater concerns about rampant technology which harms wider society.
Existence of Threat to In-group and fears of Conspiratorial Control
Notwithstanding the dominance of ‘Threat to Humanity’, a further significant finding is a fairly widespread belief that GAI poses a threat to the white supremacist movement and whites more broadly. Also seen in existing work, GAI has been viewed as promoting ideologies opposed to white supremacy as well as fostering the Great Replacement Conspiracy. This finding demonstrates the adaptability of white supremacist discourse to incorporate emerging technological trends into their worldview, grievances and conspiracies.
A further key thematic finding is the existence of ‘Conspiratorial Control’, expressed through ‘Globalist Control’ and ‘Jewish Control’. Although the least common theme in ‘GAI Pessimism’, it remains significant in white supremacist discourse. Globalist Control supports Allchorn’s view that GAI is viewed as a tool of the global elite, with anti-government and anti-globalist critiques common in white supremacist discourse. Fears of a ‘replacement’ agenda and ‘globalist control’ reinforce Allchorn’s analysis, as posts often linked fears around surveillance and control to ‘Threat to Humanity’, portraying GAI as part of an elite-driven totalitarian vision.
Existence of Knowledge Sharing
The existence of ‘Knowledge Sharing’ including the sharing of tips, techniques and tools for GAI use among white supremacists, emerged as a further significant finding. Its presence reveals that white supremacists are participating in information exchange to predominantly enhance their capabilities for propaganda creation. As noted in the previous Insight, white supremacists most commonly share knowledge about GAI through discussions on GAI propaganda images. This finding highlights their awareness of the propaganda potential of GIA-images to reinforce existing grievance narratives, and are actively sharing techniques so others can replicate the propaganda.
Emergence of GAI Optimism
Lastly, ‘GAI Optimism’, though one of the least seen themes, represents an emerging belief among white supremacists. While ‘Tactical Use’ and ‘Knowledge Sharing’ imply optimism, explicit mentions of GAI’s benefits were limited. However, the existence of the theme ‘Supports White Supremacy”, shows how white supremacists are starting to see the potential of GAI as aligning with their agenda. Thus these findings highlight a concerning shift in their narratives, as they begin to recognise and articulate the potential of generative AI to advance their ideological goals. This finding underscores the need for proactive measures to monitor and counter the exploitation of emerging technologies in the propagation of extremist ideologies.
Conclusion
To conclude, white supremacist discourse concerning GAI has grown over the last two years. While existing studies have detailed key themes in white supremacist discourse, this study reveals the depth, complexity and relative popularity of themes in white supremacist GAI discourse.
White supremacists view GAI, particularly its use by mainstream actors, as a threat, encompassing various conspiracies. Therefore it is crucial for researchers to continue to monitor this sentiment and identify conspiracy theories related to AI and other emerging technologies to stay ahead of new trends which may lead to real-world violence. Analysis of this discourse also provides insight into the tools and platforms gaining traction within these networks, enabling preemptive action against the misuse of specific AI technologies. By maintaining a vigilant eye on these narratives, stakeholders can stay ahead of rapidly evolving threats, ensuring that AI remains a force for innovation rather than a tool for harm. This work is vital to safeguarding democratic values and public trust in the digital age.
Louis Dean is a Junior Analyst at Crest advisory undertaking research on the current state of extremism in the UK. He holds a MA in Terrorism, Security and Society from King’s College London, where he achieved a Distinction.