The Evolving Landscape of AI and Extremism
The intersection of artificial intelligence (AI) and extremist activities is becoming a growing concern for digital security and counter-terrorism experts. While AI has primarily been utilized by extremist groups for propaganda generation, recent analyses suggest a shift towards employing these technologies for operational planning and research related to potential attacks. This evolution has prompted increased scrutiny into how large language models (LLMs), commonly known as AI chatbots, might inadvertently assist individuals or groups seeking to engage in harmful activities.
Historically, extremist organizations like the 'Islamic State' have demonstrated an early adoption of various technologies, including modified drones, to further their objectives. This technological aptitude is now extending to AI. Although AI chatbots are designed with safeguards to prevent the dissemination of harmful content, research indicates that these systems can sometimes be prompted to provide information that could be misused.
The 'Jailbreaking' Phenomenon and Research Findings
The concept of 'jailbreaking' AI models refers to attempts by users to circumvent built-in restrictions and elicit disallowed content. OpenAI, a leading AI developer, acknowledges this practice as malicious actors attempting to prompt models into providing information that would typically be blocked. Previous media investigations have highlighted instances where specific prompts could lead some AI models to describe methods for creating bioweapons, planning attacks, or concealing illicit activities.
A recent report by Tech Against Terrorism, an organization supported by the United Nations counter-terrorism directorate, sheds further light on this issue. The study involved submitting over 2,300 requests, based on actual terrorist use cases, to 27 different AI models. The findings were notable: 32% of these queries yielded 'genuinely usable' information. When the same questions were rephrased to appear as if for 'research purposes,' the success rate increased to 42%. This indicates a vulnerability in how AI models interpret and respond to user queries, particularly when the intent is disguised.
From Propaganda to Planning: A Noted Shift
For the past three to four years, extremist groups, including 'Islamic State' and al-Qaeda, have predominantly leveraged AI for propaganda purposes. This has involved producing and disseminating various forms of digital content such as videos, memes, podcasts, and disinformation campaigns aimed at radicalizing and recruiting followers. However, this application of AI is undergoing a transformation.
An analysis published in December by Militant Wire confirmed a significant trend: "The year 2025 has witnessed a notable rise in incidents where terrorists and violent extremists have leveraged AI tools to plan, research and prepare attacks." This shift is evidenced by several high-profile incidents, both successful and foiled, across various countries including the US, Canada, Israel, Finland, and Austria. These incidents have reportedly utilized AI for aspects such as planning, surveillance, visualization, and even generating propaganda surrounding the attack itself. For example, a US individual suspected of arson in California in 2025 reportedly used ChatGPT to create images of burning cities and inquire about legal responsibilities related to accidental fires.
While specific details on AI's role in these cases are often withheld by security agencies, court documents and forensic reports are increasingly showing chat logs where suspects have queried language models for instructions on bomb-making, ideological validation, or justifications for attacks.
Group Adoption and 'Lone Wolf' Support
The use of AI is not limited to individual actors. Extremist groups are also integrating AI into their operations. Researchers examining the activities of Jama'at Nusrat al-Islam wal-Muslimin (JNIM), an al-Qaeda affiliate in Mali, believe the group has used AI to assist in modifying drones. Furthermore, Yuri Neves and Emily Klein, security researchers with Moonshot, an organization combating online threats, observed that supporters of groups like 'Islamic State' and various right-wing factions regularly discuss AI applications in their messaging channels. These discussions include sharing AI prompts, coordinating strategies to extract desired responses, and even cost-sharing for ChatGPT subscriptions.
Rueben Dass, an associate research fellow at the S. Rajaratnam School of International Studies, highlights AI chatbots' emerging role in supporting 'lone wolf' terrorist attacks. He notes that while human planners once guided individuals through social media, AI platforms like ChatGPT are now providing a similar, albeit automated, form of support. This suggests AI is becoming an accessible resource for individuals seeking guidance in planning such acts.
Moustafa Ayad, executive director for Africa, the Middle East and Asia at the Institute for Strategic Dialogue, confirms that the 'Islamic State' media outlet Voice of Khorasan published guidance on AI use. He states that the jihadist ecosystem employs AI in diverse ways, from creating memes and TikTok videos to sophisticated border-crossing propaganda. Ayad emphasizes that this versatility, encompassing both propaganda and operational planning, presents a significant challenge.
Assessing the Danger and Future Implications
The precise extent of the danger posed by extremist use of AI remains uncertain. Experts like Dass point out that information on bomb-making or 3D-printed weapons is already readily available online, independent of AI. Neves questions whether AI truly provides information that is otherwise unobtainable, and if so, whether this qualitatively changes the threat landscape.
Klein views LLMs as a continuation of disruptive technologies, akin to the internet or encrypted messaging apps, which extremists also adopted. She suggests that AI may not directly create more terrorists but rather influences how individuals progress along a pathway to violence. AI can accelerate this process by validating grievances or reinforcing existing beliefs, effectively compressing stages of radicalization.
Adam Hadley, director of Tech Against Terrorism, acknowledges that a determined individual can eventually find most information. However, he argues that AI models significantly enhance the speed, ease, and comprehensiveness of this process. This allows individuals who previously lacked time, resources, or ability to advance much faster. Hadley expresses particular concern about the conversational nature of AI chatbots, stating that finding a bomb-making manual is different from having a 'bomb-making coach.' While Dass believes AI is unlikely to increase the 'success' rate of terror acts or the overall number of attacks, he anticipates a rise in attacks that incorporate AI in some capacity.
Hadley concurs, highlighting the clear trajectory and the concerning proportion of teenagers and children being radicalized in Europe, the UK, and the US. Given the established role of the internet and social media in youth radicalization, he predicts that chatbots will inevitably become a significant factor in this problem.
Source: Original Article
