Researchers at Tech Against Terrorism, an online watchdog supported by the United Nations counter-terrorism directorate, have published findings showing how frequently AI models provide dangerous information to would-be extremists.
The organization sent more than 2,300 requests for information drawn from “real terrorist use cases” to 27 different AI models, finding that 32% of queries produced “genuinely usable” results.
When the same queries were reframed as being for research purposes, that figure climbed to 42%, highlighting how easily guardrails can be bypassed through simple changes in framing.
This practice, known as “jailbreaking,” is described by OpenAI, the maker of ChatGPT, as “attempts by a malicious actor to prompt the model into providing disallowed content.”
Over the past three to four years, extremist groups including the Islamic State and al-Qaeda have primarily used AI to generate propaganda, producing videos, memes, podcasts and disinformation to radicalize followers.
Experts at the publication Militant Wire confirmed in a December analysis that “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.”
Headline-making attacks and several foiled plots across the US, Canada, Israel, Finland, France and Austria involved AI for planning, surveillance, visualization and propaganda.
Cambridge University released research featuring interviews with Boko Haram members in Nigeria, detailing how the group used AI models including ChatGPT, Claude, Gemini and Grok to plan attacks, design explosive devices and improve operational security.
Security researchers Yuri Neves and Emily Klein of Moonshot, writing for the Global Network on Extremism and Technology, observed extremists on Telegram “sharing AI prompts and conversation links, coordinating strategies to extract desired responses from chatbots, and cost-sharing ChatGPT subscriptions.”
Rueben Dass, an associate research fellow at the S. Rajaratnam School of International Studies in Singapore, says lone actors have shifted toward AI to receive the motivational support once provided by human planners in conflict zones.
“Previously you had this whole concept of virtual planners, where you had individuals sitting in conflict zones, who were reaching out to people on social media, trying to motivate them to carry out attacks,” Dass told DW.
Moustafa Ayad, executive director for Africa, the Middle East and Asia at the Institute for Strategic Dialogue, confirmed that the Islamic State media outlet Voice of Khorasan published guidance on using AI last year.
“Then you also have a dedicated set attempting to jailbreak AI, use it to support operational planning and readiness,” Ayad said, adding that “AI may be streamlining and supporting propaganda processes and simultaneously supporting operational planning and readiness.”
Emily Klein of Moonshot argues that AI is best understood as a continuation of earlier disruptive technologies, noting that the internet and encrypted messaging apps were also adopted by extremist actors before AI.
“So there isn’t necessarily evidence that AI is causally creating more terrorists,” Klein told DW, arguing instead that AI compresses stages of the pathway to violence by validating grievances and encouraging existing beliefs.
Adam Hadley, director of Tech Against Terrorism, acknowledged that a determined person will eventually find most information, but warned that AI changes “speed, ease and comprehensiveness” for those who previously lacked the time or resources.
“It’s one thing to find a bomb-making manual, it is quite another to have a bomb-making coach,” Hadley said, pointing to the conversational nature of AI chatbots as a particularly alarming development.
Dass cautioned that while AI may accelerate access to information, it is unlikely to make acts of terror more operationally successful, though more attacks will probably involve AI in some capacity going forward.
Hadley noted that a large proportion of those being radicalized in Europe, the UK and the US are teenagers or children, warning that “it is only a matter of time before chatbots become a significant part of the problem.”
