Boko Haram using ChatGPT, other AI tools to plan attacks, Cambridge study finds

A new report from the Cambridge Programme on AI Science & Policy has revealed that Boko Haram fighters in Nigeria are actively using frontier AI models — including ChatGPT, Claude, Gemini, and Grok — to plan attacks, design explosive devices, and improve operational security. The findings, based on interviews with members of the group, represent some of the most direct evidence yet that terrorist organizations have moved beyond using AI for propaganda and into operational planning.dw+2

A Growing Threat

The Cambridge research, released this week, found that Boko Haram factions have embedded AI into their operations more extensively and systematically than previously understood. Interviewees described using the chatbots to troubleshoot weapons, service equipment, and coordinate logistics — functions that once required specialized human expertise or physical training camps.the-decoder+2

The study arrives alongside a separate benchmark report from Tech Against Terrorism, a UN-backed watchdog, which tested 27 AI models with more than 2,300 queries drawn from real terrorist use cases. The results found that 32% of queries produced "genuinely usable" information. When the same prompts were reframed as academic research, the compliance rate rose to 42%. Full refusals accounted for 57% of responses, while 15% were classified as "hedged compliance" — meaning the model initially refused but still supplied content.dw+2

The most alarming failures came from open-source models whose safety training had been stripped through a process called "abliteration," with two such models complying with 89% and 100% of requests.themedialine

From Propaganda to Operations

Experts say the findings mark a shift. For the past several years, groups like the Islamic State and al-Qaeda primarily used AI for generating propaganda — videos, memes, and recruitment content. But 2025 saw a documented surge in AI-assisted operational planning across plots in the United States, Canada, Israel, Finland, France, and Austria.smallwarsjournal+1

"A determined person will eventually find most information," said Adam Hadley, director of Tech Against Terrorism. "But what these models change is speed, ease and comprehensiveness. People who previously lacked time, resources or ability can now get much further, much faster."dw

Calls for Action

Tech Against Terrorism has urged AI developers to treat terrorist misuse as a distinct safety category, extend refusal training beyond the most recognizable threats, and track the distribution of de-restricted open models as a national security concern. Hadley warned that with radicalization increasingly affecting teenagers across Europe and the United States, "it is only a matter of time before chatbots become a significant part of the problem."thenationalnews+1