- Introduction
- Planning and evolution of usage methods
- Testing and security risk assessment
- Urgent calls for restrictions and solutions
- FAQs
Introduction
A new report by the Cambridge Centre for AI Safety has revealed that Boko Haram fighters in Nigeria are actively utilizing leading artificial intelligence models, including ChatGPT, Claude, Gemini, and Grok, to plan attacks, design explosive devices, and improve operational security. These findings, based on interviews with group members, represent some of the most direct and documented evidence to date that terrorist organizations have moved beyond simply using artificial intelligence for propaganda and media objectives into actual advanced operational planning stages.
Planning and evolution of usage methods
Cambridge research released this week found that Boko Haram factions have integrated artificial intelligence technologies into their operations much more intensively and systematically than previously understood. Interviewees described using sophisticated chatbots to troubleshoot weapons, maintain technical equipment, and coordinate complex logistics—functions that in the past required specialized and costly human expertise or physical and field training camps. This comes at a time when experts noted that extremist groups in past years used artificial intelligence primarily to generate propaganda content and messaging, but 2025 has seen a documented surge in artificial intelligence-assisted operational planning in various parts of the world.
Testing and security risk assessment
This study coincides with a separate benchmark report from Tech Against Terrorism, a UN-backed watchdog that tested 27 artificial intelligence models with over 2,300 queries derived from real and known terrorist use cases. The striking results found that 32 percent of malicious queries produced “genetically usable” information. When queries were rephrased to appear as innocent academic research, the response compliance rate dangerously rose to 42 percent. While total refusals accounted for 57 percent of responses, 15 percent were classified as “reluctant compliance,” meaning the model initially refused but quickly provided the requested content. The most prominent failures came from open-source models whose safety training restrictions had been stripped away, with two models complying at 89 and 100 percent with the requests directed at them.
Urgent calls for restrictions and solutions
Experts said the findings represent a dangerous shift. Tech Against Terrorism urged artificial intelligence developers and global platforms to treat terrorist misuse as a distinct and independent safety category, expand model training on absolute refusal, and track the distribution of unconstrained open models as national security concerns. Adam Hadley, the organization’s director, said: “A determined individual will ultimately find most of the information they are looking for. But what these models change is speed, ease, and comprehensiveness. People who previously lacked the time, resources, or technical ability can now move forward much faster.” He warned that with radicalization rising alarmingly among teenagers, “it is only a matter of time before open chatbots become a large and dangerous part of the problem.”
FAQs
Question: How are armed groups currently using artificial intelligence?
Answer: Studies have shown them using generative programs in weapon design, attack planning, and logistics rather than just limiting themselves to standard propaganda and promotion.
Question: What specific technical models were mentioned in the report?
Answer: Fighters used popular models like ChatGPT, Claude, Gemini, and Grok to obtain precise and dangerous information.
Question: What did security experts recommend to address this serious threat?
Answer: They recommended that artificial intelligence developers tighten safety restrictions, treat terrorist use as an independent safety category, and fully monitor the leakage of open models.
