Boko Haram Used AI Chatbots to Plan Military Base Attack in Nigeria
Members of Boko Haram reportedly used AI chatbots including ChatGPT and Claude to receive tactical guidance for attacking a Nigerian military base, circumventing AI safety measures to obtain instructions on modifying motorcycles to breach defensive trenches. Research by a Cambridge University terrorism specialist based on interviews with 27 former group members found that both Boko Haram factions had institutionalized AI use for combat planning, weapons troubleshooting, and explosives design.
What happened
Boko Haram members attacked a military base in eastern Nigeria and used AI chatbots including ChatGPT, Claude, Gemini, and Grok to plan the assault. According to a former Boko Haram commander, the group consulted AI after initial attempts were blocked by defensive trenches, providing the system with information about their motorcycles, the distance needed to jump, and receiving step-by-step instructions in response.
Mechanics modified motorcycles based on chatbot guidance to increase acceleration and top speed. Riders practiced jumps by digging holes filled with broken glass and fire, with some fatal outcomes during training, before executing the attack.
Reported impact
A military base in eastern Nigeria was successfully breached by Boko Haram using AI-assisted tactical planning. The attack was enabled by circumventing safety protocols in generative AI tools through persistent questioning.
Index analysis
Research by Dr. Antonia Juelich at Cambridge University based on 27 interviews with former Boko Haram members found that both factions of Boko Haram used frontier AI systems institutionalized through specialized units and internal training. Large-language models were consulted at every stage of military activity including mission preparation, operations, and post-mission analysis, beyond the propaganda-focused AI use previously documented.
Governance controls
AI chatbots have built-in limitations intended to prevent users from soliciting information that could cause harm, but researchers have found that people can circumvent safety protocols by slowly but persistently coaxing models into divulging restricted information.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.