Boko Haram factions have embedded artificial intelligence into their operations much faster, more extensively, and more systematically than prior analysis has suggested, according to a first-of-its-kind field study on AI adoption by an active terrorist organisation.
The study, conducted by Dr Antonia Juelich and published by the Cambridge Programme on AI Science & Policy, is based on 57 face-to-face interviews with 27 former Boko Haram members in northeast Nigeria. It’s found that both major factions were using AI across weapons support, tactical planning, operational security and post-mission review – a picture sharply at odds with the prevailing view that jihadist AI use is slow and largely limited to propaganda.
Former members said the tools included ChatGPT, Claude, Gemini, Grok, Meta AI and DeepSeek. They described AI being used to help design explosive devices, identify seized military hardware, troubleshoot weapons, advise on battlefield tactics and review attacks afterwards.
Training and dedicated units turned that use into organised practice. Selected commanders were trained by outside jihadist contacts, including Islamic State-linked operatives, with support including dedicated laptops, account set-up, practical instruction and follow-up assistance on AI use.
Former members also said trained users could work around platform safeguards designed to stop AI tools from giving harmful guidance on weapons, explosives and attack planning. The findings show current safeguards have failed to stop organised terrorist misuse consistently, raising urgent questions about whether they hold against trained networks with technical support and multiple accounts.
The study also found strong enthusiasm for AI inside the group, and some former members expressed openness to mass-casualty weapons, although the group’s current use of AI remains conventional. One former commander said: “God has helped us, and so will AI.”
Dr Antonia Juelich, International Security Lead at the Cambridge Programme on AI Science & Policy, and Associate Fellow at the Leverhulme Centre for the Future of Intelligence, said:
“The findings show that adoption of AI by terrorist groups has been significantly underestimated in both scope and character. It’s time to stop treating this as a hypothetical future risk and start treating it as a present and growing national security threat.
“A safeguard is only a safeguard if it survives contact with the real threat. Governments, AI companies and researchers need to move faster, share evidence sooner, and test these systems against trained groups that share methods, switch accounts and keep coming back.”
Detailed report findings:
-
AI is being used before, during and after operations
Public evidence on terrorist use of AI has largely focused on what can be seen online, including propaganda and extremist media. Dr Juelich’s research finds AI being used inside the practical work of armed operations.
Former members described turning to AI at every stage of an operation. One said: “Anytime they did not have a clear understanding of something, they would ask the AI.”
Interviewees said AI tools were used before attacks for weapons support, including identifying seized military hardware, troubleshooting weapons and refining explosive devices. One former member said: “Before, the bomb explosion was not that big, but then they studied it. AI told us what chemicals to put in that made the explosion heavier.”
Former members also gave accounts of AI being used during operations to advise on tactical adjustments, and after attacks to review what had gone wrong and suggest improvements.
One former fighter said AI had changed how the group thought about deployment: “We sent 200 fighters because we had a lot of strength, but then 60 got killed. With the help of AI, we learned that it sometimes makes sense to only send 20. We learned more about well-coordinated attacks and deployment of smaller units.”
The study does not measure exactly how much AI improved Boko Haram’s capabilities, or claim that AI enabled attacks that would otherwise not have been possible. But former members repeatedly described practical gains, including more powerful explosives and better-coordinated attacks using smaller units.
-
AI know-how is moving through jihadist networks
The study found foreign jihadist operatives came to Boko Haram camps to train selected commanders in AI use.
Former members of Islamic State West Africa Province, the Boko Haram faction aligned with Islamic State, described repeated in-person training sessions delivered by foreign operatives between 2023 and 2025. One reported session brought together around 30 to 50 leaders and selected fighters from across the faction’s territory.
The accounts described practical sessions in which trainers provided laptops and explained to commanders how to prompt and jailbreak.
Former members of Jamā’at Ahl as-Sunnah lid-Da’wah wa’l-Jihād, the rival Boko Haram faction, also said they received AI training through different jihadist contacts.
Dr Juelich says the findings suggest AI know-how is moving through the same routes that have long carried weapons skills, drone knowledge and battlefield tactics between armed groups.
-
Boko Haram factions have built dedicated AI units
Former members said both major Boko Haram factions had created specialist AI units, sitting behind the frontline. Their role is to query AI systems, turn AI outputs into advice for the chain of command, and train commanders.
The units also managed accounts and paid subscriptions across several platforms, with help from Islamic State-linked contacts outside Nigeria.
Access was tightly controlled. Lower-ranking fighters generally could not use the tools themselves. Questions were passed up to commanders or AI unit members, who used the systems and fed answers back down.
The findings suggest AI has become part of the factions’ working machinery, with people, devices, payments and access rules built around it. That makes the capability more durable than the knowledge of any single member.
-
Platform safeguards were treated as obstacles
Former members said AI tools sometimes refused sensitive requests, including on weapons, explosives, military information and attack planning. They described those refusals as manageable.
Selected users were trained to work around restrictions, seek help from external trainers when systems refused to answer and move between accounts if one was blocked or became risky to use. One former commander said: “Boys that have received extensive training bypass the restrictions.”
Some of what the group asked of AI was general or dual-use information. But the accounts also included requests on explosives design, weapons use and attack planning, exactly the kinds of harmful guidance that platform safeguards are designed to block.
The study does not assess whether the same tactics would work against the latest safeguards. However, it shows that, during the period covered by the research, organised terrorist networks managed to find ways around platform restrictions.
-
Findings raise urgent questions for governments and AI companies
Dr Juelich says the findings should force a reset in how governments and AI labs assess terrorist use of AI. For AI developers, that means ensuring that safeguards hold against organised misuse: users who share methods, receive coaching, try again after refusals and move between accounts.
For governments and security agencies, it means treating terrorist AI adoption as a present security problem. Key questions now include how far the practice has spread, which groups have received training, and what more capable models could do in the hands of terrorist organisations.
