One former ISWAP commander told researchers there was nothing he hadn’t received an answer on. He was talking about ChatGPT. That single quote, buried inside a University of Cambridge study, tells you more about the real-world consequences of AI safety failures than any corporate transparency report ever has.
According to Daily Post, the Cambridge study is based on interviews with 27 former members of Boko Haram’s two main factions, the Islamic State West Africa Province (ISWAP) and Jama’at Ahl as-Sunnah lid-Da’wah wa’l-Jihad (JAS). Several of those former fighters said they used ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeek directly in connection with weapons and combat operations. The research focused primarily on activity during 2023 and 2024.
The picture that emerges is not some vague, theoretical misuse. It is organized and deliberate. Former fighters described groups creating dedicated AI units, staffed with engineers, weapons experts, and bomb-makers. One former JAS commander described a 23-member team operating out of a solar-powered room, answering questions, analyzing information, and training senior commanders on how to extract useful responses from AI systems. A former Boko Haram commander described consulting AI while actively making explosives, calling the technology “like a human robot.”
Study author Antonia Juelich told Al Jazeera that around 10 of the 27 people she interviewed described using AI specifically in connection with explosives. Her key concern is precise: fighters weren’t just asking generic questions. They were feeding AI systems details about the specific materials they had available and the specific problems they faced, receiving responses tailored to their exact situation. That’s not a chatbot slipping up. That’s the AI working exactly as designed, for someone with very bad intentions.
The documented uses are wide-ranging:
- Seeking instructions on recovering explosive materials from unexploded military ordnance
- Modifying commercially available drones to carry explosives
- Uploading battlefield images to ChatGPT for real-time tactical analysis during combat
- Analyzing failed attacks to understand what went wrong and how to improve
- Learning how to operate newly acquired, unfamiliar firearms
Every major AI company named in the study prohibits use of its systems for terrorism, violence, and weapons development. But prohibition in a terms-of-service document is not the same as prevention in practice. Former fighters told researchers they were trained to find ways around the safeguards specifically designed to stop these conversations. Separately, researchers at Tech Against Terrorism found active discussions among ISIL supporters about bypassing AI safety controls entirely, so-called jailbreaking. Lucas Webber of that organization warned that what researchers have found is probably just a fraction of the actual activity.
The broader context here matters. AI companies have spent years insisting their safety measures are serious and improving. But this study covers 2023 and 2024, a period when those safety claims were already being made loudly and publicly. Juelich herself noted that AI has developed considerably since then, and said plainly: “I worry about where this is heading.” So should the rest of us.