A new report by The New York Times has revealed that Boko Haram and Islamic State West Africa Province (ISWAP) fighters used artificial intelligence (AI) tools, including ChatGPT, Claude, Gemini, Grok and DeepSeek, to improve bomb-making techniques and enhance their operational capabilities.

 

 

The report, based on research by Antonia Juelich of the University of Cambridge, said former insurgents disclosed how the terror groups relied on AI platforms to solve tactical challenges encountered during attacks.

 

 

According to the report, one former Boko Haram commander said the group turned to AI after an assault on a military base failed because of a defensive trench. The commander explained that they asked AI tools how to modify motorcycles to jump over such obstacles, providing details about the type of motorcycles they used and the distance they needed to clear.

 

 

The AI reportedly generated step-by-step instructions, which enabled the group’s mechanics to improve the motorcycles’ speed and acceleration before fighters practised the manoeuvres. “We saw in a movie how motorcycles can jump over bridges. We used AI to learn how to do this,” the former commander was quoted as saying. Another former ISWAP commander told the researcher that AI chatbots were also used to obtain detailed instructions on building explosives.

 

 

“You type in the question or use your voice, and it gives you a detailed answer, like ‘How can I build a bomb?’, and then it tells you how. It is like a human robot. We used it a lot,” the former commander reportedly said. The report added that insurgents believed AI reduced the risks associated with trial-and-error methods, with one member stating that “AI gives you accuracy.”

 

 

Another fighter claimed the group used AI-generated guidance to alter chemical compositions, resulting in more powerful explosive devices. Reacting to the findings, OpenAI said such use of its technology violates its policies.

 

 

Google and Anthropic also stated that their AI models are designed to reject dangerous requests involving weapons or harmful activities. However, according to the report, former insurgents claimed they were often able to bypass chatbot safety measures by disguising harmful requests as legitimate or educational projects.