Boko Haram AI use tied to wider islamic state network

Members of Boko Haram‘s two main factions have been using artificial intelligence, including ChatGPT, Claude, Grok, and China’s DeepSeek, as early as 2023 to support terrorist operations, according to a new University of Cambridge report based on interviews with former militants. Experts warn that other extremist groups may be doing the same.

The report, which is based on interviews with more than 20 former members of Boko Haram‘s JAS and ISWAP factions, has shed light on AI’s use in extremists’ planning operations and in improving explosive devices and bombs, among other activities, according to France 24 on July 14^th.

Boko Haram is an Islamist militant group based in northeastern Nigeria. It split into two main factions in 2016: Jama’atu Ahlis Sunna Lidda’awati wal-Jihad (JAS) and Islamic State West Africa Province (ISWAP), the latter of which pledged allegiance to the Islamic State group.

Researchers previously believed, based on internet forums or Telegram groups, that extremist groups primarily used AI to generate disinformation, propaganda and recruitment material. The new interviews suggest their use extends to instructions on handling newly seized weapons, improving drone-delivered explosives and planning future raids.

The report also reveals that some Boko Haram factions have dedicated members who are trained by specialists linked to the Islamic State’s wider network across Iraq, Afghanistan, North Africa and other regions.

They train them to formulate requests that elicit the most accurate and useful answers from Large Language Models (LLMs). A common technique, called “jailbreaking,” guides users on how to phrase questions to circumvent an AI’s built-in ethical restrictions, such as those that prevent it from providing instructions for making a pipe bomb.

Antonia Juelich, a researcher at Cambridge University and the author of the report, stated that it is becoming more common to use it “not just for propaganda, but for basically an entire attack chain.”

Graig Klein, a specialist in international terrorism at Leiden University, said the interviews reinforce his own findings on jihadist groups’ use of AI.

In line with Juelich’s conclusion, Klein argued it is less likely to lead to new forms of terrorism, attacks, or weapons. Instead, it primarily acts as a force multiplier, enhancing the capabilities of these groups.

This is illustrated in the report, which cites a former ISWAP member who stated, “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 send 20. We learned more about well-coordinated attacks and deployment of smaller units.”

Klein stated that “what AI enables us to do is trial and error without necessarily having to try,” and further explained that the data can be leveraged to rethink our actions, processing it significantly faster.

He further contended that Boko Haram probably isn’t the only terrorist organisation employing AI to enhance its combat strategies.

“There’s no reason to think any other groups or ideologies would use this any differently,” he said.

Juelich echoed this, explaining that since ISWAP members were trained by specialists linked to the larger IS group, it was likely these experts were also working with other branches globally.

“The findings do suggest that this is part of a broader training and upskilling programme that is pushed by ISIS,” she said. “And that’s why I think it’s very unlikely that this has only happened in Nigeria.”

Klein, in line with many experts, also stated that the safeguards AI companies have established to prevent their services from being used for violent purposes are not up to the task.

Klein said AI was unlikely to enable entirely new forms of terrorism, as developing chemical, biological or nuclear weapons still requires difficult-to-obtain materials. Juelich, however, said the possibility of AI lowering barriers to such capabilities is something she took “very seriously.”

OpenAI, Anthropic, Google, xAI and DeepSeek did not respond to requests for comment. In a statement to SCMP, Meta said the study concerned older versions of its AI models rather than its latest model, Muse Spark, which the company said includes additional safety safeguards introduced earlier this year.

France 24 and agencies, Maghrebi.org