Boko Haram Used ChatGPT, Gemini and Claude to Plan Attacks

Boko Haram, the Nigerian terrorist organization affiliated with the Islamic State, spent years systematically using popular AI chatbots to plan attacks, repair weapons and train new fighters. That is the finding of a report by Cambridge researchers, which The New York Times obtained access to ahead of publication.

The report's author spent two years gathering material, reaching out to former members of the organization in Nigeria. Her findings show that the use of AI was neither accidental nor incidental. Boko Haram set up dedicated cells for the purpose, ran internal training sessions and shared the knowledge gained among fighters, pointing to an organized approach to adopting new tools.

How it worked in practice

According to former fighters' accounts, chatbots were used mainly to obtain technical information the organization lacked internally: on military tactics, weapons repair and maintenance, and the construction of explosive devices. AI effectively became a substitute for a military expert, available to anyone with a phone and an internet connection.

One case described in the report involves an attack on a military base that stalled because of a defensive trench dug around the facility. Former Boko Haram commanders told the researcher that the organization turned to AI tools for help, which suggested a specific modification to motorcycles so they could cross this type of terrain obstacle during the assault.

Bypassing safety filters

All the systems named in the report theoretically have built-in mechanisms that block queries related to violence, weapons or explosives. According to the fighters' accounts, however, they were able to reliably trick these filters by phrasing questions in a seemingly innocent or technical way, allowing them to get around automated content moderation systems.

AI companies respond

The New York Times asked all the companies whose tools appear in the report for comment. The responses varied widely.

We know bad actors will never stop trying to abuse our tools, and we will continue strengthening our safeguards in response - Drew Pusateri, OpenAI spokesperson

Spokespeople for Anthropic and Google limited themselves to assurances that their systems have effective and rigorous safety mechanisms, without directly addressing the specific cases described in the report. xAI and DeepSeek declined to comment at all. OpenAI and Meta stressed that using their tools for terrorist activity violates their terms of service, and said they continue to improve their filters.

What it means for the industry

The report lands at a moment when AI companies have spent months telling regulators and the public that their models are becoming safer thanks to successive layers of content moderation. Evidence that a well-organized armed group systematically bypassed these safeguards for years undercuts those assurances and shows that filters designed to catch single, obvious queries struggle against persistent, methodical probing by trained users.

For European regulators, including the bodies responsible for implementing the EU's AI Act, the case is a significant argument in the debate over mandatory safety testing of high-risk models before they are released to the public. The Boko Haram case offers a concrete, documented scenario of misuse, not just the theoretical risk that AI safety organizations have long pointed to.

The report's authors note that most of the documented incidents date to before the end of 2024, and that newer versions of the models have stricter mechanisms for refusing to answer potentially dangerous queries. That does not change the fact that once a terrorist organization acquired knowledge of effective ways to bypass safeguards, it could pass that knowledge on, regardless of how much the filters in subsequent generations of tools have since been tightened.

Sources: Terrorists Used ChatGPT and Gemini. New Report Shows Scale of the Problem (imagazine.pl), How Boko Haram Secretly Uses AI to Build Bombs, Improve Deadly Attacks (politicsnigeria.com), Terrorist groups are using every major AI chatbot for attack planning and weapons development (the-decoder.com)