Editorial illustration for Terrorist Groups Use Major AI Chatbots for Attack Planning, Weapons Development
ISIS Trained Boko Haram to Use AI for Attacks
Terrorist Groups Use Major AI Chatbots for Attack Planning, Weapons Development
Boko Haram has an AI training program, and ISIS built it. That's the core finding in a new study from Antonia Jülich, a researcher with the Cambridge Programme on AI Science & Policy, who spent time interviewing 27 former members of Boko Haram across 57 separate conversations. Her report traces how ISIS began teaching prompt engineering and jailbreak techniques as far back as 2023, then passed that knowledge to Boko Haram commanders operating in Nigeria.
The chatbots involved aren't obscure or underground tools. Jülich's interviews point to ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeek, mainstream products built by the biggest names in the industry. Both major Boko Haram factions have gone further than casual use, standing up dedicated units tasked specifically with exploring what these systems can do for the group's operations, from planning attacks to maintaining weapons stockpiles.
What separates this from earlier reporting on extremist groups experimenting with technology is the level of organization Jülich describes: structured training sessions, projector screens, assembled specialists. What follows is her account, in her own words, of how one faction pulled that knowledge into a room and showed commanders exactly how it worked.
Terrorist groups are using every major AI chatbot for attack planning and weapons development ISIS has reportedly been offering prompt engineering and jailbreak training since 2023 and has trained Boko Haram commanders in Nigeria on how to bypass AI safety filters.
Why this matters
This is the clearest evidence yet that safety filters are a speed bump, not a wall, for anyone with time and motivation to route around them. ISIS reportedly built a training pipeline for jailbreaking as far back as 2023, and Boko Haram didn't just adopt one chatbot, it stood up dedicated units covering six major platforms: ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeek. That spread tells us the vulnerabilities Jülich documented aren't quirks of one company's guardrails; they're shared weaknesses across the industry's approach to content moderation.
For developers and safety teams, the lesson is that red-teaming against a curious teenager is a different problem than red-teaming against a trained adversary with institutional knowledge passed between commanders. For founders building on these models, it's a reminder that "safety filter" is a marketing term as much as a technical one, and buyers should ask what happens when someone actually tries to break it. Researchers now have a concrete, cited case study instead of a hypothetical.
The next question is whether providers respond with real architectural fixes or another round of patched prompts.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
From Archives OpenAI Targets Families as ChatGPT Adapts for Household Use
From Archives Meta's Muse Spark 1.1 coding score hits 71.3, edges past GLM-5.2
Common Questions Answered
How did ISIS train Boko Haram in prompt engineering and jailbreak techniques?
According to researcher Antonia Jülich's study, ISIS began teaching prompt engineering and jailbreak techniques to Boko Haram commanders as far back as 2023, establishing a formal training pipeline to help them bypass AI safety filters. This knowledge transfer enabled Boko Haram to systematically circumvent security measures across multiple AI platforms.
Which AI chatbots has Boko Haram reportedly set up dedicated units to access?
Boko Haram established dedicated units covering six major AI platforms: ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeep. This widespread adoption across multiple platforms demonstrates that the vulnerabilities are not isolated to a single company's guardrails but represent systemic weaknesses across the AI industry.
What research methodology did Antonia Jülich use to document terrorist groups' AI chatbot usage?
Jülich, a researcher with the Cambridge Programme on AI Science & Policy, conducted interviews with 27 former members of Boko Haram across 57 separate conversations to trace how terrorist groups learned to use AI systems for attack planning and weapons development. Her report provides direct evidence from former members about the training and implementation of these techniques.
What does the study reveal about the effectiveness of AI safety filters against motivated actors?
The research demonstrates that AI safety filters function as a speed bump rather than a wall for determined adversaries with sufficient time and motivation. The documented jailbreaking training pipeline shows that organized groups can systematically learn to route around these protections across multiple platforms.
Further Reading
- Exploitation of Generative AI by Terrorist Groups - International Centre for Counter-Terrorism (ICCT)
- Countering terrorist propaganda in the age of AI - Atlantic Council
- Technology Evolves the Tactics: Preparing for the Rise of Terrorist AI Harms - CREST Research
- Generating Terror: The Risks of Generative AI Exploitation - Combating Terrorism Center at West Point
- Will Generative AI Fundamentally Change Terrorist Threats? - Combating Terrorism Center at West Point
