AI Used for Bombs
University of Cambridge Research Reveals ISIL Affiliates Use Major AI Chatbots for Bomb Development
Anonymous
Key Points
1
The chatbots identified in the Cambridge interviews included OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, xAI’s Grok, Meta AI and China’s DeepSeek.
2
A former ISWAP commander described the systems as exceptionally useful, saying: “There is nothing we haven’t received an answer on.”
3
The training operation involved operatives traveling to ISWAP-controlled territory to teach senior members how to use AI, indicating that access was being institutionalized rather than limited to individual fighters.
4
Tech Against Terrorism researchers said they infiltrated a pro-ISIL technical forum, where supporters exchanged guidance on using AI and circumventing chatbot safety restrictions.
5
United Nations officials warned the Security Council in February about the increased use of advanced AI by ISIL and its affiliates, providing a specific international-security context for the findings.
Terrorist affiliates linked to ISIL have systematically used AI chatbots from major technology companies to develop bombs, plan attacks, and solve battlefield problems, according to new research from the University of Cambridge. Cambridge University interviewed 27 former members of Boko Haram's ISWAP and JAS factions who described an organized training system where operatives taught senior fighters how to use the systems. A former ISWAP commander said of the AI tools: "There is nothing we haven't received an answer on."
The findings have alarmed the United Nations, technology leaders, and researchers who warn that AI capabilities are advancing faster than safety protections. Al Jazeera obtained separate material showing ISIL supporters sharing methods to bypass chatbot safety controls. The systems involved include ChatGPT, Claude, Gemini, Grok, Meta AI, and DeepSeek.
How Terrorist Groups Trained on AI Systems
The operation was far more sophisticated than individual fighters experimenting with chatbots. Cambridge researchers found that operatives traveled to ISWAP-controlled territory to conduct formal training sessions for senior members. This institutionalization meant AI access was built into the group's command structure, not left to chance.
Former fighters described using the tools for bomb-making queries, modifying commercial drones for attacks, and tactical planning. The systems provided step-by-step guidance without requiring users to possess technical expertise. One fighter said the AI answered questions no human expert would answer.
Tech Companies Face Pressure Over Safety Failures
The chatbots identified included OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini, xAI's Grok, Meta AI, and China's DeepSeek. All companies built safety restrictions meant to block weapons-related requests. Yet Al Jazeera found that ISIL supporters on technical forums routinely shared techniques to circumvent these guardrails.
Tech Against Terrorism researchers infiltrated a pro-ISIL forum and documented how supporters exchanged detailed guidance on using AI and bypassing restrictions. The techniques were simple enough that less-skilled users could replicate them. This suggests the safety systems are fundamentally inadequate.
UN Warnings Highlight AI Security Crisis
United Nations officials warned the Security Council in February that ISIL and affiliates are increasingly using advanced AI systems. The Cambridge study provides concrete evidence of what the UN suspected: terrorists now have systematic access to weapons development and attack planning tools. Researchers say AI advancement is outpacing safety protections by years.
The gap between capability and safeguards creates urgent risk. A single chatbot can serve hundreds of thousands of users worldwide. Terrorist groups exploit this scale while companies struggle to identify and block all misuse. Experts warn the problem will worsen unless restrictions are dramatically strengthened.
Sentiment
Mostly NegativeNEGATIVENEUTRALPOSITIVE
Publishers
17
Articles
12
Reach
29
A
AnonymousAuthor
