AI wasn’t supposed to help with this
Artificial intelligence has become a daily utility for millions — writing emails, summarizing meetings, tutoring students, debugging code. But the same technology that boosts productivity is now being weaponized in a far darker way. A forthcoming research paper, shared with The New York Times ahead of publication, presents evidence that members of the terrorist group Boko Haram have been using popular AI chatbots to support combat operations. We’re not talking about fringe experimentation. The report, based on interviews with 27 former members conducted in Nigeria over the past two years, indicates that AI tools like ChatGPT, Gemini, Claude, Grok, Meta AI, and DeepSeek were used to gather technical knowledge, troubleshoot weapons, and even assist in planning attacks.
How Boko Haram organized its AI operation
This wasn’t a handful of individuals messing around with chatbots in their spare time. The research describes a structured, deliberate effort. Dedicated teams handled AI-related tasks. Internal training sessions were held. Knowledge about which prompts worked and which safety measures could be bypassed was shared among members. Some users, the researchers claim, managed to circumvent the built-in guardrails that companies design specifically to stop AI from answering violent or dangerous requests.
That raises an uncomfortable question: are today’s chatbots handing out bomb-making instructions on demand? The short answer is: not exactly. Much of the data in the study covers activity through the end of 2024. AI companies insist that newer versions of their models have stronger protections. They also note that many of the prompts fall into a legal and ethical gray zone. Asking an AI how to repair a motorcycle or explaining basic chemistry isn’t inherently harmful — even though the same information can be twisted for destructive purposes.
What the AI companies are saying
OpenAI told the Times that using ChatGPT to support terrorism or violence violates its policies and that it continues improving defenses against abuse. Meta pointed out that the research mostly involved older versions of its AI models and said it has since strengthened safety measures. xAI and DeepSeek did not respond to requests for comment, according to the report.
Still, the fact that organized terrorist groups are actively probing these systems — and sometimes succeeding — underscores a challenge that isn’t going away. AI safety isn’t just about preventing students from cheating on essays or stopping people from generating fake images. It’s about ensuring that powerful tools don’t become force multipliers for violence.
Does AI actually make terrorists more dangerous?
Security experts caution against overstating the threat. Planning and executing attacks still depend on real-world logistics: funding, communication, human coordination, physical materials. A chatbot can’t replace any of that. What AI appears to do, based on this research, is make less-experienced members somewhat more capable. It’s an efficiency gain, not a superpower.
But that distinction doesn’t make the findings any less troubling. As AI models grow more capable and more accessible, the pressure on companies to lock down their systems will only intensify. The line between legitimate use and abuse is often blurry, and bad actors are actively testing where it bends.
What this means for the future of AI safety
The Boko Haram case is a concrete example of a fear that has shadowed AI development for years: that the same tools we rely on for WhatsApp HD photo sending or everyday productivity could be turned against us. The report is a reminder that the conversation about AI safety needs to include national security, not just ethics boards and content moderation policies.
Companies will need to keep updating their safeguards faster than adversaries can probe them. Researchers will need to keep studying how extremist groups actually use these tools — not just theorizing about it. And the public should understand that the AI safety debate is not an abstract, academic exercise. It’s happening now, in real time, with real consequences.
For now, the takeaway is straightforward: AI has already fallen into the wrong hands. The question is whether the people building it can stay ahead of the people abusing it.
