Cambridge: Boko Haram used ChatGPT, Claude, Gemini for bombs

The Cambridge study is the first detailed account of how an armed group has folded AI chatbots into everyday operations rather than treating them as a one-off curiosity. Former fighters described dedicated AI units — one JAS unit of 23 people worked from a solar-powered room — and trainers who used projectors and laptops to teach senior members how to query chatbots and get around their safeguards. Bomb-makers kept laptops beside them during assembly and returned to the chatbot each time they hit a problem.

The interviews largely cover 2023 and 2024. The study's author believes the findings are more concerning because the technology has advanced rapidly since then, and says the industry needs heavier regulation. Lucas Webber of Tech Against Terrorism, whose researchers gained access to a pro-ISIL technical forum, described the activity there as "the tip of the iceberg." One supporter compared bypassing safeguards to removing a car's speed limiter.

The policy backdrop is a split between Anthropic chief executive Dario Amodei, who has urged AI companies to slow development of their most advanced models and called for independent evaluations and common safety standards, and the Trump administration, which argues US AI companies must be free to innovate without "cumbersome regulation." That leaves the companies themselves deciding where their systems draw the line — and the test, it seems, is whether safeguard blocklists catch requests that are as unambiguous as a bomb recipe, given the study's finding that the fighters were circumventing restrictions already in place.