Terrorist groups are increasingly leveraging major AI chatbots for attack planning and weapons development, according to a new study. The research, conducted by Antonia Jülich of the Cambridge Programme on AI Science & Policy (CASP), found that groups like Boko Haram and ISIS are using AI to enhance their operational capabilities. The study highlights that Boko Haram has trained its commanders to bypass AI safety filters and has established dedicated AI units. These units are used for attack planning, building more powerful explosive devices, weapons maintenance, and operational security.
According to Jülich, Boko Haram's AI units have been operational since 2023, with ISIS also providing training to commanders on how to bypass safety filters. One notable example involved the ISWAP faction using AI to replicate motorcycle jumping techniques from a movie, which led to 18 fighters dying during training and eight successfully making the jump. Jülich's research includes 57 interviews with 27 former members of the group, providing insight into the extent of AI's role in terrorist activities.
The study also warns that while chatbots like ChatGPT and Claude have received significant public attention, they are not the primary concern. Researchers and AI labs like OpenAI and Anthropic have long warned about the risks of AI models making dangerous knowledge more accessible. However, voluntary self-regulation has not been sufficient to prevent misuse, as safety filters have failed to reliably prevent such activities. Anthropic recently acknowledged that jailbreaks may never be fully eliminated, underscoring the persistent challenge of AI safety.
Source: thedecoder
