Terrorist groups using ChatGPT, Claude, Gemini to plan attacks and build weapons

Voluntary self-regulation by AI companies has proven insufficient to prevent terrorist misuse. Although researchers and companies like OpenAI and Anthropic have long warned that AI models could make dangerous knowledge more accessible, the study documents that safety filters on mainstream chatbots—the same ones used by millions globally—have failed to provide reliable protection. Anthropic has recently admitted that jailbreaks will likely never be fully eliminated, suggesting that the vulnerability may be inherent to large language models themselves.

The distinction researchers draw between mainstream chatbots and specialized AI systems is telling. While ChatGPT and Claude primarily surface and organize existing knowledge, the deeper concern lies in potential misuse of domain-specific AI tools in life sciences—where AI could potentially generate genuinely novel and dangerous information. The fact that Boko Haram commanders established dedicated AI units and trained fighters systematically indicates a coordinated effort to operationalize the technology. The group's prior consideration of mass-casualty weapons, combined with demonstrated enthusiasm for AI capabilities, suggests that as terrorist organizations become more sophisticated in their use of these tools, the risk escalates from conventional weapons enhancement to entirely new categories of threat.