안전 필터 실패 속 Boko Haram, ChatGPT로 폭탄 제작
Cambridge 연구에 따르면 ISIS가 모델 탈옥(jailbreak) 기법을 훈련시킨 후 Boko Haram이 공격 계획과 폭탄 제작에 ChatGPT, Claude, Grok을 사용하고 있는 것으로 나타났습니다.
a study by the cambridge programme on ai science & policy confirms that boko haram factions established dedicated ai units to use tools like chatgpt, claude, and gemini for logistics, explosives construction, and military tactics. isis reportedly provided training on bypassing safety filters, allowing groups to extract sensitive information despite guardrails, with one instance resulting in casualties from replicated movie stunts. the findings indicate that current safety mechanisms are unreliable and that ai lowers the barrier to acquiring dangerous knowledge, validating researcher fears about biological and chemical weapon assistance. this failure of voluntary self-regulation suggests a need for stricter policy or technical interventions before groups graduate from conventional to mass-casualty weapon planning.