KT, 한국 문화 특화 AI 벤치마크 'KSAFE-MM' 공개

KT's KSAFE-MM establishes the first comprehensive safety benchmark for multimodal AI models reflecting Korean cultural contexts, challenging English-dominated evaluation standards. The automated pipeline enables rapid deployment of culturally-tailored safety benchmarks across languages without deep cultural expertise. This open-source contribution sets a new precedent for region-specific AI safety governance, potentially influencing global approaches to localized AI alignment.