생성형 AI 유해 콘텐츠의 한국형 분류체계 개발: 법률, 정책, 사회문화적 측면에서

다국어 초록 (Multilingual Abstract)

The proliferation of generative artificial intelligence (Generative AI), particularly large language models (LLMs), raises concerns about the creation of illegal and unethical content caused by users’ intentional misuse. While existing AI safety research has primarily focused on the technical limitations of models, systematic responses to user-induced harm have been relatively insufficient. This study centers its analysis on the concept of “User-Induced Harm,” and constructs a “Korean Classification System for Generative AI Harmful Content” by comprehensively analyzing the Korean legal system (14 laws), major AI company policies (7 companies), and socio-cultural specificities (including domestic studies). The proposed framework comprises 8 main categories and 23 subtypes, offering an integrated classification system that reflects legal grounds, corporate policy alignment, and Korean societal sensitivity and values. This research contributes to establishing AI content safety standards tailored to Korean society and presents a concrete framework for managing AI-related risks from a user-centric perspective.