법령 정보 확인 및 규제 판별에서의 생성형 AI 활용 가능성: ChatGPT 기반 탐색적 분석

This study explores the applicability of generative artificial intelligence (Generative AI), particularly ChatGPT, to legal information retrieval and regulatory classification. Two tasks were designed: one involved generating legal provisions based on input information such as statute name, article number, and title; the other involved classifying whether a given provision is regulatory or non-regulatory. The study evaluated classification accuracy, content similarity, execution consistency, and performance variation across conditions. Results showed that in the provision generation task, ChatGPT often produced inaccurate, inconsistent, or hallucinated content, with low similarity to reference texts. In contrast, the classification task achieved a mean accuracy of 66.9%, with high consistency and reliability, suggesting potential utility as a supportive tool for identifying regulatory provisions. However, the frequent misclassification of non-regulatory provisions remains a key limitation. As an initial empirical assessment of generative AI in legal and regulatory contexts, this study provides a foundation for future research comparing domain-specific AI models and exploring their use in regulatory support tasks.