AI 학습용 데이터의 특허 보호 방안

This article explores the evolving legal framework governing the patentability of AI training data—a critical asset whose importance has grown in tandem with advances in artificial intelligence (AI). In the AI context, data functions not merely as an informational resource but as a core enabler of technological innovation through processing, analysis, and machine learning applications. While data may be protected under copyright law or trade secret regimes, it generally falls outside the scope of patentable subject matter, as it does not inherently involve the application of natural laws. However, when data is integrated with technical means—such as specific algorithms or processing mechanisms—and is concretely defined as a claim element, it may be eligible for patent protection. This legal gray area has led to a surge in patent applications seeking to capture data-centric inventions. To secure protection, applicants have increasingly adopted claim formats including terms such as “…data,” “…dataset,” “…data structure,” “…structured data,” and “…data storage medium.” Among these, the Korean Intellectual Property Office (KIPO) currently recognizes only “data storage medium” claims as product inventions under its Technical Examination Practice Guide, offering no explicit guidance on the eligibility of “dataset” or “data structure” claims. This lack of clarity has led to inconsistent examination outcomes and growing legal uncertainty. To resolve these challenges, this article advocates for the revision of the Guide to explicitly recognize “dataset” and “data structure” claims as eligible product inventions when they satisfy the requisite technical criteria. At the same time, it should clarify that claims directed solely to “data”—which are not considered product inventions in jurisdictions such as the United States, European Union, or Japan—or to “structured data” (recognized only in Japan), do not satisfy the requirements for patentable subject matter in Korea.