인공지능 검색 사업자의 법적 책임, 그리고 과제

The main issue of AI search results will be how to legally define the results in that they are presented in a new form by selecting, summarizing, and reorganizing information according to users' questions rather than simply listing links, unlike internet searches. AI results can be regarded as compilations or database outputs because they are generated as a single response by combining information from various sources. However, if the outcome does not involve human creativity, it is difficult to protect it as an editorial work or work under the current copyright law, so the AI-generated results remain legally close to the public domain. This means that no one can copyright the AI-generated text or images, and it creates a structure that makes it difficult for infringement to be established even if others use them freely. In addition, the scope of responsibility of operators providing AI search services is discussed. For traditional online service providers (OSPs), the principle of limiting liability called ‘Safe Harbor’ has been applied, which is a principle that service providers are not responsible for user-created content under certain conditions. However, in the sense that AI search generates content by AI rather than users, it is difficult to apply the existing OSP liability limitation principle as it is. In particular, when a service provider directly intervenes in the creation result or adjusts an algorithm to induce a specific result, this is more likely to take responsibility as an information provider beyond simple technical intermediation. On the other hand, if the creation result is entirely automated and there is no intervention of the provider, there is room for the existing OSP liability limitation logic to be applied by analogy. Finally, whether AI search results infringe copyright is an important legal issue. If the response output by AI is substantially similar to the expression of another person's work, it may correspond to infringement of reproduction rights or public transmission rights. However, if AI search simply summarizes or modifies information, it may correspond to ‘fair use’ or ‘citation’, and in this case, copyright infringement is not established. The problem is that if there is an omission or error in the source labeling among the answers provided by AI, and if a hallucination that falsely presents a non-existent source or content appears, the requirements for fair use are not met, which may increase the risk of copyright infringement.