생성형 AI 검색 서비스(Google AI Overview)와 법적 책임- 독일 하급심 판례를 중심으로 -

초록

With the advancement of artificial intelligence (AI) technology, the paradigm of internet search is undergoing a fundamental transformation—from a passive infrastructure that merely mediates hyperlinks and snippets, to an active determiner of information that directly generates and provides complete, sentence-form answers by combining large language models (LLMs) with retrieval-augmented generation (RAG) technology. Generative AI search services such as Google's AI Overview dramatically enhance information accessibility, but at the same time give rise to novel legal issues, including defamation and reputational harm caused by hallucinations, unauthorized use of third parties' copyrighted works, and distortions of trademark rights and competitive order. Against this backdrop, the question of whether a platform providing such a service remains a mere intermediary of third-party information—and thus retains the protection of liability exemption provisions—or has instead transitioned into the status of an actor that directly generates the resulting output—and is thus excluded from such exemptions—constitutes the starting point for the entire body of discussion across the relevant areas of law. This paper examines this issue with particular focus on two recent lower court decisions rendered in Germany. The Regional Court of Munich I (2026) characterized the hallucinated statements produced by the AI Overview as the platform's own independent statement, thereby excluding the application of the hosting liability privilege under the Digital Services Act (DSA) and affirming direct tortious liability for the infringement of corporate personality rights and the right to an established and operating business. By contrast, the Regional Court of Berlin II (2026), addressing output generated by the same AI Overview function, held that such output did not constitute use of a trademark under trademark law and amounted to nothing more than a technical computational result derived from third-party information, thereby denying liability for both trademark infringement and unfair competition. Furthermore, this paper compares and analyzes the liability doctrines that the German Federal Court of Justice (BGH) has traditionally applied to conventional search services and autocomplete functions—namely, limited liability as an indirect disturber (Störer) and liability for omission following notice—against the AI Overview service, and examines in parallel the corresponding positions taken by Korean case law and scholarship. The analysis reveals that the Korean legal system likewise adopts, centered on Article 44-2 of the Act on Promotion of Information and Communications Network Utilization and Information Protection, an ex post, knowledge-based liability structure premised on the mediation of third-party information, aligning it with the German approach. However, the German lower courts' reasoning—that this very premise is absent from the outset in the case of AI Overview services—suggests that, under Korean law as well, a distinct construction of direct liability may be required, rather than a straightforward analogical application of the existing liability doctrine developed for portal operators. In conclusion, this paper assesses that generative AI search services are, in theory, likely to move in the direction of bearing direct liability as principal actors, while also proposing, as a policy-oriented compromise reconciling the competing values of digital industry innovation and user protection, the possibility of transitionally applying a liability-for-omission framework. This paper notes, however, that such an approach can only serve as a transitional solution, and that where identical or similar errors recur even after notice has been given, liability must shift to direct liability for affirmative conduct.

키워드