생성형 인공지능 할루시네이션 현상의 법적 책임 구조와 기본권 보호를 위한 규범설계

This article reconceptualizes hallucinations in generative artificial intelligence as structural and institutional risks rather than mere technical errors, and examines their implications for fundamental rights and legal responsibility. Generative AI operates through probabilistic language generation without mechanisms for factual verification, which enables the production of non-existent statutes, judicial precedents, or statistical data presented as credible information. Such hallucinations undermine informational reliability and may result in infringements of constitutionally protected rights, including reputation, informational self-determination, and freedom of expression. In particular, when generative AI is utilized in the public sector and professional domains not only as a decision-support tool but as a de facto basis for decision-making, technological risks are transformed into institutional risks and ultimately crystallize into constitutional risks. Drawing on comparative case studies, including Mata v. Avianca, this article identifies gaps in responsibility attribution and deficiencies in existing verification obligations. It argues that regulatory responses should focus on ex post liability mechanisms and the clarification of human final decision-making responsibility, while respecting the prohibition of prior restraint and the principle of proportionality. The article ultimately proposes a normative framework for legal responsibility that prioritizes the protection of fundamental rights in the era of generative artificial intelligence.