AI 환경에서의 주식회사 이사의 절차적 정당성과 기업지배구조 장치 - MFW 절차와 설명가능 AI(XAI)를 통한 이사의 충실의무의 절차적 정당화 -

초록

The black-box character (opacity) of AI decision-making makes it difficult to verify the substantive legitimacy of an outcome after the fact. Where a director relies on an AIʼs judgment yet cannot confirm its internal logic, this amounts to abandoning ʻʻinformed judgmentʼʼ and may result in a breach of the duty of care and the duty of loyalty. In an AI-driven decision-making environment, the basis of legitimacy must therefore be sought not in the efficiency of outcomes but in procedural transparency and explainability. Through a comparative analysis with Delaware law, this article examines how AI decision-making can be procedurally justified. This article combines the procedural-legitimacy principle of the Delaware MFW procedure with explainable AI (XAI) techniques to develop three lines of analysis. First, drawing on Kahn v. M&F Worldwide (2014) and In re Match Group (2024), it argues that a procedural structure—in which directors examine the AIʼs ʻʻreasoningʼʼ rather than its ʻʻconclusion,ʼʼ subject it to verification by an independent AI Special Committee, and grant final approval through a Human-in-the-loop mechanism—serves as the evidentiary basis establishing the ʻʻinformed judgmentʼʼ element that the Business Judgment Rule already requires, and that its absence defeats the Ruleʼs application. Second, it shows that the visualization of algorithmic reasoning through XAI techniques such as SHAP and LIME serves as procedural evidence of ʻʻgood faith oversight,ʼʼ and that the Human-in-the-loop obligation under Article 34(1)(iv) of the AI Basic Act furnishes a positive-law basis for prohibiting the complete delegation of decision-making authority to AI. Third, it identifies two new problems that AI use raises for the duty of loyalty: a conflict of interest at the AI-system selection stage (such as pressure to adopt algorithms within designated-director structures), and ʻʻepistemic abdication,ʼʼ whereby a director ratifies AI outputs without substantive scrutiny. It proposes a ʻʻBJR 2.0 (demonstrable stewardship)ʼʼ standard, under which a director qualifies for Business Judgment Rule protection only by demonstrating that the algorithmic tool was interrogated as rigorously as a conflicted advisor. To keep these procedural devices from degenerating into ʻʻsimulacral governance,ʼʼ the boardʼs capacity for substantive interrogation of AI and its independent judgment are indispensable. The way in which Delawareʼs SB21 weakened the minority-shareholder-protection function of MFW case law offers a comparative lesson that Korea must not compromise minimum standards when designing a procedural safe harbor for the AI Special Committee.

키워드