AI 환경에서의 주식회사 이사의 감시의무와 내부통제시스템의 실질화 - 개정 상법 제382조의3·인공지능기본법·Caremark-Marchand 법리의 통합적 분석 -

초록

As artificial intelligence increasingly assumes a central role in corporate decision-making, directors’ oversight duties have undergone a qualitative transformation. The 2025 amendment to Article 382-3 of the Korean Commercial Act extended directors’ fiduciary obligations from the corporation alone to “the corporation and its shareholders,” opening a direct legal pathway under which AI governance failures may constitute a breach of the duty of fair treatment of shareholders. The AI Basic Act (in force January 2026) gives concrete statutory content to directors’ duty of care through the obligation to confirm high-impact AI status (Article 33), the Human-in-the-loop mandate (Article 34(1)(iv)), and the AI impact assessment (Article 35). These provisions fill the regulatory gap exposed by the Delaware Bingle decision, providing a normative basis for potentially converting AI risks from business risk to compliance risk for Korean corporate law purposes. This article functionally reconstructs the Delaware Caremark–Marchand framework as a duty of substantive AI internal control. Because AI may constitute a Mission Critical Risk when deployed in core business functions, boards must establish real monitoring systems covering the transparency, fairness, and security of AI algorithms; willful ignorance of AI red flags constitutes conscious disregard. A director who approves AI outputs without independent scrutiny engages in “algorithmic conscious disregard,” amounting to bad faith and excluding Business Judgment Rule protection. The article further demonstrates convergence between the German Stock Corporation Act § 91 framework and the Korean Supreme Court’s line of oversight decisions, which from 1985 to 2025 has steadily converged toward the substantive operation of internal control systems, and proposes three legislative reforms: (i) codification of algorithmic legality review within the compliance officer regime; (ii) mandatory Algorithmic Impact Assessment with public disclosure; and (iii) an AI Governance Code of Best Practice on a comply-or-explain basis. Delaware’s SB21, which weakened MFW minority-shareholder protections, offers a comparative benchmark against which Korean procedural safe harbor design must not compromise minimum standards; detailed analysis is reserved for subsequent research.

키워드