Who Is Accountable? When AI Makes the Wrong Decision? Rethinking Corporate Governance
TL;DR
It is argued that accountability for algorithmic harm cannot rest on a single actor or a single governance layer, and responsibility instead needs to be distributed across the people and functions that design, approve, deploy, and supervise an AI system.
Abstract
Corporate boards now sit atop decision architectures that no longer belong to them alone. Machine learning systems screen credit applications, price insurance risk, recommend mergers, flag fraud, and increasingly shape the strategic judgments that directors and executives once reached through experience and deliberation. When one of these systems produces a harmful, discriminatory, or commercially damaging outcome, existing governance doctrine struggles to identify who answers for it. This paper examines the widening gap between algorithmic decision-making and the accountability structures built for human agents. Drawing on agency, stakeholder, stewardship, institutional, and resource dependence theories, together with enterprise risk management and responsible AI scholarship, the paper argues that accountability for algorithmic harm cannot rest on a single actor or a single governance layer. Responsibility instead needs to be distributed across the people and functions that design, approve, deploy, and supervise an AI system, with each layer answerable for a distinct category of failure: design flaws, oversight lapses, deployment misjudgment, and monitoring neglect. The paper's central contribution is a multi-level Corporate AI Accountability Governance Framework that assigns differentiated responsibilities to the board, executive leadership, a dedicated AI governance committee, risk management, internal audit, technology teams, external vendors, and regulators. The framework is built around a feature that conventional governance controls were never designed to handle: AI systems continue to change after deployment, so a one-time approval cannot substitute for ongoing supervision. The paper closes with practical implications for boards preparing for algorithmic oversight, a comparative reading of regulatory expectations across the European Union, the United States, the United Kingdom, and selected Asia-Pacific economies, and a research agenda for scholars working on the next phase of digital corporate governance.