인공지능(AI) 감사 도입 방안 연구

Artificial Intelligence (AI) auditing is the process of evaluating whether AI systems function as expected and comply with ethical and legal standards. It involves verifying that the systems operate accurately as intended and do not produce biased or discriminatory outcomes. Currently, there is no statutory audit specifically for AI systems. However, if an audited company uses AI systems to prepare financial statements or to implement internal controls based on the system’s judgments, these systems may become subject to audit. The rapid advancement of AI technology introduces new risks, necessitating continuous consideration of AI audit systems, standards, and methodologies. While traditional IT audits assess efficiency, security, and legal compliance, AI audits must also examine additional factors such as fairness, bias, and unpredictability, alongside existing evaluation elements. The timing of AI audits varies depending on the subject and type of audit, and audit timing can be determined based on the AI lifecycle stages. AI auditors must meet various qualification criteria, including expertise, independence, and legal responsibility. AI auditing faces several challenges, such as the absence of audit standards, ambiguity in the scope of audits, and high costs. To address these issues, it is necessary to establish the direction for designing AI audit systems and to develop reasonable audit plans. Specifically, when designing AI audit systems, a balance between audit quality and cost should be considered, and a flexible and continuous audit framework that reflects the complexity and rapid pace of change in AI systems should be built. Reasonable audit plans include clarifying audit objectives, utilizing a risk-based approach, leveraging internal and external audits, and introducing AI audit tools. Details such as the subject and scope of the audit, the timing and frequency of audits, and the qualification criteria for auditors are also important. Organizations conducting AI audits should enhance their expertise through team-based audits and protect audit personnel by clarifying the scope of legal responsibility and establishing disclaimers. Fostering AI audit talent is essential, and the development of new audit standards and procedures to address generative AI audits is also necessary.