Strengthening transparency has emerged as a pivotal issue in promoting the responsible development of artificial intelligence (AI). As the prevailing framework for corporate information disclosure, Environmental, Social, and Governance (ESG) reporting shares an inherent synergy with AI governance; both are rooted in the pursuit of sustainable development and the disclosure of specific matters to investors and broader stakeholders. This study analyzes the status of artificial intelligence (AI) information disclosure in the ESG (Environmental, Social, and Governance) reports of listed companies across the United States, Europe, and China, finding that: (1) ESG reports have emerged as a primary channel for business organizations to disclose AI-related information; (2) significant disparities exist in disclosure levels across four key AI-related domains—development, application, manufacturing, and consumption; and (3) disclosure density varies considerably across E, S, and G dimensions, with the Governance (G) pillar exhibiting the most comprehensive information. Based on an empirical analysis of the ESG-AI disclosure framework, this study proposes an optimization scheme for ESG-AI reporting, clearly defining mandatory ESG-AI disclosure obligations for listed companies and employing the “comply or explain” mechanism to balance corporate transparency with operational efficiency while adhering to the “Double Materiality” principle by disclosing model training energy consumption and ecological impacts under Environmental (E) matters, addressing employment, employee training, marketing labeling, and customer privacy under Social (S) matters, and elaborating on corporate AI strategies, risk management protocols, and governance policies under Governance (G) matters. Regarding procedural safeguards, taking China as a case study, centralized disclosure could be implemented through the National Enterprise Credit Information Publicity System, complemented by an assurance system for listed company reports to enhance the accessibility and accuracy of information disclosure.
Junkai Chen (2026) studied this question.