Ethical Risks of Generative Large Language Models and Corporate Social Responsibility Performance: Evidence from China

The ethical risks associated with generative large language models (GLLMs) raise concerns about their implications for corporate social responsibility (CSR). Drawing on firm-year observations for Chinese A-share companies over 2015–2024, we investigate how GLLM-related ethical risk relates to CSR performance. A firm-level disclosure-based measure of GLLM-related ethical risk signals is constructed from annual-report text and analyzed using two-way fixed-effect models. The results show a negative and statistically significant association, with higher ethical risk corresponding to lower subsequent CSR performance. This association persists across robustness checks and instrumental variable analyses. Firm-clustered bootstrap analyses reveal significant negative indirect associations through reputational pressure and financing constraints, supporting the proposed mediating pathways. Managerial compliance attention strengthens the negative association between ethical risk and CSR performance. These findings extend research on the antecedents of CSR by identifying the ethical risks of emerging technologies as a relevant organizational concern. They also highlight the importance of integrating ethical AI governance with CSR planning to support responsible technology adoption and sustained corporate sustainability commitments.