DeepGreen Detects Corporate Greenwashing Using LLMs

This paper proposes DeepGreen, a dual-stage LLM-driven system to detect corporate greenwashing in 9,369 A-share annual reports published between 2021 and 2023. Validation shows high reliability and ablation experiments indicate Retrieval-Augmented Generation (RAG) reduces hallucinations versus longer input windows; empirical IV, PSM, and placebo tests find DeepGreen’s greenwashing signals positively associate with environmental penalties. Findings suggest green investors, firm size, and green assets can weaken this penalty correlation, informing targeted ESG oversight.