DETECTING THE FINGERPRINTS OF FRAUD: A FIVE-STAGE COMPUTATIONAL FRAMEWORK FOR ESG VERIFICATION

ESG reporting suffers from fragmentation across 600+ global frameworks, subjective metrics, and pervasive greenwashing. This paper presents an AI framework integrating LayoutLMv3 for multimodal document processing, ClimateBERT for greenwashing detection, and XGBoost for emission estimation. Evaluation across 139 corporate sustainability reports yields: 95% CO2 extraction precision, 92% recall, a 20% improvement in data completeness, a 25% enhancement in consistency, R2=0.85 enabling 80% Scope 3 gap-filling, a 12% anomaly detection rate, and a 90% workload reduction (40→2 hours per report).5 These results position AI as a structural prerequisite for credible ESG disclosure under EU CSRD and IFRS S1/S2.   

Keywords: ESG Reporting, Artificial Intelligence, Greenwashing Detection, LayoutLMv3, ClimateBERT, XGBoost, Scope 3 Emissions, CSRD, Algorithmic Bias.