Greenwashing alert system for EU green bonds
The CONSOB–University of Trento prototype

S. Paterlini, A. Nicolodi, M. Gentile, V. Foglia Manzillo, M.R. Sancilio, P. Deriu

FinTech no. 14 - July 2025 [PDF] 
 

Abstract

The demand for green bonds – financial instruments used to fund environmentally sustainable projects – has recently registered an increasing trend by remaining on high levels since 2021. At the same time, concerns about greenwashing have risen as well, potentially undermining investor confidence, harming market integrity, and slowing down the transition to a sustainable economy.
According to ESMA, identifying greenwashing cases could be challenging due to misleading financial disclosure or omission of information. Indeed, sustainability claims could suffer from absence of substantial backing or evidence, cheap talk, inconsistency (i.e., mismatch between the company’s sustainability claims and its actual practices), cherry-picking, complexity and lack of transparency. In addition, market pressure due to consumer demands or investor expectations could lead companies to overstate their environmental credentials and expertise. Lastly, the absence of a specific and unique regulation on this topic at both national and EU levels could undermine greenwashing detection as well. Similarly, academic literature underlines various but somewhat interrelated interpretations of greenwashing such as ‘selective disclosure’, ‘unsubstantiated or misleading claims’, ‘gap between environmental information disclosed and actual environmental performance’ and ‘overly positive beliefs about an organization’s environmental performance’. All these definitions refer to firms manipulating communication with the aim of creating a favourable, social and eco-friendly company image.
The aim of our research is to develop a first prototype that should be grounded in Artificial Intelligence (AI), which could support supervision activity by providing alerts for potential cases of greenwashing (i.e., greenwashing alert system). Financial authorities increasingly recognize the need for supervisory technology tools (SupTech) to enhance their oversight capabilities. SupTech leverages advanced technologies such as AI and machine learning (ML) to improve the efficiency and effectiveness of regulatory supervision. These tools enable authorities to quickly process vast amounts of data, potentially identifying risks early, and supporting compliance with regulatory standards.
The developed prototype relies on large language models such as ClimateBERT and ESGBERT, combined with a proprietary dictionary that maps a defined set of keywords to each Sustainable Development Goal (SDG), based on the alignment of the Green Bond Principles (GBP) with the SDG framework. First, the prototype can identify environmental phrases and environmental claims in selected documents such as sustainability reports, tagging them accordingly as ‘environmental’ or ‘environmental claim’. Second, the prototype highlights the sentiment of statements and classifies them into three categories: risk, opportunity, or neutral, depending on the tone of the content. Third, the prototype performs a dictionary-based search to extract SDG-related phrases, assigns them to the corresponding SDG and then computes an SDG mismatch measure that quantifies discrepancies between the declared and the detected SDGs. The overall system provides structured insights that can support regulatory and supervisory efforts. By automating these processes, the prototype can significantly reduce the time which analysts need to manually review sustainability reports and mitigate potential biases that can arise from subjective interpretation. This systematic approach could not only improve efficiency but also strengthen the reliability and transparency of analyses. The prototype has not yet been fully validated and tested, primarily due to the limited number of confirmed greenwashing cases; nonetheless, it shows potential for future implementation as a greenwashing alert system.

Authors
Sandra Paterlini - University of Trento, Department of Economics and Management (sandra.paterlini@unitn.it);
Andrea Nicolodi - University of Trento (andrea.nicolodi-1@unitn.it);
Monica Gentile - CONSOB, Research and Regulation Department (m.gentile@consob.it);
Vincenzo Foglia Manzillo - CONSOB, Issuers Supervisory Department (v.fogliamanzillo@consob.it);
Maria Raffaella Sancilio - CONSOB, General Directorate (m.sancilio@consob.it);
Paola Deriu - CONSOB, Head of the Research and Regulation Department (p.deriu@consob.it);

The authors are the only responsible for errors and imprecisions. The opinions expressed here are those of the authors and do not necessarily reflect those of CONSOB.

JEL Classifications: Q01, Q56, G18, G20, G38, C63, O31.

Keywords: sustainable finance, green bonds, SupTech tools, machine learning, Artificial Intelligence