Unmasking Greenwashing with AI

As sustainability becomes a central focus in corporate strategy, businesses are eager to showcase their environmental credentials to appeal to investors, consumers and regulators. But not all that glitters is green.

The rise of greenwashing - the exaggeration or fabrication of eco-friendly practices - has become a significant challenge, eroding trust and distorting critical data that actuaries rely on for decision-making.

To combat this challenge, advanced technologies like Natural Language Processing (NLP) can be used to uncover misleading claims and promote accountability in sustainability practices. This will assist actuaries and industry professionals to make informed and reliable decisions.

Drawing insights from my recent presentation,  AI-Powered Detection of Corporate Greenwashing , delivered at the UNSW Workshop in Risk and Actuarial Frontiers: Data Science and AI in Actuarial Practice, I've highlighted how NLPs can aid actuarial practice. 

Why greenwashing is an actuarial concern

Greenwashing presents several risks that directly impact actuarial work, including:

• Distorted risk assessments: Misleading claims can result in inaccurate evaluation of sustainability-linked risks.
• Regulatory compliance: Non-compliance with disclosure standards due to greenwashing can lead to penalties.
• Investment decisions: ESG-focused portfolios may inadvertently include companies with overstated environmental commitments, impacting long-term performance and 'true to label' considerations.
• Legal liability, regulatory penalties, and professional indemnity claims for misrepresenting sustainability credentials which impacts reserving and capital requirements, as insurers and financial institutions may face growing climate litigation risks.

How NLPs detect greenwashing

By analysing corporate disclosures and other textual data, NLP equips actuaries with scalable and efficient tools to detect misleading claims and validate sustainability, offering advantages over manual review by processing vast amounts of information quickly and systematically. This is achieved through:

1. Sentiment analysis which evaluates tone in corporate sustainability statements, flagging overly positive language that may indicate exaggerated claims.
2. Named Entity Recognition (NER) which identifies key entities - like companies, emissions metrics and sustainability initiatives - and links claims to measurable outcomes, enabling validation against external datasets.
3. Topic modelling which analyses themes in reports, highlighting the prevalence of broad statements over actionable commitments, which can be further validated for consistency and specificity. It also highlights neglect of critical areas, such as Scope 3 emissions, which account for indirect environmental impacts.
4. Relationship extraction which maps connections between entities, such as linking a company to claimed environmental projects, and detects mismatches between reported initiatives and actual data.
5. Discrepancy mapping which compares self-reported emissions and sustainability claims with independent data sources to flag potential contradictions, such as rising CO2 levels despite sustainability progress claims.

Constructing the Green Authenticity Index (GAI)

A significant aspect of my presentation was the construction of the Green Authenticity Index (GAI). The GAI applies the Stacey Matrix to evaluate corporate sustainability claims, ensuring a structured assessment of transparency and credibility. It focuses on two key dimensions. This index is designed to quantify the sincerity of corporate sustainability claims.

Key dimensions of the GAI

The GAI evaluates two primary dimensions:

• Certainty: Measures the clarity, factuality and specificity of corporate language. Precise commitments (e.g., "reduce carbon emissions by 20% by 2023) receive higher scores than vague statements (e.g., "strive for sustainability").
• Agreement: Assesses the alignment between reported claims and external data, such as ESG scores, emissions reports and third-party audits.

Alternative dimensions (e.g., complexity of sustainability challenges, impact of actions, or intent) were considered but found less effective, while certainty and agreement provide the clearest, most actionable insights for greenwashing detection.

Data inputs include:

• Corporate disclosures: Reports, sustainability updates and filings.
• External data: ESG ratings, independent audits and verified emissions data.
• Unstructured sources: Social media posts, press releases and news articles.

Weighting and scoring

• NLP models assign weights to linguistic patterns, sentiment and factual consistency.
• Sensitivity analysis ensures that no single factor disproportionately influences the score.

Outputs

• Companies receive a comprehensive score that ranks their sincerity in sustainability efforts.
• Visual tools enable actuaries to compare companies across sectors or regions, providing actionable insights.

Case Study: Assessing a Retailer's Green Pledge with GAI

A global retailer announces a "Net Zero by 2035" commitment, prompting an actuarial team to assess its credibility for an institutional investor considering ESG-linked bonds.

Using the GAI, the team evaluates certainty and agreement. The company provides a roadmap with specific targets, audited Scope 1 and 2 emissions, and verified carbon offsets. However, it omits Scope 3 emissions - critical in retail supply chains - raising concerns about the completeness of its commitment.

Comparing external data, ESG ratings (e.g., MSCI, MarketPsych) score the company highly, but NGOs and media reports flag unverified carbon offset claims. Public sentiment is mixed, with some praising the goal while others question its feasibility.

GAI assigns a moderate certainty score due to missing Scope 3 data and a low-to-moderate agreement score reflecting conflicting third-party opinions. The findings suggest a potential greenwashing risk, requiring further due diligence before endorsing the company's ESG-linked financial products.

Challenges, limitations and the evolving nature of greenwashing

While the GAI is a valuable tool for assessing corporate sustainability sincerity, it has limitations. Since it relies heavily on NLP, it may miss non-textual factors, such as internal carbon pricing or operational changes not reflected in reports.

Companies skilled in narrative control might score well despite weak sustainability efforts. The GAI is not a substitute for deeper financial analysis. If used in isolation, it may lead to misjudgments about a company's true sustainability performance, as it does not directly model climate risks or investment exposures.

Additionally, external data bias can distort results. ESG ratings and public sentiment may be influenced by selective disclosures or strong PR campaigns, making some companies appear more credible than they actually are.

As Abraham Lincoln famously remarked:

Corporations will evolve and adapt their strategies in response to new detection technologies, learning how to navigate and circumvent some of the methodologies aimed at exposing greenwashing.

This principle underscores the importance of continuously improving tools like the GAI and ensuring that detection methodologies remain robust, adaptive and transparent.

Implications for actuarial practice

The integration of NLP and tools like the GAI into actuarial frameworks offers transformative potential through:

• Enhanced risk assessment by identifying inconsistencies in sustainability claims and refining environmental risk models, such as the DICE model for economic climate impacts, the MSCI Climate VaR for financial risk projections etc., By integrating verifiable emissions data and external ESG assessments, these models will be able to better quantify risks, improving forecasts for carbon pricing, supply chain disruptions, and extreme weather impacts.
• Informed investment strategies by supporting ESG portfolios by distinguishing credible efforts from deceptive practices.
• Regulatory compliance by helping companies align with sustainability reporting. The GAI helps assess whether a company's claims match independent data sources, such as ESG ratings, carbon footprint reports, and industry benchmarks.

A more sustainable future

The integration of advanced technologies like NLP into actuarial practice is a strategic response to the increasing importance of sustainability metrics. Tools like the GAI enable actuaries to quantify corporate sincerity, identify inconsistencies, and promote accountability in sustainability reporting.

Adopting the GAI enhances transparency and accountability in sustainability reporting by verifying corporate claims against independent data. As sustainability regulations and expectations evolve, GAI helps companies align disclosures with measurable actions, reducing greenwashing risks and improving trust among investors, regulators, and society.