Why does AI need oversight in ESG reporting

Artificial intelligence is emerging in every part of the sustainability landscape as a growing number of companies use AI-powered solutions for reporting, risk management and decision-making. The question is whether these new tools strengthen sustainability information or simply add new levels of complexity, writes Reuters.

Even with adjustments to Europe's new sustainability rules, companies are facing increasing pressure. The Corporate Sustainability Reporting Directive (CSRD) and the Corporate Sustainability Due Diligence Directive (CSDDD) require companies to work with larger volumes of data from more sources, often in tight deadlines.

The ability of artificial intelligence to interpret unstructured data, to compare disclosures on a large scale and identify patterns means that it can help organizations manage these tensions, but its use also requires caution.

Companies often store relevant information across multiple systems, making it difficult to achieve consistency. Tools like Sweep's carbon management platform use machine learning models to identify discrepancies in supplier data and flag them for review, helping businesses improve their accuracy. This is particularly useful for dual materiality assessments, where gaps in supplier data and inconsistent reporting formats remain the norm.

Filling these gaps remains one of the most challenging parts of reporting. Companies like Watershed have developed proxy approaches for Scope 3 emissions that draw on broad datasets to estimate emission factors where supplier information is lacking.

These estimates are not a substitute for direct data, and their use comes with challenges-different models can yield significantly different results, and proxy-based approaches remain sensitive to incorrect assumptions. However, when used correctly, they help companies move from broad averages to more specific, decision-relevant estimates.

Greater access to inside information

AI can also make internal data more accessible. Large organizations often struggle to extract the information needed for sustainability disclosure, especially for governance, risk processes, and supply chain due diligence. AI-powered search and tagging tools can help find evidence and match it to disclosure requirements.

But these systems can also introduce new risks if they reveal outdated or incorrect information. Ensuring that internal data is version-controlled and auditable remains a critical human task, even when artificial intelligence is involved.

The usefulness of sustainability data also depends on whether it informs financial decisions, oversight, and strategy. The non-profit organization Ceres, for example, uses artificial intelligence to analyze climate disclosures from hundreds of insurers. By applying consistent labels across the TCFD pillars, it identifies patterns and gaps that would otherwise take analysts months to identify. The approach improves comparability and provides clarity about where companies are in line with reporting expectations.

These capabilities are also valuable for supervisors. The Bank for International Settlements has explored how AI can help prudential regulators go beyond firm-by-firm analysis to understand system-wide risks. AI can compare disclosures, highlight missing metrics, and identify patterns in reports. This gives supervisors a clearer picture of systemic risks and practices.

Challenges

Investors face similar challenges. The first CSRD reporting cycle provided much richer information, but many investors found it difficult to review and compare these reports.

Tools like those developed by Climate Aligned use plain language analysis to compare companies' transition plans and climate reporting, allowing investors and other financial stakeholders to understand whether commitments are capital expenditure-related and whether activities are aligned with taxonomies.

Artificial intelligence is also applied to risk management in globalized supply chains, which are complex and vulnerable to disruption.

In the early stages of the COVID-19 pandemic, satellite data showed reduced pollution in parts of China, which provided an early signal for factory closures. Similar dynamics can be seen in microchip shortages or reduced nuclear power generation during periods of low river flow.

The amount of Earth observation data has also grown rapidly thanks to satellite imagery, climate models, and sensor networks, generating terabytes of data and trillions of observations.

The non-governmental organization ClimateTRACE uses nearly a billion data points to track methane and other greenhouse gas emissions around the world. This type of analysis would be far beyond the capacity of individual analysts without artificial intelligence. This not only increases transparency, but also reveals risks that might otherwise go unnoticed.

Google's Deepmind just released WeatherNext 2, which uses artificial intelligence to significantly improve the accuracy of storm forecasts. This could provide additional days of warning, allowing for earlier evacuations and reduced damage. Better forecasts of drought or crop failures could help identify famine risks before market signals emerge. With more time to prepare, governments and aid agencies can act to mitigate impacts.

It is important to note that signals generated by artificial intelligence require human judgment to properly assess their implications and integrate them into decision-making processes.

This is true for all applications of AI in sustainability. One concern is the rise of cheap AI-generated reports. These tools can produce quickly produced documents, but with limited quality control. The results may not reflect a company's strategy or specific challenges, and without expert oversight, the reports pose significant reputational or regulatory risks.

Another challenge is information overload. AI-powered reporting and synthetic data sets create vast amounts of new data, but much of it may not be useful for decision-making. This only creates noise that analysts, investors, and regulators have to sift through.

Data security is an additional risk. Poorly managed AI tools can create vulnerabilities that expose sensitive customer or personal information. The threat to data security comes both from AI-enabled cyberattacks and from organizations failing to properly manage their own data when using AI.

The final problem is the potential for systemic correlations. If many market participants rely on such AI models, the diversity of judgments can decrease. This increases the volatility of market bubbles and crashes, according to research by the London Stock Exchange (LSE). Markets depend on differentiated perspectives, and overreliance on such approaches can undermine that stability.

Ultimately, AI has an important role to play in managing sustainability risks in an increasingly volatile world, but only if it is used wisely and in a well-managed environment. Most importantly, humans must remain in the driving seat.