As in every other field, AI has the potential to revolutionize the way financial regulation and supervision is managed. AI could not only make climate reporting easier for companies, but also make the data more reliable for regulators.
One of the key opportunities that AI brings to financial surveillance and regulation is to analyze large data sets and identify new opportunities. Several research projects are investigating how AI can assess climate risk cases from text and how it can be used to track and disclose the environmental impacts of supply chains.
However, questions remain about its effectiveness and whether its energy use is worth the cost.
Central banks are experimenting with AI
Central banks are already exploring ways to use AI in both their operations, including understanding climate risks.
A Banque de France research report examining how AI can help companies estimate carbon emissions found that AI predicts carbon intensity in 69% of cases, but has difficulties in extreme cases, such as heavy polluters. In Vietnam, the vice president of the National Bank of Vietnam said AI could be used to improve ESG reporting.
Meanwhile, the Bank for International Settlements (BIS) has two projects in its innovation hub focused on climate-related AI applications: Project Gaia and Project Symbiosis. Project Gaia uses large-scale language models (LLMs) to automatically extract climate-related indicators from publicly available reports. According to BIS, the aim is to overcome the lack of global reporting standards for comparing information on climate-related risks. The Gaia project is ongoing, with additional use cases and “relevant to a much broader context than climate-related data analysis,” says the BIS overview of the project.
Project Symbiosis builds on the work done at Gaia using various machine learning subsets such as LLM, deep learning, and natural language. The partnership will use AI to collect, interpret, and calculate Scope 3 emissions, identify opportunities to reduce emissions, and explore how that data can be used to match suppliers with funding sources to decarbonize supply chains.
With 95% of the financial sector’s emissions falling under Scope 3, the project aims to “show how new technologies can provide viable technological pathways to positively impact core stakeholders by reducing the critical information gaps that hinder climate change.”
The results of this project are consistent with other AI and risk research and could provide the basis for more standardized emissions and help improve issues around standardizing scope 3 calculations.
The usefulness of AI in climate risk assessment
Interest in using AI to reduce climate risks extends beyond central banks.
Several projects are exploring ways to apply AI to satellite and other imagery data, developing multi-input models that compile data points from images and text together.
This includes aspects like using image segmentation to find out the carbon footprint of a particular region or classifying it based on plants known to grow in that area, said Peter Schwendner, a machine learning expert at the Zurich University of Applied Sciences.
When executed with intention and leveraged in a focused manner using AI production, the impact can be truly meaningful and positive.
– Mattia Romani, Systemiq Partner
He said that from a regulatory perspective, such use of AI is “to improve market transparency, and then financial markets need to improve asset allocation, with the aim of allocating more capital, for example, to sustainable assets. This should work for both investment and lending.”
This is exactly the type of project that spaceborne AI company Kuva Space is hoping to scale up. A Finnish company has partnered with WWF-Indonesia to explore how AI can be applied to hyperspectral imaging to understand changes in regional coastal ecosystems.
Hyperspectral imaging uses advanced satellite cameras to capture images of hard-to-reach areas. Kuva Space’s AI system can flag potential changes or anomalies, such as changes in the health of seagrass, an important ocean carbon store, for scientists on the ground to see.
Although the program is in a pilot phase, the project could have a bigger impact by not only tracking ecosystems but could help regulators and investors identify and track project areas in areas such as blue carbon.
Malachy Escola, Kuva Space’s commercial director, said investors and regulators want this information, but they don’t have it. “We’re really going to make it more science-based or evidence-based and provide that information to decision-makers so they can have confidence.”
What you need: Better data (and more data)
But for artificial intelligence to be truly effective at understanding climate risk and natural loss, Schwendner says more data is needed, especially from companies.
Companies themselves don’t need to crunch numbers or create models, but they do need to provide raw data for analysis.
“Academics and data providers are very keen to work with this data, so I don’t think there is a need for the general public to invest heavily here…This can be achieved, for example, through collaboration between scientists and data providers, but the raw data about the company’s operations needs to be available,” he said.
This includes information such as which companies are sourcing raw materials, in what quantities, and from where.
“Right now, we only know this very roughly. We need to know the exact numbers. Once we know that, and once we know the production volume, and where the energy is coming from in this local production facility… we can estimate the environmental impact. Then we can estimate the climate impact.”
More data was expected this year as a result of the EU’s climate disclosure rules, but the EU’s sustainable omnibus measures may mean less information is available than expected.
Will AI be good or bad for the environment?
Another serious question about using AI to reduce climate change risks is whether it is worth considering the impact on energy use and the environment.
AI consumes large amounts of energy and water. Those working on AI projects in the climate field recognize the contradiction that using something for environmental purposes itself uses large amounts of natural resources.
Experts say the biggest issue in using AI is the types of models being used and how they are applied. Training models requires large amounts of energy, increasing power demands on data centers. It accounted for 1.5% of total energy demand in 2024 and is projected to account for 10% of energy demand growth by 2030.
Even researchers at the BIS Innovation Hub say in their report on Project Symbiosis that “the use of AI could generate large amounts of emissions, even as electricity grids around the world continue to slowly decarbonize at varying rates.”
But some say AI could have a positive impact on the environment.
Related articles
A study by the London School of Economics found that AI could reduce global emissions by 3.2 billion to 5.4 billion tonnes of CO2 equivalent by 2035 if applied in key areas such as innovating resource efficiency, driving behavior change, modeling climate and policy system interventions, and managing resilience and adaptation.
“If done with intention and leveraged in a focused manner using AI production, the impact can be really, really meaningful and positive,” said Mattia Romani, a partner at Systemiq and one of the report authors.
AI can also be used to streamline the collection and access of corporate data. Romani said responsible data sharing is problematic and regulators need to step in to ensure safe data practices and “enable private actors to provide data without risking competition or legal exposure.”
If AI is used for practical applications such as reducing emissions, Romani said its intentional use could justify increased energy costs.
“If we continue to use AI to sell more products on Instagram, I think the emissions associated with the additional power will be significant.”
Moria Costa is an award-winning American journalist based in Paris.
This article was originally published by Green Central Banking. Please see the original text here.
weekly summary
Get all our stories in one place every Wednesday at noon (EST).
