TASR: A Trustworthy LLM-based Framework for TCFD-Aligned Sustainability Report Analysis

Bo Huang, Sheng Yang, Wenjun Lin, Yan Yan, Jing Lu
Proceedings of the The 39th Canadian Conference on Artificial Intelligence, PMLR 318:77-88, 2026.

Abstract

Reliable and transparent assessment of environmental, social, and governance (ESG) disclosures is critical for sustainable finance, regulatory oversight, and risk-aware decision-making. However, existing sustainability reporting evaluations rely on costly manual reviews or third-party ratings, which limit reproducibility. This work proposes a trustworthy large language model (LLM)-based framework for automated sustainability report analysis aligned with the Task Force on Climate-related Financial Disclosures (TCFD). We propose TASR (Trustworthy Analysis for Sustainability Report), a three-stage framework for TCFD-aligned sustainability report analysis that integrates LLM-based scoring, benchmarking against third-party ESG ratings, and downstream predictive modeling. Experiments on 100 sustainability reports from U.S. oil, gas, and mining companies demonstrate strong alignment with Bloomberg Environmental Disclosure scores (Spearman’s $\rho$ = 0.70) and high score stability across repeated evaluations. Furthermore, predictive models trained on the LLM-generated TCFD scores achieve meaningful predictive performance in forecasting disclosure benchmarks, highlighting their practical utility for sustainability rating. The results suggest that LLM-based TCFD scoring offers a potentially scalable and transparent alternative for sustainability disclosure assessment.

Cite this Paper

BibTeX
@InProceedings{pmlr-v318-huang26a, title = {TASR: A Trustworthy LLM-based Framework for TCFD-Aligned Sustainability Report Analysis}, author = {Huang, Bo and Yang, Sheng and Lin, Wenjun and Yan, Yan and Lu, Jing}, booktitle = {Proceedings of the The 39th Canadian Conference on Artificial Intelligence}, pages = {77--88}, year = {2026}, editor = {Bouzar-Benlabiod, Lydia and Leung, Carson}, volume = {318}, series = {Proceedings of Machine Learning Research}, month = {25--29 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v318/main/assets/huang26a/huang26a.pdf}, url = {https://proceedings.mlr.press/v318/huang26a.html}, abstract = {Reliable and transparent assessment of environmental, social, and governance (ESG) disclosures is critical for sustainable finance, regulatory oversight, and risk-aware decision-making. However, existing sustainability reporting evaluations rely on costly manual reviews or third-party ratings, which limit reproducibility. This work proposes a trustworthy large language model (LLM)-based framework for automated sustainability report analysis aligned with the Task Force on Climate-related Financial Disclosures (TCFD). We propose TASR (Trustworthy Analysis for Sustainability Report), a three-stage framework for TCFD-aligned sustainability report analysis that integrates LLM-based scoring, benchmarking against third-party ESG ratings, and downstream predictive modeling. Experiments on 100 sustainability reports from U.S. oil, gas, and mining companies demonstrate strong alignment with Bloomberg Environmental Disclosure scores (Spearman’s $\rho$ = 0.70) and high score stability across repeated evaluations. Furthermore, predictive models trained on the LLM-generated TCFD scores achieve meaningful predictive performance in forecasting disclosure benchmarks, highlighting their practical utility for sustainability rating. The results suggest that LLM-based TCFD scoring offers a potentially scalable and transparent alternative for sustainability disclosure assessment.} }
Endnote
%0 Conference Paper %T TASR: A Trustworthy LLM-based Framework for TCFD-Aligned Sustainability Report Analysis %A Bo Huang %A Sheng Yang %A Wenjun Lin %A Yan Yan %A Jing Lu %B Proceedings of the The 39th Canadian Conference on Artificial Intelligence %C Proceedings of Machine Learning Research %D 2026 %E Lydia Bouzar-Benlabiod %E Carson Leung %F pmlr-v318-huang26a %I PMLR %P 77--88 %U https://proceedings.mlr.press/v318/huang26a.html %V 318 %X Reliable and transparent assessment of environmental, social, and governance (ESG) disclosures is critical for sustainable finance, regulatory oversight, and risk-aware decision-making. However, existing sustainability reporting evaluations rely on costly manual reviews or third-party ratings, which limit reproducibility. This work proposes a trustworthy large language model (LLM)-based framework for automated sustainability report analysis aligned with the Task Force on Climate-related Financial Disclosures (TCFD). We propose TASR (Trustworthy Analysis for Sustainability Report), a three-stage framework for TCFD-aligned sustainability report analysis that integrates LLM-based scoring, benchmarking against third-party ESG ratings, and downstream predictive modeling. Experiments on 100 sustainability reports from U.S. oil, gas, and mining companies demonstrate strong alignment with Bloomberg Environmental Disclosure scores (Spearman’s $\rho$ = 0.70) and high score stability across repeated evaluations. Furthermore, predictive models trained on the LLM-generated TCFD scores achieve meaningful predictive performance in forecasting disclosure benchmarks, highlighting their practical utility for sustainability rating. The results suggest that LLM-based TCFD scoring offers a potentially scalable and transparent alternative for sustainability disclosure assessment.
APA
Huang, B., Yang, S., Lin, W., Yan, Y. & Lu, J.. (2026). TASR: A Trustworthy LLM-based Framework for TCFD-Aligned Sustainability Report Analysis. Proceedings of the The 39th Canadian Conference on Artificial Intelligence, in Proceedings of Machine Learning Research 318:77-88 Available from https://proceedings.mlr.press/v318/huang26a.html.

Related Material