Researchers have been exploring whether large language models can be used to explain complex decisions made by credit risk models. In a study published on arXiv, the authors evaluated three pipelines that use different types of data and explanation methods, including one that combines tabular and network data. The results show that while the language models can generate narratives that name the influential factors in a decision, they are less reliable when stating the direction of influence. This has implications for the governance of risk models and the deployment of large language models in regulated credit settings. --- Why it matters: This study matters to AI researchers because it highlights the importance of explanation quality in high-stakes tasks like credit decisioning. The findings also have specific implications for the use of large language models in regulated industries, where transparency and accountability are crucial. Source: https://arxiv.org/abs/2608.17715