AI Mortgage Tools Show Stark Bias Against Non-English Names

Leading general-purpose AI models demonstrate a pronounced bias, incorrectly identifying transactions linked to non-English names as potentially foreign at a significantly higher rate than those with English names. This issue is critical as the mortgage industry rapidly adopts AI, with a substantial majority of lenders exploring or implementing the technology in their workflows.

To address these concerns, researchers have introduced MortarBench, an open-source benchmark tool. This platform allows for standardized testing of AI models against common mortgage origination tasks, aiming to provide a clearer understanding of their accuracy and potential biases for lenders, regulators, and consumers alike.