Biases in the Blind Spot: Detecting What LLMs Fail to Mention

TL;DR

This paper presents an automated, black-box method for detecting hidden, task-specific biases in large language models by analyzing their reasoning traces without relying on predefined bias categories.

Contribution

The authors introduce a novel pipeline that automatically uncovers unverbalized biases in LLMs, surpassing traditional manual bias detection methods.

Findings

Automatically discovers previously unknown biases such as language fluency and formality.

Validates known biases like gender, race, and religion.

Works across multiple models and decision tasks.

Abstract

Large Language Models (LLMs) often provide chain-of-thought (CoT) reasoning traces that appear plausible, but may hide internal biases. We call these *unverbalized biases*. Monitoring models via their stated reasoning is therefore unreliable, and existing bias evaluations typically require predefined categories and hand-crafted datasets. In this work, we introduce a fully automated, black-box pipeline for detecting task-specific unverbalized biases. Given a task dataset, the pipeline uses LLM autoraters to generate candidate bias concepts. It then tests each concept on progressively larger input samples by generating positive and negative variations, and applies statistical techniques for multiple testing and early stopping. A concept is flagged as an unverbalized bias if it yields statistically significant performance differences while not being cited as justification in the model's CoTs. We evaluate our pipeline across seven LLMs on three decision tasks (hiring, loan approval, and university admissions). Our technique automatically discovers previously unknown biases in these models (e.g., Spanish fluency, English proficiency, writing formality). In the same run, the pipeline also validates biases that were manually identified by prior work (gender, race, religion, ethnicity). More broadly, our proposed approach provides a practical, scalable path to automatic task-specific bias discovery.