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

Iván Arcuschin, David Chanin, Adrià Garriga-Alonso +1

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. Monitorin…