Intel / article

Chain-of-Models: Cross-Model Auditing for Bias-Robust LLM Judges

Source: arxiv.org
Author: Qian Wang, Zhanzhi Lou, Zhenheng Tang, Nuo Chen, Bingsheng He
Date: Published Aug 3, 2026

Context

Choosing an AI reviewer requires testing how it audits another model, because its own standalone performance may not predict that . Chain-of-Models has a second model inspect the first model's reasoning before producing a final judgment. Across nine models, four bias types, and four factual datasets, the researchers found that the best auditor depended on the bias being tested. A selection rule given the bias type achieved 88.4% accuracy on the held-out , compared with 82.4% for the strongest fixed auditor. The result supports evaluating reviewer pairings for a defined failure mode rather than choosing solely by standalone scores.

Terms in this piece · Glossary

agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
LLM-as-judge — Using one model to score another's output against a rubric, so quality can be measured at a scale human grading cannot reach.

Key quotes

“standalone bias resistance does not predict audit effectiveness”

“the best auditor is bias-specific”

Read the source arxiv.org