# Inside the Unfair Judge: A Mechanistic Interpretability Account of LLM-as-Judge Bias

arXiv: https://arxiv.org/abs/2607.11871 · HTML: https://mlsanity.com/paper/2607.11871/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Probing deception, bias and beliefs, one of 13 directions in Mechanistic interpretability.
- Builds on 36 of its 114 references that are in the map.
- First posted on arXiv 13 Jul 2026.

## Placed in

- Mechanistic interpretability / [Probing deception, bias and beliefs](https://mlsanity.com/mechinterp/probing-deception-bias-and-beliefs/)

## Builds on (most cited first, 30 of 36)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [DeepSeek-V3 Technical Report](https://mlsanity.com/paper/2412.19437/) (2024)
- [Phi-4 Technical Report](https://mlsanity.com/paper/2412.08905/) (2024)
- [Injecting Domain-Specific Knowledge into Large Language Models: A Comprehensive Survey](https://mlsanity.com/paper/2502.10708/) (2025)
- [DrugAgent: Automating AI-aided Drug Discovery Programming through LLM Multi-Agent Collaboration](https://mlsanity.com/paper/2411.15692/) (2024)
- [Treble Counterfactual VLMs: A Causal Approach to Hallucination](https://mlsanity.com/paper/2503.06169/) (2025)
- [Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments](https://mlsanity.com/paper/2602.01244/) (2026)
- [DPU: Dynamic Prototype Updating for Multimodal Out-of-Distribution Detection](https://mlsanity.com/paper/2411.08227/) (2024)
- [Building a Foundational Guardrail for General Agentic Systems via Synthetic Data](https://mlsanity.com/paper/2510.09781/) (2025)
- [AD-LLM: Benchmarking Large Language Models for Anomaly Detection](https://mlsanity.com/paper/2412.11142/) (2024)
- [Ponder & Press: Advancing Visual GUI Agent towards General Computer Control](https://mlsanity.com/paper/2412.01268/) (2024)
- [Graph Neural Network Explanations are Fragile](https://mlsanity.com/paper/2406.03193/) (2024)
- [Auditing Multi-Agent LLM Reasoning Trees Outperforms Majority Vote and LLM-as-Judge](https://mlsanity.com/paper/2602.09341/) (2026)
- [Do LLMs "Feel"? Emotion Circuits Discovery and Control](https://mlsanity.com/paper/2510.11328/) (2025)
- [StealthRank: LLM Ranking Manipulation via Stealthy Prompt Optimization](https://mlsanity.com/paper/2504.05804/) (2025)
- [SocialMaze: A Benchmark for Evaluating and Enhancing Social Reasoning in Large Language Models in Complex Social Environments](https://mlsanity.com/paper/2505.23713/) (2025)
- [Defenses Against Prompt Attacks Learn Surface Heuristics](https://mlsanity.com/paper/2601.07185/) (2026)
- [A Fano-Style Accuracy Upper Bound for LLM Single-Pass Reasoning in Multi-Hop QA](https://mlsanity.com/paper/2509.21199/) (2025)
- [ProbeLLM: Automating Principled Diagnosis of LLM Failures](https://mlsanity.com/paper/2602.12966/) (2026)
- [MMAC-Copilot: Multi-modal Agent Collaboration Operating Copilot](https://mlsanity.com/paper/2404.18074/) (2024)
- [AD-AGENT: A Multi-agent Framework for End-to-end Anomaly Detection](https://mlsanity.com/paper/2505.12594/) (2025)
- [QUITE: A Query Rewrite System Beyond Rules with LLM Agents](https://mlsanity.com/paper/2506.07675/) (2025)
- [Provably Robust Explainable Graph Neural Networks against Graph Perturbation Attacks](https://mlsanity.com/paper/2502.04224/) (2025)
- [Graph Synthetic Out-of-Distribution Exposure with Large Language Models](https://mlsanity.com/paper/2504.21198/) (2025)
- [Evaluate Bias without Manual Test Sets: A Concept Representation Perspective for LLMs](https://mlsanity.com/paper/2505.15524/) (2025)
- ["Someone Hid It": Query-Agnostic Black-Box Attacks on LLM-Based Retrieval](https://mlsanity.com/paper/2602.00364/) (2026)
- [Emergent Risks in Generative Multi-Agent Systems](https://mlsanity.com/paper/2603.27771/) (2026)
- [The Cylindrical Representation Hypothesis for Language Model Steering](https://mlsanity.com/paper/2605.01844/) (2026)
- [Under the Shadow of Babel: How Language Shapes Reasoning in LLMs](https://mlsanity.com/paper/2506.16151/) (2025)

## Built on by (most cited first, 3 of 3)

- [A Judge Should Know What Changed:Construct Validity for LLM-as-a-Judge Evaluation](https://mlsanity.com/paper/2608.24419/) (2026)
- [GTA: Graph Theory Agent and Benchmark for Algorithmic Graph Reasoning with LLMs](https://mlsanity.com/paper/2609.12265/) (2026)
- [Do MLLM Judges Judge the Edit? Auditing Bias in Image Editing Evaluation with Verified Quality Preservation](https://mlsanity.com/paper/2610.01670/) (2026)

## Abstract (arXiv)

Existing studies of LLM-as-judge scoring bias work predominantly at the input-output level: they perturb inputs, measure score deltas, and propose prompt-level mitigations. We argue that the same biases admit a representation-level account in the judge's hidden state, complementary to the input-output view and operationally useful in ways it does not afford. We report three findings, across seven judges, seven bias types, and nine benchmarks. Geometry: baseline judging inputs occupy a tight activation manifold while biased inputs are displaced along a low-dimensional, type-specific subspace that sharpens with depth and is recovered consistently by three families of estimators. Causal control: steering hidden states along this subspace drives scoring in both directions, forward shifts reproducing biased scoring on clean inputs and reverse shifts restoring baseline scoring on biased ones, while matched-norm random directions produce shifts an order of magnitude smaller. Operational: a simple linear projection onto the same bias-direction features anticipates judge failures on three entirely unseen benchmarks, substantially outperforming text-based alternatives. Reading bias as activation geometry, rather than as input-output noise, unifies geometric structure, causal control, and operational prediction within a single framework. The project page is available at https://xzx34.github.io/unfair-judge/