Grounded but Not Immune: Confirmation Bias in Agent-as-a-judge
cs.AI, cs.CL
Submitted: 2026-11-03
Updated: 2026-11-03
Comments: Preprint. Under Review
Project page: https://aj-bench.github.io
License: http://creativecommons.org/licenses/by/4.0/
The gist: Evaluating open-ended agent tasks increasingly falls to an Agent-as-a-Judge: an evaluator handed tools and the agent's full trajectory so it can verify a result against the environment.
Terminology
Abstract
Evaluating open-ended agent tasks increasingly falls to an Agent-as-a-Judge: an evaluator handed tools and the agent's full trajectory so it can verify a result against the environment. The premise is that a judge that gathers its own evidence should be grounded by it. We show that this grounding does not immunize the judge against confirmation bias: a single false pre-screening note, attributed to an authority and carrying no evidential bearing on the correct verdict, overturns up to 40% of a tool-using judge's correct verdicts across four judges and five datasets, even though the judge collects decisive evidence. The shift is driven by specific properties of the cue. We treat this cue as a controlled diagnostic of confirmation bias rather than an adversarial attack. Our central contribution is to explain how confirmation bias changes the judge's behavior in a trajectory. We classify changes in judge behavior and attribute changes to one of three failure modes: over-trust (verify, conclude correctly, then defer), hallucination (fabricate support for the flip), and under-investigation (stop at the first corroborating signal). For the two Claude judges, the failure is predominantly over-trust (67-75% of harmful flips), while smaller judges have error modes spread across all three. Lightweight mitigations blunt but never remove the bias.
Sources
- Chain-of-Thought Reasoning In The Wild Is Not Always Faithful
- J1: Exploring Simple Test-Time Scaling for LLM-as-a-Judge
- From Yes-Men to Truth-Tellers: Addressing Sycophancy in Large Language Models with Pinpoint Tuning
- Herd Behavior: Investigating Peer Influence in LLM-based Multi-Agent Systems
- Large Language Models Cannot Self-Correct Reasoning Yet
- Can You Trick the Grader? Adversarial Persuasion of LLM Judges
- SWE-bench: Can Language Models Resolve Real-World GitHub Issues?
- Measuring Faithfulness in Chain-of-Thought Reasoning
- Are LLM-Judges Robust to Expressions of Uncertainty? Investigating the effect of Epistemic Markers on LLM-based Evaluation
- An Empirical Study on Failures in Automated Issue Solving
- The Silent Judge: Unacknowledged Shortcut Bias in LLM-as-a-Judge
- When Your AI Agent Succumbs to Peer-Pressure: Studying Opinion-Change Dynamics of LLMs
- Don't Judge Code by Its Cover: Exploring Biases in LLM Judges for Code Evaluation
- gpt-oss-120b & gpt-oss-20b Model Card
- LLM Evaluators Recognize and Favor Their Own Generations
- Discovering Language Model Behaviors with Model-Written Evaluations
- Agentic Rubrics as Contextual Verifiers for SWE Agents
- Automation Bias in AI-Assisted Medical Decision-Making under Time Pressure in Computational Pathology
- Towards Understanding Sycophancy in Language Models
- AJ-Bench: Benchmarking Agent-as-a-Judge for Environment-Aware Evaluation
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents