Llm reviews are less reliable than static tests suggest
Breaking the Illusion of Review Reliability under Static Evaluation: SCOPE Fuzzing for LLM-based Scientific Reviewers
Artificial Intelligence
Summary
The paper shows that big AI models used for reviewing scientific papers may not be as reliable as previously thought. Previous tests only changed small, predictable parts of papers to see if the AI noticed problems. The authors find this method misses many hidden weaknesses and that AI reviews can be biased depending on the paper's initial score. They built a new tool called SCOPE-Fuzzer that changes papers in smarter, more varied ways to reveal these hidden problems. This tool helps better understand when and how AI reviewers might make mistakes.
What this means in practice
- •For peer review platform developers: Improve AI-based reviewer tools by detecting subtle weaknesses with dynamic content changes to increase review accuracy and fairness.
- •For software testing teams: Adapt the SCOPE-Fuzzer approach to test other AI systems by exploring diverse input mutations guided by feedback for better vulnerability detection.
Authors
Zhuo Chen, Hao Zeng, Jiawei Liu, Guoxiu He, Le Cai, Liu Haotan, Li Wenbo, Yong Huang, Wei Lu
Abstract
The rapid growth of submissions and reviewing workload has accelerated the use of large language models (LLMs) in peer review. Prior studies suggest that LLM-based reviewers can penalize content perturbations, such as overclaiming, indicating a certain degree of reliability. Yet these conclusions are largely based on a narrow set of perturbation strategies instantiated with static templates, providing limited evidence of actual reliability. In this paper, we construct a three-level evaluation framework covering perturbations to surface presentation, argumentative logic, and value judgment. Experiments on representative LLM-based reviewers reveal two limitations of static evaluation: stratified vulnerability, where perturbation effects depend on whether the paper's original review score is high or low, and perturbation undercoverage, where a single template misses vulnerabilities exposed by diverse realizations. To address these limitations, we propose SCOPE-Fuzzer, a strategy-aware fuzzer that combines feedback-driven strategy selection with adaptive mutation of paper content. By iteratively probing reviewers with dynamic perturbations, SCOPE-Fuzzer consistently uncovers vulnerabilities overlooked by static evaluation and other baselines.