paperMarch 2025Unreviewed

Know Thy Judge: On the Robustness Meta-Evaluation of LLM Safety Judges

Francisco Eiras, Eliott Zemour, Eric Lin, Vaikkunth Mugunthan

arXiv.org

Abstract

Large Language Model (LLM) based judges form the underpinnings of key safety evaluation processes such as offline benchmarking, automated red-teaming, and online guardrailing. This widespread requirement raises the crucial question: can we trust the evaluations of these evaluators? In this paper, we highlight two critical challenges that are typically overlooked: (i) evaluations in the wild where factors like prompt sensitivity and distribution shifts can affect performance and (ii) adversarial

Framework mappings

NIST AI Risk Management Framework

MEASUREMeasure

Suggested from the entry's categories.

Cite

@misc{eiras2025know,
  title = {{Know Thy Judge: On the Robustness Meta-Evaluation of LLM Safety Judges}},
  author = {Francisco Eiras and Eliott Zemour and Eric Lin and Vaikkunth Mugunthan},
  year = {2025},
  month = mar,
  eprint = {2503.04474},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2503.04474},
  url = {https://www.semanticscholar.org/paper/0ffb356aab98ae69c717f8b2969c3fed0592a048}
}