paperFebruary 2026Unreviewed

A Coin Flip for Safety: LLM Judges Fail to Reliably Measure Adversarial Robustness

Leo Schwinn, Moritz Ladenburger, Tim Beyer, Mehrnaz Mofakhami, G. Gidel, Stephan Gunnemann

arXiv.org

Abstract

Automated \enquote{LLM-as-a-Judge} frameworks have become the de facto standard for scalable evaluation across natural language processing. For instance, in safety evaluation, these judges are relied upon to evaluate harmfulness in order to benchmark the robustness of safety against adversarial attacks. However, we show that existing validation protocols fail to account for substantial distribution shifts inherent to red-teaming: diverse victim models exhibit distinct generation styles, attacks

Framework mappings

MITRE ATLAS

AML.T0043Craft Adversarial Data

NIST AI Risk Management Framework

MEASUREMeasure

Suggested from the entry's categories.

Cite

@misc{schwinn2026coin,
  title = {{A Coin Flip for Safety: LLM Judges Fail to Reliably Measure Adversarial Robustness}},
  author = {Leo Schwinn and Moritz Ladenburger and Tim Beyer and Mehrnaz Mofakhami and G. Gidel and Stephan Gunnemann},
  year = {2026},
  month = feb,
  eprint = {2603.06594},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2603.06594},
  url = {https://www.semanticscholar.org/paper/e8bd6ec12610833e7de437d7b2a5b4fc7c89fd9d}
}