When Auditors Fabricate: Batch-Size Degradation and Confident Hallucination in LLM Detection of Planted Document Contamination

Abstract

Large language models are increasingly proposed as automated auditors of document quality, yet their reliability as detectors of planted errors is poorly characterised. We construct a contaminated corpus of 150 academic papers spanning supply chain management and medical research, injecting 450 known contaminants of three types: typographical corruption, semantic reversal, and absurd out-of-context insertion. We then evaluate Google Gemini 3.0 Pro's ability to recover a 180-contaminant answer-ke

Categories

Framework mappings

Suggested from the entry's categories.

@misc{parekh2026when,
  title = {{When Auditors Fabricate: Batch-Size Degradation and Confident Hallucination in LLM Detection of Planted Document Contamination}},
  author = {Karan Parekh and Sanjana Pendyala Ravinder and Sana Mhapsekar and Medina Maloku},
  year = {2026},
  month = sep,
  eprint = {2609.09696},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.09696}
}