Segment-Level Coherence for Robust Harmful Intent Probing in LLMs

Abstract

Large Language Models (LLMs) are increasingly exposed to adaptive jailbreaking, particularly in high-stakes Chemical, Biological, Radiological, and Nuclear (CBRN) domains. Although streaming probes enable real-time monitoring, they still make systematic errors. We identify a core issue: existing methods often rely on a few high-scoring tokens, leading to false alarms when sensitive CBRN terms appear in benign contexts. To address this, we introduce a streaming probing objective that requires mul

Categories

Framework mappings

Suggested from the entry's categories.

@misc{he2026segmentlevel,
  title = {{Segment-Level Coherence for Robust Harmful Intent Probing in LLMs}},
  author = {Xuanli He and Bilgehan Sel and Faizan Ali and Jenny Bao and Hoagy Cunningham and Jerry Wei},
  year = {2026},
  month = apr,
  eprint = {2604.14865},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2604.14865}
}