← 研踪 PaperTrailarXiv:2610.27908 · v2
Polarity Blindness in LLM Safety: A Systematic Study of Negation, Antonym Substitution, and Adversarial Framing
首次提交 2026-10-20 17:22
版本更新 2026-11-09 22:34
雷达观测 2026-11-10 15:22
摘要
Negation remains a persistent challenge for language models across NLP tasks and modalities. We show that this weakness can undermine LLM safety guardrails through a structural vulnerability we term polarity blindness: the failure to recognize harmful intent after polarity-altering linguistic transformations. We study two forms of polarity inversion: grammatical negation and antonym replacement, and use synonym replacement as a polarity-preserving control. We further test whether an adversarial textual wrapper amplifies these failures. Using a 260-prompt benchmark across 10 state-of-the-art LLMs and 3 languages (English, Spanish, Czech), stratified by harmfulness potential and evaluated on a validated 0-5 harm scale, we find that polarity blindness affects most evaluated models, with mean harm scores reaching 4.12 for the most vulnerable model. The effect persists across typologically differing negation systems, suggesting that polarity blindness reflects a broader limitation in semantic robustness rather than a language- or grammar-specific failure.
作者备注：Code and data: https://github.com/dmlls/llm-safety-polarity-blindness
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v2 · 2026-11-09 22:34 · 最新观测版本
Polarity Blindness in LLM Safety: A Systematic Study of Negation, Antonym Substitution, and Adversarial Framing
Negation remains a persistent challenge for language models across NLP tasks and modalities. We show that this weakness can undermine LLM safety guardrails through a structural vulnerability we term polarity blindness: the failure to recognize harmful intent after polarity-altering linguistic transformations. We study two forms of polarity inversion: grammatical negation and antonym replacement, and use synonym replacement as a polarity-preserving control. We further test whether an adversarial textual wrapper amplifies these failures. Using a 260-prompt benchmark across 10 state-of-the-art LLMs and 3 languages (English, Spanish, Czech), stratified by harmfulness potential and evaluated on a validated 0-5 harm scale, we find that polarity blindness affects most evaluated models, with mean harm scores reaching 4.12 for the most vulnerable model. The effect persists across typologically differing negation systems, suggesting that polarity blindness reflects a broader limitation in semantic robustness rather than a language- or grammar-specific failure.
查看此版本解读 →查看此版本原文 ↗