@inproceedings{ghorbanpour-fraser-2026-sensitivity,
title = "On the Sensitivity of Instruction-tuned {LLM}s to Harmful Sentences in Long Inputs",
author = "Ghorbanpour, Faeze and
Fraser, Alexander",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1484/",
pages = "33923--33944",
ISBN = "979-8-89176-459-0",
abstract = "Large language models (LLMs) increasingly operate on long inputs, yet their behavior when harmful sentences are sparsely embedded within such inputs remains poorly understood. We present a sensitivity analysis that probes how well LLMs detect and localize harmful sentences embedded in long mixed inputs, when harmful content is sparsely distributed among non-harmful sentences. We construct long inputs by combining neutral and harmful sentences, and systematically vary four factors: input length (600{--}30,000 tokens), the proportion of harmful sentences (0.05{--}0.50), harm realization (explicit vs. implicit), and the position of harmful sentences within the input (beginning, middle, end), enabling a controlled stress-test evaluation. Experiments across toxic, offensive, and hate content, and across LLaMA-3.1, Qwen-2.5, and Mistral, reveal consistent patterns: sensitivity is non-monotonic with respect to harmful prevalence, peaking at moderate levels; sensitivity degrades as input length increases; harmful sentences placed earlier in the input are more strongly prioritized; and explicit harm is more reliably identified than implicit harm. These findings provide a systematic view of how LLMs prioritize harmful sentences in long input under controlled stress conditions, highlighting both emerging strengths and remaining challenges for safety-related use."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="ghorbanpour-fraser-2026-sensitivity">
<titleInfo>
<title>On the Sensitivity of Instruction-tuned LLMs to Harmful Sentences in Long Inputs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Faeze</namePart>
<namePart type="family">Ghorbanpour</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alexander</namePart>
<namePart type="family">Fraser</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Large language models (LLMs) increasingly operate on long inputs, yet their behavior when harmful sentences are sparsely embedded within such inputs remains poorly understood. We present a sensitivity analysis that probes how well LLMs detect and localize harmful sentences embedded in long mixed inputs, when harmful content is sparsely distributed among non-harmful sentences. We construct long inputs by combining neutral and harmful sentences, and systematically vary four factors: input length (600–30,000 tokens), the proportion of harmful sentences (0.05–0.50), harm realization (explicit vs. implicit), and the position of harmful sentences within the input (beginning, middle, end), enabling a controlled stress-test evaluation. Experiments across toxic, offensive, and hate content, and across LLaMA-3.1, Qwen-2.5, and Mistral, reveal consistent patterns: sensitivity is non-monotonic with respect to harmful prevalence, peaking at moderate levels; sensitivity degrades as input length increases; harmful sentences placed earlier in the input are more strongly prioritized; and explicit harm is more reliably identified than implicit harm. These findings provide a systematic view of how LLMs prioritize harmful sentences in long input under controlled stress conditions, highlighting both emerging strengths and remaining challenges for safety-related use.</abstract>
<identifier type="citekey">ghorbanpour-fraser-2026-sensitivity</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1484/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>33923</start>
<end>33944</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T On the Sensitivity of Instruction-tuned LLMs to Harmful Sentences in Long Inputs
%A Ghorbanpour, Faeze
%A Fraser, Alexander
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F ghorbanpour-fraser-2026-sensitivity
%X Large language models (LLMs) increasingly operate on long inputs, yet their behavior when harmful sentences are sparsely embedded within such inputs remains poorly understood. We present a sensitivity analysis that probes how well LLMs detect and localize harmful sentences embedded in long mixed inputs, when harmful content is sparsely distributed among non-harmful sentences. We construct long inputs by combining neutral and harmful sentences, and systematically vary four factors: input length (600–30,000 tokens), the proportion of harmful sentences (0.05–0.50), harm realization (explicit vs. implicit), and the position of harmful sentences within the input (beginning, middle, end), enabling a controlled stress-test evaluation. Experiments across toxic, offensive, and hate content, and across LLaMA-3.1, Qwen-2.5, and Mistral, reveal consistent patterns: sensitivity is non-monotonic with respect to harmful prevalence, peaking at moderate levels; sensitivity degrades as input length increases; harmful sentences placed earlier in the input are more strongly prioritized; and explicit harm is more reliably identified than implicit harm. These findings provide a systematic view of how LLMs prioritize harmful sentences in long input under controlled stress conditions, highlighting both emerging strengths and remaining challenges for safety-related use.
%U https://aclanthology.org/2026.emnlp-main.1484/
%P 33923-33944
Markdown (Informal)
[On the Sensitivity of Instruction-tuned LLMs to Harmful Sentences in Long Inputs](https://aclanthology.org/2026.emnlp-main.1484/) (Ghorbanpour & Fraser, EMNLP 2026)
ACL