@inproceedings{wang-etal-2026-foresight-enhancing,
title = "{F}ore{S}ight: Enhancing Risk Monitoring via Early Safety Signal Distillation",
author = "Wang, Hanling and
Wei, Chenlong and
Xu, Ling and
Niu, Hanyan and
Cao, Qi and
Huang, Shizhou and
Yang, Yang and
Zhu, Xiaohui and
Zhu, Yao",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.701/",
pages = "14713--14729",
ISBN = "979-8-89176-462-0",
abstract = "As large language models (LLMs) are increasingly deployed, harmful generations have become a critical safety concern. Existing safeguards operate at the input, output, or streaming-generation stages, while early-risk methods that rely on surface tokens or output logits often suffer from weak initial signals, and internals-based detectors using dense representations may retain highly entangled and redundant safety-irrelevant information. It therefore remains unclear whether the earliest post-generation hidden states already contain reliable signals about final-response harmfulness. To address this gap, we propose ForeSight, a first-token output-risk forecasting framework that distills weak and redundant early safety signals into compact, layer-aware risk representations. Experiments across five safety benchmarks and two target models demonstrate that ForeSight achieves superior and efficient early-risk forecasting while using only first-token hidden states. The code is available at: https://anonymous.4open.science/r/Foresight-B067"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-foresight-enhancing">
<titleInfo>
<title>ForeSight: Enhancing Risk Monitoring via Early Safety Signal Distillation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hanling</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chenlong</namePart>
<namePart type="family">Wei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ling</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanyan</namePart>
<namePart type="family">Niu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shizhou</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaohui</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yao</namePart>
<namePart type="family">Zhu</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>As large language models (LLMs) are increasingly deployed, harmful generations have become a critical safety concern. Existing safeguards operate at the input, output, or streaming-generation stages, while early-risk methods that rely on surface tokens or output logits often suffer from weak initial signals, and internals-based detectors using dense representations may retain highly entangled and redundant safety-irrelevant information. It therefore remains unclear whether the earliest post-generation hidden states already contain reliable signals about final-response harmfulness. To address this gap, we propose ForeSight, a first-token output-risk forecasting framework that distills weak and redundant early safety signals into compact, layer-aware risk representations. Experiments across five safety benchmarks and two target models demonstrate that ForeSight achieves superior and efficient early-risk forecasting while using only first-token hidden states. The code is available at: https://anonymous.4open.science/r/Foresight-B067</abstract>
<identifier type="citekey">wang-etal-2026-foresight-enhancing</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.701/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>14713</start>
<end>14729</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ForeSight: Enhancing Risk Monitoring via Early Safety Signal Distillation
%A Wang, Hanling
%A Wei, Chenlong
%A Xu, Ling
%A Niu, Hanyan
%A Cao, Qi
%A Huang, Shizhou
%A Yang, Yang
%A Zhu, Xiaohui
%A Zhu, Yao
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F wang-etal-2026-foresight-enhancing
%X As large language models (LLMs) are increasingly deployed, harmful generations have become a critical safety concern. Existing safeguards operate at the input, output, or streaming-generation stages, while early-risk methods that rely on surface tokens or output logits often suffer from weak initial signals, and internals-based detectors using dense representations may retain highly entangled and redundant safety-irrelevant information. It therefore remains unclear whether the earliest post-generation hidden states already contain reliable signals about final-response harmfulness. To address this gap, we propose ForeSight, a first-token output-risk forecasting framework that distills weak and redundant early safety signals into compact, layer-aware risk representations. Experiments across five safety benchmarks and two target models demonstrate that ForeSight achieves superior and efficient early-risk forecasting while using only first-token hidden states. The code is available at: https://anonymous.4open.science/r/Foresight-B067
%U https://aclanthology.org/2026.findings-emnlp.701/
%P 14713-14729
Markdown (Informal)
[ForeSight: Enhancing Risk Monitoring via Early Safety Signal Distillation](https://aclanthology.org/2026.findings-emnlp.701/) (Wang et al., Findings 2026)
ACL
- Hanling Wang, Chenlong Wei, Ling Xu, Hanyan Niu, Qi Cao, Shizhou Huang, Yang Yang, Xiaohui Zhu, and Yao Zhu. 2026. ForeSight: Enhancing Risk Monitoring via Early Safety Signal Distillation. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 14713–14729, Budapest, Hungary. Association for Computational Linguistics.