@inproceedings{manoranjan-etal-2026-shard,
title = "{SHARD}: Safe and Helpful Alignment via Self-Reframing Distillation",
author = "Manoranjan, Viswonathan and
Gupta, Amogh and
Vijjini, Anvesh Rao and
Hofweber, Thomas and
Chaturvedi, Snigdha",
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.1460/",
pages = "31076--31103",
ISBN = "979-8-89176-462-0",
abstract = "Large language models often struggle with sensitive prompts. They may refuse outright, provide generic safety boilerplate, or fail to address the user{'}s legitimate informational needs that can be answered safely. We introduce SHARD, a self-reframing distillation method to improve safety-helpfulness. It first rewrites sensitive prompts to surface benign intent using philosophical guidelines, then reframes its original responses into safe, more helpful ones, and finally fine-tunes the model on its self-reframed responses. Across DNA and the English subset of LINGUASAFE, SHARD improves helpfulness for most model families while preserving safety. It also remains competitive with distillation from a larger teacher model, suggesting that models can internalize safe and helpful behavior elicited from the models themselves."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="manoranjan-etal-2026-shard">
<titleInfo>
<title>SHARD: Safe and Helpful Alignment via Self-Reframing Distillation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Viswonathan</namePart>
<namePart type="family">Manoranjan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Amogh</namePart>
<namePart type="family">Gupta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anvesh</namePart>
<namePart type="given">Rao</namePart>
<namePart type="family">Vijjini</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thomas</namePart>
<namePart type="family">Hofweber</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Snigdha</namePart>
<namePart type="family">Chaturvedi</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>Large language models often struggle with sensitive prompts. They may refuse outright, provide generic safety boilerplate, or fail to address the user’s legitimate informational needs that can be answered safely. We introduce SHARD, a self-reframing distillation method to improve safety-helpfulness. It first rewrites sensitive prompts to surface benign intent using philosophical guidelines, then reframes its original responses into safe, more helpful ones, and finally fine-tunes the model on its self-reframed responses. Across DNA and the English subset of LINGUASAFE, SHARD improves helpfulness for most model families while preserving safety. It also remains competitive with distillation from a larger teacher model, suggesting that models can internalize safe and helpful behavior elicited from the models themselves.</abstract>
<identifier type="citekey">manoranjan-etal-2026-shard</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1460/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>31076</start>
<end>31103</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SHARD: Safe and Helpful Alignment via Self-Reframing Distillation
%A Manoranjan, Viswonathan
%A Gupta, Amogh
%A Vijjini, Anvesh Rao
%A Hofweber, Thomas
%A Chaturvedi, Snigdha
%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 manoranjan-etal-2026-shard
%X Large language models often struggle with sensitive prompts. They may refuse outright, provide generic safety boilerplate, or fail to address the user’s legitimate informational needs that can be answered safely. We introduce SHARD, a self-reframing distillation method to improve safety-helpfulness. It first rewrites sensitive prompts to surface benign intent using philosophical guidelines, then reframes its original responses into safe, more helpful ones, and finally fine-tunes the model on its self-reframed responses. Across DNA and the English subset of LINGUASAFE, SHARD improves helpfulness for most model families while preserving safety. It also remains competitive with distillation from a larger teacher model, suggesting that models can internalize safe and helpful behavior elicited from the models themselves.
%U https://aclanthology.org/2026.findings-emnlp.1460/
%P 31076-31103
Markdown (Informal)
[SHARD: Safe and Helpful Alignment via Self-Reframing Distillation](https://aclanthology.org/2026.findings-emnlp.1460/) (Manoranjan et al., Findings 2026)
ACL
- Viswonathan Manoranjan, Amogh Gupta, Anvesh Rao Vijjini, Thomas Hofweber, and Snigdha Chaturvedi. 2026. SHARD: Safe and Helpful Alignment via Self-Reframing Distillation. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 31076–31103, Budapest, Hungary. Association for Computational Linguistics.