@inproceedings{seth-etal-2026-safetune,
title = "{SAFETUNE} : A Unified, Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned {LLM}s",
author = "Seth, Pratinav and
Sadhu, Saisab and
Kaushal, Anshul and
Sankarapu, Vinay Kumar",
editor = "Agrawal, Sweta and
Abad, Alberto",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-demos.73/",
pages = "771--785",
ISBN = "979-8-89176-461-3",
abstract = "Methods for repairing safety alignment in fine-tuned Large Language Models (LLMs) are evolving quickly, but the literature is fragmented across the repair pipeline: each method is typically tested on its own drifted checkpoint, with its own repair implementation, and judged against its own set of baselines. This makes methods hard to compare or select, and calls for a framework that standardizes drift-induction, repair, and evaluation while remaining flexible enough to support new paradigms and architectures. To address this gap, we introduce SAFETUNE, a source-available library that unifies drift-induction, repair, and evaluation behind a standardized pipeline. SAFETUNE hosts four repair paradigms: weight patching, safety-constrained retraining, gradient-based unlearning, and inference-time steering, with implementations audited and debugged against their original papers. A documented intervention guide suggests a starting repair for a checkpoint{'}s safety-capability profile, with the full pipeline runnable in four lines of code."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="seth-etal-2026-safetune">
<titleInfo>
<title>SAFETUNE : A Unified, Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Pratinav</namePart>
<namePart type="family">Seth</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Saisab</namePart>
<namePart type="family">Sadhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anshul</namePart>
<namePart type="family">Kaushal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vinay</namePart>
<namePart type="given">Kumar</namePart>
<namePart type="family">Sankarapu</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: System Demonstrations</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sweta</namePart>
<namePart type="family">Agrawal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alberto</namePart>
<namePart type="family">Abad</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">979-8-89176-461-3</identifier>
</relatedItem>
<abstract>Methods for repairing safety alignment in fine-tuned Large Language Models (LLMs) are evolving quickly, but the literature is fragmented across the repair pipeline: each method is typically tested on its own drifted checkpoint, with its own repair implementation, and judged against its own set of baselines. This makes methods hard to compare or select, and calls for a framework that standardizes drift-induction, repair, and evaluation while remaining flexible enough to support new paradigms and architectures. To address this gap, we introduce SAFETUNE, a source-available library that unifies drift-induction, repair, and evaluation behind a standardized pipeline. SAFETUNE hosts four repair paradigms: weight patching, safety-constrained retraining, gradient-based unlearning, and inference-time steering, with implementations audited and debugged against their original papers. A documented intervention guide suggests a starting repair for a checkpoint’s safety-capability profile, with the full pipeline runnable in four lines of code.</abstract>
<identifier type="citekey">seth-etal-2026-safetune</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-demos.73/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>771</start>
<end>785</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SAFETUNE : A Unified, Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMs
%A Seth, Pratinav
%A Sadhu, Saisab
%A Kaushal, Anshul
%A Sankarapu, Vinay Kumar
%Y Agrawal, Sweta
%Y Abad, Alberto
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing: System Demonstrations
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-461-3
%F seth-etal-2026-safetune
%X Methods for repairing safety alignment in fine-tuned Large Language Models (LLMs) are evolving quickly, but the literature is fragmented across the repair pipeline: each method is typically tested on its own drifted checkpoint, with its own repair implementation, and judged against its own set of baselines. This makes methods hard to compare or select, and calls for a framework that standardizes drift-induction, repair, and evaluation while remaining flexible enough to support new paradigms and architectures. To address this gap, we introduce SAFETUNE, a source-available library that unifies drift-induction, repair, and evaluation behind a standardized pipeline. SAFETUNE hosts four repair paradigms: weight patching, safety-constrained retraining, gradient-based unlearning, and inference-time steering, with implementations audited and debugged against their original papers. A documented intervention guide suggests a starting repair for a checkpoint’s safety-capability profile, with the full pipeline runnable in four lines of code.
%U https://aclanthology.org/2026.emnlp-demos.73/
%P 771-785
Markdown (Informal)
[SAFETUNE : A Unified, Faithful Library for Auditing and Repairing Safety Drift in Fine-Tuned LLMs](https://aclanthology.org/2026.emnlp-demos.73/) (Seth et al., EMNLP 2026)
ACL