@inproceedings{vetter-etal-2026-data,
title = "Data Attribution of Emergent Misalignment with Persona Features",
author = "Vetter, Clemens and
Kacz{\'e}r, David and
Flek, Lucie and
Mai, Florian",
editor = "Arad, Dana and
Belinkov, Yonatan and
Chen, Hanjie and
Feldhus, Nils and
Mueller, Aaron and
Mohebbi, Hosein and
Sarti, Gabriele and
Tutek, Martin",
booktitle = "Proceedings of the 9th {B}lackbox{NLP} Workshop: Analyzing and Interpreting Neural Networks for {NLP}",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.blackboxnlp-1.10/",
pages = "136--165",
ISBN = "979-8-89176-476-7",
abstract = "Emergent misalignment (EM) is the phenomenon where fine-tuning a language model on a narrow task leads to harmful behavior in unrelated domains. A leading mechanistic account attributes EM to persona features: latent directions representing misaligned personas, acquired during pre-training, are amplified during fine-tuning on misaligned data. We ask where these features come from, and whether naturally occurring human-written text suffices to induce EM. Using Sparse Autoencoder (SAE) based model diffing across four open-weight models, we find that misalignment fine-tuning amplifies features related to jailbreak personas, sarcasm, deception, and manipulation, while suppressing safety-relevant and assistant-identity features. Steering individual features controls EM in both directions: it induces misalignment rates of up to 62{\%} in aligned models and re-aligns misaligned models to near-baseline rates. Attributing the causal features to one million web documents from an open pre-training corpus retrieves recurring narratives about villainous characters, domination, and harmful agency. Yet fine-tuning on these human-written documents does not reliably induce EM, whereas synthetic instruction-response pairs derived from the same content do, even across models. Semantic relevance alone is thus not sufficient: response structure or model-generated phrasing plays an important role in inducing EM."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="vetter-etal-2026-data">
<titleInfo>
<title>Data Attribution of Emergent Misalignment with Persona Features</title>
</titleInfo>
<name type="personal">
<namePart type="given">Clemens</namePart>
<namePart type="family">Vetter</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Kaczér</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lucie</namePart>
<namePart type="family">Flek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Florian</namePart>
<namePart type="family">Mai</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 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dana</namePart>
<namePart type="family">Arad</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yonatan</namePart>
<namePart type="family">Belinkov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanjie</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nils</namePart>
<namePart type="family">Feldhus</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aaron</namePart>
<namePart type="family">Mueller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hosein</namePart>
<namePart type="family">Mohebbi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gabriele</namePart>
<namePart type="family">Sarti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Martin</namePart>
<namePart type="family">Tutek</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">979-8-89176-476-7</identifier>
</relatedItem>
<abstract>Emergent misalignment (EM) is the phenomenon where fine-tuning a language model on a narrow task leads to harmful behavior in unrelated domains. A leading mechanistic account attributes EM to persona features: latent directions representing misaligned personas, acquired during pre-training, are amplified during fine-tuning on misaligned data. We ask where these features come from, and whether naturally occurring human-written text suffices to induce EM. Using Sparse Autoencoder (SAE) based model diffing across four open-weight models, we find that misalignment fine-tuning amplifies features related to jailbreak personas, sarcasm, deception, and manipulation, while suppressing safety-relevant and assistant-identity features. Steering individual features controls EM in both directions: it induces misalignment rates of up to 62% in aligned models and re-aligns misaligned models to near-baseline rates. Attributing the causal features to one million web documents from an open pre-training corpus retrieves recurring narratives about villainous characters, domination, and harmful agency. Yet fine-tuning on these human-written documents does not reliably induce EM, whereas synthetic instruction-response pairs derived from the same content do, even across models. Semantic relevance alone is thus not sufficient: response structure or model-generated phrasing plays an important role in inducing EM.</abstract>
<identifier type="citekey">vetter-etal-2026-data</identifier>
<location>
<url>https://aclanthology.org/2026.blackboxnlp-1.10/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>136</start>
<end>165</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Data Attribution of Emergent Misalignment with Persona Features
%A Vetter, Clemens
%A Kaczér, David
%A Flek, Lucie
%A Mai, Florian
%Y Arad, Dana
%Y Belinkov, Yonatan
%Y Chen, Hanjie
%Y Feldhus, Nils
%Y Mueller, Aaron
%Y Mohebbi, Hosein
%Y Sarti, Gabriele
%Y Tutek, Martin
%S Proceedings of the 9th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-476-7
%F vetter-etal-2026-data
%X Emergent misalignment (EM) is the phenomenon where fine-tuning a language model on a narrow task leads to harmful behavior in unrelated domains. A leading mechanistic account attributes EM to persona features: latent directions representing misaligned personas, acquired during pre-training, are amplified during fine-tuning on misaligned data. We ask where these features come from, and whether naturally occurring human-written text suffices to induce EM. Using Sparse Autoencoder (SAE) based model diffing across four open-weight models, we find that misalignment fine-tuning amplifies features related to jailbreak personas, sarcasm, deception, and manipulation, while suppressing safety-relevant and assistant-identity features. Steering individual features controls EM in both directions: it induces misalignment rates of up to 62% in aligned models and re-aligns misaligned models to near-baseline rates. Attributing the causal features to one million web documents from an open pre-training corpus retrieves recurring narratives about villainous characters, domination, and harmful agency. Yet fine-tuning on these human-written documents does not reliably induce EM, whereas synthetic instruction-response pairs derived from the same content do, even across models. Semantic relevance alone is thus not sufficient: response structure or model-generated phrasing plays an important role in inducing EM.
%U https://aclanthology.org/2026.blackboxnlp-1.10/
%P 136-165
Markdown (Informal)
[Data Attribution of Emergent Misalignment with Persona Features](https://aclanthology.org/2026.blackboxnlp-1.10/) (Vetter et al., BlackboxNLP 2026)
ACL