@inproceedings{hong-etal-2026-interference,
title = "Interference-Aware Composition of Steering Vectors",
author = "Hong, Giwon and
Radevski, Gorjan and
Glava{\v{s}}, Goran and
Gashteovski, Kiril and
Ponti, Edoardo and
Lawrence, Carolin and
Minervini, Pasquale",
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.1024/",
pages = "21627--21658",
ISBN = "979-8-89176-462-0",
abstract = "Steering vectors are a flexible mechanism for controlling large language models (LLMs) along interpretable behaviours. Multi-behaviour steering, however, remains challenging due to interference. Individual steering vectors have conflicting directions or override one another, which not only impedes controllability but also often degrades generation results. In this work, we first analyse how multi-behaviour compositions of steering vectors affect model outputs, characterising how interactions between steering directions result in non-additive, position-dependent distributional shifts in generation. We then introduce an interference-aware allocation (IAA) method for multi-behaviour steering that minimises the cross-behaviour interference by dynamically adjusting the magnitudes of steering vectors at each generation step. Without instruction prompting, our method ties the best average on Sycophancy (76.4), achieves the best average on Refusal (80.8), and is far more fluent on Myopic-reward than the strongest baseline (67.1 vs. 41.7). With full behavioural instructions, it improves the all-prompt average from 88.2/94.8/81.2 to 95.7/98.7/95.6 while maintaining high fluency (87.3/90.5/87.7)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hong-etal-2026-interference">
<titleInfo>
<title>Interference-Aware Composition of Steering Vectors</title>
</titleInfo>
<name type="personal">
<namePart type="given">Giwon</namePart>
<namePart type="family">Hong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gorjan</namePart>
<namePart type="family">Radevski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Goran</namePart>
<namePart type="family">Glavaš</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kiril</namePart>
<namePart type="family">Gashteovski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Edoardo</namePart>
<namePart type="family">Ponti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolin</namePart>
<namePart type="family">Lawrence</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pasquale</namePart>
<namePart type="family">Minervini</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>Steering vectors are a flexible mechanism for controlling large language models (LLMs) along interpretable behaviours. Multi-behaviour steering, however, remains challenging due to interference. Individual steering vectors have conflicting directions or override one another, which not only impedes controllability but also often degrades generation results. In this work, we first analyse how multi-behaviour compositions of steering vectors affect model outputs, characterising how interactions between steering directions result in non-additive, position-dependent distributional shifts in generation. We then introduce an interference-aware allocation (IAA) method for multi-behaviour steering that minimises the cross-behaviour interference by dynamically adjusting the magnitudes of steering vectors at each generation step. Without instruction prompting, our method ties the best average on Sycophancy (76.4), achieves the best average on Refusal (80.8), and is far more fluent on Myopic-reward than the strongest baseline (67.1 vs. 41.7). With full behavioural instructions, it improves the all-prompt average from 88.2/94.8/81.2 to 95.7/98.7/95.6 while maintaining high fluency (87.3/90.5/87.7).</abstract>
<identifier type="citekey">hong-etal-2026-interference</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1024/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>21627</start>
<end>21658</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Interference-Aware Composition of Steering Vectors
%A Hong, Giwon
%A Radevski, Gorjan
%A Glavaš, Goran
%A Gashteovski, Kiril
%A Ponti, Edoardo
%A Lawrence, Carolin
%A Minervini, Pasquale
%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 hong-etal-2026-interference
%X Steering vectors are a flexible mechanism for controlling large language models (LLMs) along interpretable behaviours. Multi-behaviour steering, however, remains challenging due to interference. Individual steering vectors have conflicting directions or override one another, which not only impedes controllability but also often degrades generation results. In this work, we first analyse how multi-behaviour compositions of steering vectors affect model outputs, characterising how interactions between steering directions result in non-additive, position-dependent distributional shifts in generation. We then introduce an interference-aware allocation (IAA) method for multi-behaviour steering that minimises the cross-behaviour interference by dynamically adjusting the magnitudes of steering vectors at each generation step. Without instruction prompting, our method ties the best average on Sycophancy (76.4), achieves the best average on Refusal (80.8), and is far more fluent on Myopic-reward than the strongest baseline (67.1 vs. 41.7). With full behavioural instructions, it improves the all-prompt average from 88.2/94.8/81.2 to 95.7/98.7/95.6 while maintaining high fluency (87.3/90.5/87.7).
%U https://aclanthology.org/2026.findings-emnlp.1024/
%P 21627-21658
Markdown (Informal)
[Interference-Aware Composition of Steering Vectors](https://aclanthology.org/2026.findings-emnlp.1024/) (Hong et al., Findings 2026)
ACL
- Giwon Hong, Gorjan Radevski, Goran Glavaš, Kiril Gashteovski, Edoardo Ponti, Carolin Lawrence, and Pasquale Minervini. 2026. Interference-Aware Composition of Steering Vectors. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 21627–21658, Budapest, Hungary. Association for Computational Linguistics.