@inproceedings{zhang-etal-2026-activation,
title = "Activation-Space Anchored Access Control for Multi-Class Permission Reasoning in Large Language Models",
author = "Zhang, Zhaopeng and
Sun, Pengcheng and
Zhang, Lan and
Tang, Chen and
Lai, Jiewei and
Zhuang, Xirong and
Wang, Yunhao and
Jin, Hui and
Liu, Li",
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.908/",
pages = "19224--19240",
ISBN = "979-8-89176-462-0",
abstract = "Large language models (LLMs) are increasingly deployed over knowledge bases for efficient knowledge retrieval and question answering. However, LLMs can inadvertently answer beyond a user{'}s permission scope, leaking sensitive content and complicating deployment under fine-grained access-control requirements. We identify a geometric regularity in intermediate activations: representations associated with different permission scopes exhibit consistent and distinguishable structure in activation space. Building on this observation, we propose Activation-Space Anchored Access Control (AAAC), a training-free framework for multi-class permission control. AAAC constructs a compact bank of permission-specific activation anchors from a small offline calibration set and requires no model fine-tuning. At inference time, the trusted permission metadata is used by the controller to steer query activations toward authorized anchors and away from unauthorized ones, thereby mitigating the risk of over-privileged generation. Across three LLM families, AAAC achieves permission violation rates of 0.005{--}0.009 and prompt-based attack success rates of 0.084{--}0.101, while maintaining competitive utility with 10.3{\%}{--}25.2{\%} inference overhead."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-activation">
<titleInfo>
<title>Activation-Space Anchored Access Control for Multi-Class Permission Reasoning in Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zhaopeng</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pengcheng</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chen</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiewei</namePart>
<namePart type="family">Lai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xirong</namePart>
<namePart type="family">Zhuang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yunhao</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hui</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Li</namePart>
<namePart type="family">Liu</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 (LLMs) are increasingly deployed over knowledge bases for efficient knowledge retrieval and question answering. However, LLMs can inadvertently answer beyond a user’s permission scope, leaking sensitive content and complicating deployment under fine-grained access-control requirements. We identify a geometric regularity in intermediate activations: representations associated with different permission scopes exhibit consistent and distinguishable structure in activation space. Building on this observation, we propose Activation-Space Anchored Access Control (AAAC), a training-free framework for multi-class permission control. AAAC constructs a compact bank of permission-specific activation anchors from a small offline calibration set and requires no model fine-tuning. At inference time, the trusted permission metadata is used by the controller to steer query activations toward authorized anchors and away from unauthorized ones, thereby mitigating the risk of over-privileged generation. Across three LLM families, AAAC achieves permission violation rates of 0.005–0.009 and prompt-based attack success rates of 0.084–0.101, while maintaining competitive utility with 10.3%–25.2% inference overhead.</abstract>
<identifier type="citekey">zhang-etal-2026-activation</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.908/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>19224</start>
<end>19240</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Activation-Space Anchored Access Control for Multi-Class Permission Reasoning in Large Language Models
%A Zhang, Zhaopeng
%A Sun, Pengcheng
%A Zhang, Lan
%A Tang, Chen
%A Lai, Jiewei
%A Zhuang, Xirong
%A Wang, Yunhao
%A Jin, Hui
%A Liu, Li
%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 zhang-etal-2026-activation
%X Large language models (LLMs) are increasingly deployed over knowledge bases for efficient knowledge retrieval and question answering. However, LLMs can inadvertently answer beyond a user’s permission scope, leaking sensitive content and complicating deployment under fine-grained access-control requirements. We identify a geometric regularity in intermediate activations: representations associated with different permission scopes exhibit consistent and distinguishable structure in activation space. Building on this observation, we propose Activation-Space Anchored Access Control (AAAC), a training-free framework for multi-class permission control. AAAC constructs a compact bank of permission-specific activation anchors from a small offline calibration set and requires no model fine-tuning. At inference time, the trusted permission metadata is used by the controller to steer query activations toward authorized anchors and away from unauthorized ones, thereby mitigating the risk of over-privileged generation. Across three LLM families, AAAC achieves permission violation rates of 0.005–0.009 and prompt-based attack success rates of 0.084–0.101, while maintaining competitive utility with 10.3%–25.2% inference overhead.
%U https://aclanthology.org/2026.findings-emnlp.908/
%P 19224-19240
Markdown (Informal)
[Activation-Space Anchored Access Control for Multi-Class Permission Reasoning in Large Language Models](https://aclanthology.org/2026.findings-emnlp.908/) (Zhang et al., Findings 2026)
ACL
- Zhaopeng Zhang, Pengcheng Sun, Lan Zhang, Chen Tang, Jiewei Lai, Xirong Zhuang, Yunhao Wang, Hui Jin, and Li Liu. 2026. Activation-Space Anchored Access Control for Multi-Class Permission Reasoning in Large Language Models. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 19224–19240, Budapest, Hungary. Association for Computational Linguistics.