<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kim-lee-2026-linearly">
    <titleInfo>
        <title>Linearly Readable, Not Robust: How Linear Jailbreak Probes Conflate Benign Register with Adversarial Intent</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">JaeSeong</namePart>
        <namePart type="family">Kim</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Suan</namePart>
        <namePart type="family">Lee</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>A cheap, now-standard safety filter attaches a linear probe to a frozen LLM’s hidden states to flag jailbreak prompts before generation, reaching near-perfect in-distribution accuracy at negligible cost. We show this single-layer probe is not a usable real-time filter, explain why, and give a fix. Across two model families, five attack types, ten languages, and three sets of hard-but-benign prompts, we find: (i) published probes are badly miscalibrated, so the headline 98–100% detection is an artifact of the chosen threshold (expected calibration error 0.20–0.33, and 0.07–0.20 even after a leakage-controlled retraining); (ii) most of the signal is benign writing style—its “register”—since a model-free style classifier reproduces ~0.92 of the probe’s accuracy; (iii) a 5% false-positive guarantee set on in-distribution benign prompts rises to 68–98% on diverse traffic; and (iv) a single 20-token suffix drops recall to near zero. Going beyond correlational “shortcut” diagnoses, we use concept erasure (LEACE) to causally remove a 3-dimensional register subspace: this cuts ~60 points of over-refusal (false flags on benign prompts) while leaving detection essentially intact (AUC drop \ensuremathłeq0.003), and the same subspace recurs across three architectures (verified against a shuffled-source-label null). Our recipe—register erasure plus deployment-time distribution-free calibration—achieves a guaranteed \ensuremathłeq5% false-positive rate on held-out diverse benign prompts at 0.945–0.961 recall, versus 0.18–0.29 for the un-erased probe. Cheap probing is a measurement instrument, deployable as a filter only via this recipe.</abstract>
    <identifier type="citekey">kim-lee-2026-linearly</identifier>
    <identifier type="doi">10.18653/2026.findings-emnlp.1648</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.1648/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>35071</start>
            <end>35088</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Linearly Readable, Not Robust: How Linear Jailbreak Probes Conflate Benign Register with Adversarial Intent
%A Kim, JaeSeong
%A Lee, Suan
%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 kim-lee-2026-linearly
%X A cheap, now-standard safety filter attaches a linear probe to a frozen LLM’s hidden states to flag jailbreak prompts before generation, reaching near-perfect in-distribution accuracy at negligible cost. We show this single-layer probe is not a usable real-time filter, explain why, and give a fix. Across two model families, five attack types, ten languages, and three sets of hard-but-benign prompts, we find: (i) published probes are badly miscalibrated, so the headline 98–100% detection is an artifact of the chosen threshold (expected calibration error 0.20–0.33, and 0.07–0.20 even after a leakage-controlled retraining); (ii) most of the signal is benign writing style—its “register”—since a model-free style classifier reproduces ~0.92 of the probe’s accuracy; (iii) a 5% false-positive guarantee set on in-distribution benign prompts rises to 68–98% on diverse traffic; and (iv) a single 20-token suffix drops recall to near zero. Going beyond correlational “shortcut” diagnoses, we use concept erasure (LEACE) to causally remove a 3-dimensional register subspace: this cuts ~60 points of over-refusal (false flags on benign prompts) while leaving detection essentially intact (AUC drop \ensuremathłeq0.003), and the same subspace recurs across three architectures (verified against a shuffled-source-label null). Our recipe—register erasure plus deployment-time distribution-free calibration—achieves a guaranteed \ensuremathłeq5% false-positive rate on held-out diverse benign prompts at 0.945–0.961 recall, versus 0.18–0.29 for the un-erased probe. Cheap probing is a measurement instrument, deployable as a filter only via this recipe.
%R 10.18653/2026.findings-emnlp.1648
%U https://aclanthology.org/2026.findings-emnlp.1648/
%U https://doi.org/10.18653/2026.findings-emnlp.1648
%P 35071-35088

Markdown (Informal)

[Linearly Readable, Not Robust: How Linear Jailbreak Probes Conflate Benign Register with Adversarial Intent](https://aclanthology.org/2026.findings-emnlp.1648/) (Kim & Lee, Findings 2026)

• Linearly Readable, Not Robust: How Linear Jailbreak Probes Conflate Benign Register with Adversarial Intent (Kim & Lee, Findings 2026)

ACL

• JaeSeong Kim and Suan Lee. 2026. Linearly Readable, Not Robust: How Linear Jailbreak Probes Conflate Benign Register with Adversarial Intent. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 35071–35088, Budapest, Hungary. Association for Computational Linguistics.