@inproceedings{chen-etal-2026-armor,
title = "{ARMOR}-{PRM}: Adversarially Robust Process Reward Models via Counterfactual Margins and Invariance Regularization",
author = "Chen, Hongyu and
Chai, Guangjia and
Nan, Haoxian and
Wang, Ruixue and
Jiang, Peilin",
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.1287/",
pages = "27250--27264",
ISBN = "979-8-89176-462-0",
abstract = "Process Reward Models (PRMs) are vulnerable to reward hacking: policies exploit surface cues such as confident phrasing or structured formatting in place of logical correctness, which yields an inverted-U collapse under optimization pressure. We propose ARMOR-PRM, which augments standard PRM training with three regularizers: a counterfactual margin loss, an adversarial invariance loss with low-score anchoring, and a gradient penalty on the reward head. On ProcessBench (three splits; ``math'' excluded due to training data overlap), ARMOR-PRM improves Detection Rate by 14.9 percentage points and AUROC by 4.1 percentage points over Vanilla PRM, and is the only variant whose Best-of-N reranking accuracy improves monotonically through N=128. A domain-shift diagnostic further shows that objective-level robustness and data coverage are distinct but complementary axes: ARMOR-PRM responds best to iterative hard-negative retraining, suggesting a practical two-stage deployment path."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2026-armor">
<titleInfo>
<title>ARMOR-PRM: Adversarially Robust Process Reward Models via Counterfactual Margins and Invariance Regularization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hongyu</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guangjia</namePart>
<namePart type="family">Chai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haoxian</namePart>
<namePart type="family">Nan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruixue</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peilin</namePart>
<namePart type="family">Jiang</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>Process Reward Models (PRMs) are vulnerable to reward hacking: policies exploit surface cues such as confident phrasing or structured formatting in place of logical correctness, which yields an inverted-U collapse under optimization pressure. We propose ARMOR-PRM, which augments standard PRM training with three regularizers: a counterfactual margin loss, an adversarial invariance loss with low-score anchoring, and a gradient penalty on the reward head. On ProcessBench (three splits; “math” excluded due to training data overlap), ARMOR-PRM improves Detection Rate by 14.9 percentage points and AUROC by 4.1 percentage points over Vanilla PRM, and is the only variant whose Best-of-N reranking accuracy improves monotonically through N=128. A domain-shift diagnostic further shows that objective-level robustness and data coverage are distinct but complementary axes: ARMOR-PRM responds best to iterative hard-negative retraining, suggesting a practical two-stage deployment path.</abstract>
<identifier type="citekey">chen-etal-2026-armor</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1287/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>27250</start>
<end>27264</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ARMOR-PRM: Adversarially Robust Process Reward Models via Counterfactual Margins and Invariance Regularization
%A Chen, Hongyu
%A Chai, Guangjia
%A Nan, Haoxian
%A Wang, Ruixue
%A Jiang, Peilin
%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 chen-etal-2026-armor
%X Process Reward Models (PRMs) are vulnerable to reward hacking: policies exploit surface cues such as confident phrasing or structured formatting in place of logical correctness, which yields an inverted-U collapse under optimization pressure. We propose ARMOR-PRM, which augments standard PRM training with three regularizers: a counterfactual margin loss, an adversarial invariance loss with low-score anchoring, and a gradient penalty on the reward head. On ProcessBench (three splits; “math” excluded due to training data overlap), ARMOR-PRM improves Detection Rate by 14.9 percentage points and AUROC by 4.1 percentage points over Vanilla PRM, and is the only variant whose Best-of-N reranking accuracy improves monotonically through N=128. A domain-shift diagnostic further shows that objective-level robustness and data coverage are distinct but complementary axes: ARMOR-PRM responds best to iterative hard-negative retraining, suggesting a practical two-stage deployment path.
%U https://aclanthology.org/2026.findings-emnlp.1287/
%P 27250-27264
Markdown (Informal)
[ARMOR-PRM: Adversarially Robust Process Reward Models via Counterfactual Margins and Invariance Regularization](https://aclanthology.org/2026.findings-emnlp.1287/) (Chen et al., Findings 2026)
ACL