@inproceedings{sai-etal-2026-robust,
title = "Robust and Efficient Guardrails with Latent Reasoning",
author = "Sai, Siddharth and
Wen, Xiaofei and
Chen, Muhao",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2518/",
pages = "57256--57268",
ISBN = "979-8-89176-459-0",
abstract = "Maintaining the safety of large language models (LLMs) is crucial as they are increasingly deployed in real-world applications. Existing safety guardrails typically rely on single-pass classification or, more recently, distilled reasoning. Reasoning-based guardrails significantly outperform classification-only baselines, but they incur substantial query latency and token overhead that make them impractical for high-throughput deployment. To address this challenge, we propose CoLaGuard, a guardrail model that transfers multi-step safety reasoning into a continuous latent space through a stage-wise training curriculum, enabling direct hidden-state propagation at inference. Evaluated across ten prompt and response moderation settings from eight unique safety benchmarks on ten prompt- and response-moderation settings spanning eight safety benchmarks, CoLaGuard improves macro-F1 by 8.24 points over Llama Guard 3 and matches our explicit reasoning baseline, GuardReasoner, in macro-F1 while delivering a 12.9$\times$ speedup and 22.4$\times$ reduction in token usage. Our results suggest that latent reasoning offers a practical alternative to explicit rationale generation for deployable guardrails, jointly improving safety robustness and inference efficiency rather than treating them as competing objectives."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sai-etal-2026-robust">
<titleInfo>
<title>Robust and Efficient Guardrails with Latent Reasoning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Siddharth</namePart>
<namePart type="family">Sai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaofei</namePart>
<namePart type="family">Wen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Muhao</namePart>
<namePart type="family">Chen</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 2026 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Maintaining the safety of large language models (LLMs) is crucial as they are increasingly deployed in real-world applications. Existing safety guardrails typically rely on single-pass classification or, more recently, distilled reasoning. Reasoning-based guardrails significantly outperform classification-only baselines, but they incur substantial query latency and token overhead that make them impractical for high-throughput deployment. To address this challenge, we propose CoLaGuard, a guardrail model that transfers multi-step safety reasoning into a continuous latent space through a stage-wise training curriculum, enabling direct hidden-state propagation at inference. Evaluated across ten prompt and response moderation settings from eight unique safety benchmarks on ten prompt- and response-moderation settings spanning eight safety benchmarks, CoLaGuard improves macro-F1 by 8.24 points over Llama Guard 3 and matches our explicit reasoning baseline, GuardReasoner, in macro-F1 while delivering a 12.9\times speedup and 22.4\times reduction in token usage. Our results suggest that latent reasoning offers a practical alternative to explicit rationale generation for deployable guardrails, jointly improving safety robustness and inference efficiency rather than treating them as competing objectives.</abstract>
<identifier type="citekey">sai-etal-2026-robust</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2518/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>57256</start>
<end>57268</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Robust and Efficient Guardrails with Latent Reasoning
%A Sai, Siddharth
%A Wen, Xiaofei
%A Chen, Muhao
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F sai-etal-2026-robust
%X Maintaining the safety of large language models (LLMs) is crucial as they are increasingly deployed in real-world applications. Existing safety guardrails typically rely on single-pass classification or, more recently, distilled reasoning. Reasoning-based guardrails significantly outperform classification-only baselines, but they incur substantial query latency and token overhead that make them impractical for high-throughput deployment. To address this challenge, we propose CoLaGuard, a guardrail model that transfers multi-step safety reasoning into a continuous latent space through a stage-wise training curriculum, enabling direct hidden-state propagation at inference. Evaluated across ten prompt and response moderation settings from eight unique safety benchmarks on ten prompt- and response-moderation settings spanning eight safety benchmarks, CoLaGuard improves macro-F1 by 8.24 points over Llama Guard 3 and matches our explicit reasoning baseline, GuardReasoner, in macro-F1 while delivering a 12.9\times speedup and 22.4\times reduction in token usage. Our results suggest that latent reasoning offers a practical alternative to explicit rationale generation for deployable guardrails, jointly improving safety robustness and inference efficiency rather than treating them as competing objectives.
%U https://aclanthology.org/2026.emnlp-main.2518/
%P 57256-57268
Markdown (Informal)
[Robust and Efficient Guardrails with Latent Reasoning](https://aclanthology.org/2026.emnlp-main.2518/) (Sai et al., EMNLP 2026)
ACL
- Siddharth Sai, Xiaofei Wen, and Muhao Chen. 2026. Robust and Efficient Guardrails with Latent Reasoning. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 57256–57268, Budapest, Hungary. Association for Computational Linguistics.