SafeRoute: Adaptive Model Selection for Efficient and Accurate Safety Guardrails in Large Language Models,

  • Deploying large language models (LLMs) in real-world applications requires robust safety guard models to detect and block harmful user prompts. While large safety guard models achieve strong performance, their computational cost is substantial. To mitigate this, smaller distilled models are used, but they often underperform on “hard” examples where the larger model provides accurate predictions. We observe that many inputs can be reliablyhandled by the smaller model, while only a small fraction require the larger model’s capacity. Motivated by this, we propose SafeRoute, a binary router that distinguishes hard examples from easy ones. Our method selectively applies the larger safety guard model to the data that the router considers hard, improving efficiency while maintaining accuracy compared to solely using the larger safety guard model. Experimental results on multiple benchmark datasets demonstrate that our adaptive model selection significantly enhances the trade-off between computational cost and safety performance,Deploying large language models (LLMs) in real-world applications requires robust safety guard models to detect and block harmful user prompts. While large safety guard models achieve strong performance, their computational cost is substantial. To mitigate this, smaller distilled models are used, but they often underperform on “hard” examples where the larger model provides accurate predictions. We observe that many inputs can be reliablyhandled by the smaller model, while only a small fraction require the larger model’s capacity. Motivated by this, we propose SafeRoute, a binary router that distinguishes hard examples from easy ones. Our method selectively applies the larger safety guard model to the data that the router considers hard, improving efficiency while maintaining accuracy compared to solely using the larger safety guard model. Experimental results on multiple benchmark datasets demonstrate that our adaptive model selection significantly enhances the trade-off between computational cost and safety performance, outperforming relevant baselines.show moreshow less

Download full text files

Export metadata

Additional Services

Search Google Scholar
Metadaten
Author:Seanie Lee, Dong Bok Lee, Dominik Wagner, Minki Kang, Haebin Seong, Tobias BockletORCiD, Juho Lee, Sung Ju Hwang
URL / DOI:https://aclanthology.org/2025.findings-acl.105.pdf
Parent Title (English):Findings of the Association for Computational Linguistics: ACL 2025
Publisher:Association for Computational Linguistics
Place of publication:Vienna, Austria
Document Type:Article
Language:English
Date of first Publication:2025/07/01
Reviewed:Begutachtet/Reviewed
Release Date:2025/11/26
Pagenumber:2053-2069
institutes:Fakultät Informatik
Zentrum für Künstliche Intelligenz (KIZ)
Research Themes:Digitalisierung & Künstliche Intelligenz
Licence (German):Creative Commons - CC BY - Namensnennung 4.0 International
Verstanden ✔
Diese Webseite verwendet technisch erforderliche Session-Cookies. Durch die weitere Nutzung der Webseite stimmen Sie diesem zu. Unsere Datenschutzerklärung finden Sie hier.