Distillability of LLM Security Logic: Predicting Attack Success Rate of Outline Filling Attack via Ranking Regression

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Tianyu, Xi, Zihang, Hua, Jingyu, Zhong, Sheng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911290049429504
author Zhang, Tianyu
Xi, Zihang
Hua, Jingyu
Zhong, Sheng
author_facet Zhang, Tianyu
Xi, Zihang
Hua, Jingyu
Zhong, Sheng
contents In the realm of black-box jailbreak attacks on large language models (LLMs), the feasibility of constructing a narrow safety proxy, a lightweight model designed to predict the attack success rate (ASR) of adversarial prompts, remains underexplored. This work investigates the distillability of an LLM's core security logic. We propose a novel framework that incorporates an improved outline filling attack to achieve dense sampling of the model's security boundaries. Furthermore, we introduce a ranking regression paradigm that replaces standard regression and trains the proxy model to predict which prompt yields a higher ASR. Experimental results show that our proxy model achieves an accuracy of 91.1 percent in predicting the relative ranking of average long response (ALR), and 69.2 percent in predicting ASR. These findings confirm the predictability and distillability of jailbreak behaviors, and demonstrate the potential of leveraging such distillability to optimize black-box attacks.
format Preprint
id arxiv_https___arxiv_org_abs_2511_22044
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Distillability of LLM Security Logic: Predicting Attack Success Rate of Outline Filling Attack via Ranking Regression
Zhang, Tianyu
Xi, Zihang
Hua, Jingyu
Zhong, Sheng
Cryptography and Security
Artificial Intelligence
In the realm of black-box jailbreak attacks on large language models (LLMs), the feasibility of constructing a narrow safety proxy, a lightweight model designed to predict the attack success rate (ASR) of adversarial prompts, remains underexplored. This work investigates the distillability of an LLM's core security logic. We propose a novel framework that incorporates an improved outline filling attack to achieve dense sampling of the model's security boundaries. Furthermore, we introduce a ranking regression paradigm that replaces standard regression and trains the proxy model to predict which prompt yields a higher ASR. Experimental results show that our proxy model achieves an accuracy of 91.1 percent in predicting the relative ranking of average long response (ALR), and 69.2 percent in predicting ASR. These findings confirm the predictability and distillability of jailbreak behaviors, and demonstrate the potential of leveraging such distillability to optimize black-box attacks.
title Distillability of LLM Security Logic: Predicting Attack Success Rate of Outline Filling Attack via Ranking Regression
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2511.22044