Semantic-Preserving Adversarial Attacks on LLMs: An Adaptive Greedy Binary Search Approach

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Zhang, Chong, Li, Xiang, Wang, Jia, Liang, Shan, Xue, Haochen, Jin, Xiaobo
Format:	Preprint
Published:	2025
Subjects:	Computation and Language Cryptography and Security
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

_version_	1866915355990949888
author	Zhang, Chong Li, Xiang Wang, Jia Liang, Shan Xue, Haochen Jin, Xiaobo
author_facet	Zhang, Chong Li, Xiang Wang, Jia Liang, Shan Xue, Haochen Jin, Xiaobo
contents	Large Language Models (LLMs) increasingly rely on automatic prompt engineering in graphical user interfaces (GUIs) to refine user inputs and enhance response accuracy. However, the diversity of user requirements often leads to unintended misinterpretations, where automated optimizations distort original intentions and produce erroneous outputs. To address this challenge, we propose the Adaptive Greedy Binary Search (AGBS) method, which simulates common prompt optimization mechanisms while preserving semantic stability. Our approach dynamically evaluates the impact of such strategies on LLM performance, enabling robust adversarial sample generation. Through extensive experiments on open and closed-source LLMs, we demonstrate AGBS's effectiveness in balancing semantic consistency and attack efficacy. Our findings offer actionable insights for designing more reliable prompt optimization systems. Code is available at: https://github.com/franz-chang/DOBS
format	Preprint
id	arxiv_https___arxiv_org_abs_2506_18756
institution	arXiv
publishDate	2025
record_format	arxiv
spellingShingle	Semantic-Preserving Adversarial Attacks on LLMs: An Adaptive Greedy Binary Search Approach Zhang, Chong Li, Xiang Wang, Jia Liang, Shan Xue, Haochen Jin, Xiaobo Computation and Language Cryptography and Security Large Language Models (LLMs) increasingly rely on automatic prompt engineering in graphical user interfaces (GUIs) to refine user inputs and enhance response accuracy. However, the diversity of user requirements often leads to unintended misinterpretations, where automated optimizations distort original intentions and produce erroneous outputs. To address this challenge, we propose the Adaptive Greedy Binary Search (AGBS) method, which simulates common prompt optimization mechanisms while preserving semantic stability. Our approach dynamically evaluates the impact of such strategies on LLM performance, enabling robust adversarial sample generation. Through extensive experiments on open and closed-source LLMs, we demonstrate AGBS's effectiveness in balancing semantic consistency and attack efficacy. Our findings offer actionable insights for designing more reliable prompt optimization systems. Code is available at: https://github.com/franz-chang/DOBS
title	Semantic-Preserving Adversarial Attacks on LLMs: An Adaptive Greedy Binary Search Approach
topic	Computation and Language Cryptography and Security
url	https://arxiv.org/abs/2506.18756