A methodological analysis of prompt perturbations and their effect on attack success rates

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Machado, Tiago, de Macedo, Maysa Malfiza Garcia, de Paula, Rogerio Abreu, Grave, Marcelo Carpinette, Adebiyi, Aminat, de Souza, Luan Soares, Santarelli, Enrico, Pinhanez, Claudio
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912707517612032
author Machado, Tiago
de Macedo, Maysa Malfiza Garcia
de Paula, Rogerio Abreu
Grave, Marcelo Carpinette
Adebiyi, Aminat
de Souza, Luan Soares
Santarelli, Enrico
Pinhanez, Claudio
author_facet Machado, Tiago
de Macedo, Maysa Malfiza Garcia
de Paula, Rogerio Abreu
Grave, Marcelo Carpinette
Adebiyi, Aminat
de Souza, Luan Soares
Santarelli, Enrico
Pinhanez, Claudio
contents This work aims to investigate how different Large Language Models (LLMs) alignment methods affect the models' responses to prompt attacks. We selected open source models based on the most common alignment methods, namely, Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Reinforcement Learning with Human Feedback (RLHF). We conducted a systematic analysis using statistical methods to verify how sensitive the Attack Success Rate (ASR) is when we apply variations to prompts designed to elicit inappropriate content from LLMs. Our results show that even small prompt modifications can significantly change the Attack Success Rate (ASR) according to the statistical tests we run, making the models more or less susceptible to types of attack. Critically, our results demonstrate that running existing 'attack benchmarks' alone may not be sufficient to elicit all possible vulnerabilities of both models and alignment methods. This paper thus contributes to ongoing efforts on model attack evaluation by means of systematic and statistically-based analyses of the different alignment methods and how sensitive their ASR is to prompt variation.
format Preprint
id arxiv_https___arxiv_org_abs_2511_10686
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A methodological analysis of prompt perturbations and their effect on attack success rates
Machado, Tiago
de Macedo, Maysa Malfiza Garcia
de Paula, Rogerio Abreu
Grave, Marcelo Carpinette
Adebiyi, Aminat
de Souza, Luan Soares
Santarelli, Enrico
Pinhanez, Claudio
Computation and Language
62, 68,
I.2.7
This work aims to investigate how different Large Language Models (LLMs) alignment methods affect the models' responses to prompt attacks. We selected open source models based on the most common alignment methods, namely, Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Reinforcement Learning with Human Feedback (RLHF). We conducted a systematic analysis using statistical methods to verify how sensitive the Attack Success Rate (ASR) is when we apply variations to prompts designed to elicit inappropriate content from LLMs. Our results show that even small prompt modifications can significantly change the Attack Success Rate (ASR) according to the statistical tests we run, making the models more or less susceptible to types of attack. Critically, our results demonstrate that running existing 'attack benchmarks' alone may not be sufficient to elicit all possible vulnerabilities of both models and alignment methods. This paper thus contributes to ongoing efforts on model attack evaluation by means of systematic and statistically-based analyses of the different alignment methods and how sensitive their ASR is to prompt variation.
title A methodological analysis of prompt perturbations and their effect on attack success rates
topic Computation and Language
62, 68,
I.2.7
url https://arxiv.org/abs/2511.10686