Super Suffixes: Bypassing Text Generation Alignment and Guard Models Simultaneously

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Adiletta, Andrew, Adiletta, Kathryn, Derya, Kemal, Sunar, Berk
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915671569334272
author Adiletta, Andrew
Adiletta, Kathryn
Derya, Kemal
Sunar, Berk
author_facet Adiletta, Andrew
Adiletta, Kathryn
Derya, Kemal
Sunar, Berk
contents The rapid deployment of Large Language Models (LLMs) has created an urgent need for enhanced security and privacy measures in Machine Learning (ML). LLMs are increasingly being used to process untrusted text inputs and even generate executable code, often while having access to sensitive system controls. To address these security concerns, several companies have introduced guard models, which are smaller, specialized models designed to protect text generation models from adversarial or malicious inputs. In this work, we advance the study of adversarial inputs by introducing Super Suffixes, suffixes capable of overriding multiple alignment objectives across various models with different tokenization schemes. We demonstrate their effectiveness, along with our joint optimization technique, by successfully bypassing the protection mechanisms of Llama Prompt Guard 2 on five different text generation models for malicious text and code generation. To the best of our knowledge, this is the first work to reveal that Llama Prompt Guard 2 can be compromised through joint optimization. Additionally, by analyzing the changing similarity of a model's internal state to specific concept directions during token sequence processing, we propose an effective and lightweight method to detect Super Suffix attacks. We show that the cosine similarity between the residual stream and certain concept directions serves as a distinctive fingerprint of model intent. Our proposed countermeasure, DeltaGuard, significantly improves the detection of malicious prompts generated through Super Suffixes. It increases the non-benign classification rate to nearly 100%, making DeltaGuard a valuable addition to the guard model stack and enhancing robustness against adversarial prompt attacks.
format Preprint
id arxiv_https___arxiv_org_abs_2512_11783
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Super Suffixes: Bypassing Text Generation Alignment and Guard Models Simultaneously
Adiletta, Andrew
Adiletta, Kathryn
Derya, Kemal
Sunar, Berk
Cryptography and Security
Artificial Intelligence
I.2.7
The rapid deployment of Large Language Models (LLMs) has created an urgent need for enhanced security and privacy measures in Machine Learning (ML). LLMs are increasingly being used to process untrusted text inputs and even generate executable code, often while having access to sensitive system controls. To address these security concerns, several companies have introduced guard models, which are smaller, specialized models designed to protect text generation models from adversarial or malicious inputs. In this work, we advance the study of adversarial inputs by introducing Super Suffixes, suffixes capable of overriding multiple alignment objectives across various models with different tokenization schemes. We demonstrate their effectiveness, along with our joint optimization technique, by successfully bypassing the protection mechanisms of Llama Prompt Guard 2 on five different text generation models for malicious text and code generation. To the best of our knowledge, this is the first work to reveal that Llama Prompt Guard 2 can be compromised through joint optimization. Additionally, by analyzing the changing similarity of a model's internal state to specific concept directions during token sequence processing, we propose an effective and lightweight method to detect Super Suffix attacks. We show that the cosine similarity between the residual stream and certain concept directions serves as a distinctive fingerprint of model intent. Our proposed countermeasure, DeltaGuard, significantly improves the detection of malicious prompts generated through Super Suffixes. It increases the non-benign classification rate to nearly 100%, making DeltaGuard a valuable addition to the guard model stack and enhancing robustness against adversarial prompt attacks.
title Super Suffixes: Bypassing Text Generation Alignment and Guard Models Simultaneously
topic Cryptography and Security
Artificial Intelligence
I.2.7
url https://arxiv.org/abs/2512.11783