Adversarial Decoding: Generating Readable Documents for Adversarial Objectives

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Collin, Zhang, Tingwei, Shmatikov, Vitaly
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916644847091712
author Zhang, Collin
Zhang, Tingwei
Shmatikov, Vitaly
author_facet Zhang, Collin
Zhang, Tingwei
Shmatikov, Vitaly
contents We design, implement, and evaluate adversarial decoding, a new, generic text generation technique that produces readable documents for different adversarial objectives. Prior methods either produce easily detectable gibberish, or cannot handle objectives that include embedding similarity. In particular, they only work for direct attacks (such as jailbreaking) and cannot produce adversarial text for realistic indirect injection, e.g., documents that (1) are retrieved in RAG systems in response to broad classes of queries, and also (2) adversarially influence subsequent generation. We also show that fluency (low perplexity) is not sufficient to evade filtering. We measure the effectiveness of adversarial decoding for different objectives, including RAG poisoning, jailbreaking, and evasion of defensive filters, and demonstrate that it outperforms existing methods while producing readable adversarial documents.
format Preprint
id arxiv_https___arxiv_org_abs_2410_02163
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Adversarial Decoding: Generating Readable Documents for Adversarial Objectives
Zhang, Collin
Zhang, Tingwei
Shmatikov, Vitaly
Computation and Language
Cryptography and Security
Machine Learning
We design, implement, and evaluate adversarial decoding, a new, generic text generation technique that produces readable documents for different adversarial objectives. Prior methods either produce easily detectable gibberish, or cannot handle objectives that include embedding similarity. In particular, they only work for direct attacks (such as jailbreaking) and cannot produce adversarial text for realistic indirect injection, e.g., documents that (1) are retrieved in RAG systems in response to broad classes of queries, and also (2) adversarially influence subsequent generation. We also show that fluency (low perplexity) is not sufficient to evade filtering. We measure the effectiveness of adversarial decoding for different objectives, including RAG poisoning, jailbreaking, and evasion of defensive filters, and demonstrate that it outperforms existing methods while producing readable adversarial documents.
title Adversarial Decoding: Generating Readable Documents for Adversarial Objectives
topic Computation and Language
Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2410.02163