arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2404.02637cs.CRcs.AIcs.DC
收藏

劫持大型语言模型应用的词表攻击

Vocabulary Attack to Hijack Large Language Model Applications

Ostbayerische Technische Hochschule Amberg-Weiden(东巴伐利亚应用科学大学安贝格-魏登分校)

Patrick Levi, Christoph P. Neumann

AI总结:

本文提出一种通过从模型词表中插入优化单词来劫持大型语言模型应用的方法,对Llama2和Flan-T5家族模型实验证明,该方法隐蔽难测,且攻击者模型可不同于目标模型。

摘要:

大型语言模型(LLMs)的快速发展正在推动越来越多的应用。随着用户数量的增加,我们也看到越来越多的攻击者试图智取这些系统。他们希望模型泄露机密信息、特定的虚假信息或表现出攻击性行为。为此,他们通过插入分隔符或系统地改写指令来操纵LLM的指令,直到达到目的。我们的方法不同。它从模型词表中插入单词。我们使用优化过程和来自另一个LLM(攻击者LLM)的嵌入来找到这些单词。我们分别对来自Llama2和Flan-T5家族的两个流行开源LLM进行目标劫持来证明我们的方法。我们提出两个主要发现。首先,我们的方法创建了不引人注目的指令,因此很难被检测到。对于许多攻击案例,我们发现即使插入单个单词也足够了。其次,我们证明可以使用与目标模型不同的模型来执行我们的攻击。

英文摘要:

The fast advancements in Large Language Models (LLMs) are driving an increasing number of applications. Together with the growing number of users, we also see an increasing number of attackers who try to outsmart these systems. They want the model to reveal confidential information, specific false information, or offensive behavior. To this end, they manipulate their instructions for the LLM by inserting separators or rephrasing them systematically until they reach their goal. Our approach is different. It inserts words from the model vocabulary. We find these words using an optimization procedure and embeddings from another LLM (attacker LLM). We prove our approach by goal hijacking two popular open-source LLMs from the Llama2 and the Flan-T5 families, respectively. We present two main findings. First, our approach creates inconspicuous instructions and therefore it is hard to detect. For many attack cases, we find that even a single word insertion is sufficient. Second, we demonstrate that we can conduct our attack using a different model than the target model to conduct our attack with.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑