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 2502.11533cs.CL
收藏

合并不熟悉的 LLM 时需谨慎:一种能够窃取隐私的网络钓鱼模型

Be Cautious When Merging Unfamiliar LLMs: A Phishing Model Capable of Stealing Privacy

Zhejiang University(浙江大学);University of Virginia(弗吉尼亚大学)

Zhenyuan Guo, Yi Shi, Wenlong Meng, Chen Gong, Chengkun Wei, Wenzhi Chen

AI总结:

本文提出 PhiMM 隐私攻击,利用伪装成具备专门能力的网络钓鱼模型,在模型合并后通过指令窃取 PII 与 MI,实验显示隐私泄露显著上升。

摘要:

模型合并是大型语言模型(LLM)中一种广泛使用的技术,它将多个面向特定任务的 LLM 集成到一个统一模型中,使合并后的模型能够继承这些 LLM 的专门能力。大多数面向特定任务的 LLM 来自开源社区,且未经过严格审计,这可能给模型合并带来风险。本文强调了一种被忽视的隐私风险:一个不安全的模型可能会损害参与模型合并的其他 LLM 的隐私。具体而言,我们提出了 PhiMM,这是一种隐私攻击方法,利用精心构造的隐私网络钓鱼指令数据集训练一个能够窃取隐私的网络钓鱼模型。此外,我们引入了一种新颖的模型伪装方法,通过模仿一种专门能力来隐藏攻击意图,诱使用户合并该网络钓鱼模型。一旦受害者合并了网络钓鱼模型,攻击者就可以通过使用网络钓鱼指令查询合并后的模型,提取个人身份信息(PII)或推断成员信息(MI)。实验结果表明,合并网络钓鱼模型会增加隐私泄露风险。与合并前的结果相比,PII 泄露平均增加了 3.9%,MI 泄露平均增加了 17.4%。我们通过一个链接发布了 PhiMM 的代码。

英文摘要:

Model merging is a widespread technology in large language models (LLMs) that integrates multiple task-specific LLMs into a unified one, enabling the merged model to inherit the specialized capabilities of these LLMs. Most task-specific LLMs are sourced from open-source communities and have not undergone rigorous auditing, potentially imposing risks in model merging. This paper highlights an overlooked privacy risk: \textit{an unsafe model could compromise the privacy of other LLMs involved in the model merging.} Specifically, we propose PhiMM, a privacy attack approach that trains a phishing model capable of stealing privacy using a crafted privacy phishing instruction dataset. Furthermore, we introduce a novel model cloaking method that mimics a specialized capability to conceal attack intent, luring users into merging the phishing model. Once victims merge the phishing model, the attacker can extract personally identifiable information (PII) or infer membership information (MI) by querying the merged model with the phishing instruction. Experimental results show that merging a phishing model increases the risk of privacy breaches. Compared to the results before merging, PII leakage increased by 3.9\% and MI leakage increased by 17.4\% on average. We release the code of PhiMM through a link.

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。

↑