arXiv cs.AI· Karima Makhlouf, Lamiaa Basyoni, Syed Khaderi, Gabriel Marquez, Peter Sotomango, Mahmoud Awawdah, Sami Zhioua·· 2 天前AI 评分31
LLM 隐私威胁的消融研究：四类攻击的敏感性差异
On the Privacy of LLMs: An Ablation Study
AI 导读
一项针对 LLM 隐私的消融研究统一评估了成员推理（MIA）、属性推理（AIA）、数据提取（DEA）和后门攻击（BA）四类代表性攻击，发现各自敏感因素不同：MIA 强依赖数据集特征与攻击策略，AIA 受模型与训练特征影响，DEA 随模型规模和训练数据重复度上升，BA 则始终保持高成功率。
正文 · 原文
Abstract:Large language models (LLMs) face diverse privacy threats, yet existing studies typically evaluate individual attacks in isolation, offering limited guidance on which system factors drive privacy vulnerabilities. This paper presents a unified threat model and a systematic ablation study of four representative attacks: Membership Inference (MIA), Attribute Inference (AIA), Data Extraction (DEA), and Backdoor Attacks (BA). Our findings reveal attack-specific sensitivities: membership inference depends strongly on dataset characteristics and attack strategy, attribute inference is influenced by model and training characteristics, data extraction increases with model scale and training-data repetition, and backdoor attacks achieve consistently high success rates. Based on these findings, we derive practical recommendations for privacy-aware LLM deployment, including prioritizing membership leakage assessments in retrieval-augmented systems, minimizing repeated personal information in training data, monitoring attribute inference risks, and safeguarding fine-tuning pipelines against backdoors.
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.02255 [cs.CR] |
| (or arXiv:2605.02255v2 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2605.02255 arXiv-issued DOI via DataCite |
Submission history
From: Karima Makhlouf [view email]
[v1]
Mon, 4 May 2026 06:06:41 UTC (2,203 KB)
[v2]
Thu, 15 Oct 2026 08:34:36 UTC (3,353 KB)
来源：arXiv cs.AI · arxiv.org