I Know What You Said: Unveiling Hardware Cache Side-Channels in Local Large Language Model Inference

Zibo Gao, Junjie Hu, Feng Guo, Yixin Zhang, Yinglong Han, Siyuan Liu, Haiyang Li, Zhiqiang Lv, Institute of Information Engineering, Chinese Academy of Sciences, School of Cyber Security

34th USENIX Security Symposium · Day 1

This paper, presented at USENIX Security, unveils a critical and previously unexplored security vulnerability in locally deployed Large Language Models (LLMs): **hardware cache side-channel leakage**. Authored by researchers from the University of Chinese Academy of Sciences, the work demonstrates how an unprivileged adversary can eavesdrop on a victim's local LLM inference process to reconstruct both the sensitive input prompts and the generated output responses. This discovery directly challenges the prevailing assumption that local LLMs inherently offer superior privacy compared to their cloud-based counterparts.

AI review

This is real research. Novel attack surface, serious engineering to make it work, and results that should make every 'local LLM for privacy' pitch sweat. The combination of cache side-channels with LLM-based denoising to reconstruct both input and output text is genuinely clever and the 98%+ semantic similarity numbers are damning.