Similarity Is Not Validity: Defending LLM Semantic Caches Against Poisoning

Reported by arXiv cs.CR (Zihan Zhang, Shuangjie Yao, Zesen Liu, Zhixiang Zhang, Wai Ip Lai, Dung Hiu Hilton Yeung, Chun Kit Zhang, Fuchen Ma, Yuanyuan Yuan, Yu Jiang, Dongdong She)

Summary by Cyber Security News30 Sep 2026, 06:00 CEST

About this summary. This is a short, independently written summary of an article first published by arXiv cs.CR. Cyber Security News did not report or verify the underlying story. Read the original: https://arxiv.org/abs/2609.35908

arXiv:2609.35908v1 Announce Type: new Abstract: Semantic caches reduce LLM serving costs by reusing previously generated answers for semantically similar queries. However, retrieval is based solely on embedding similarity between the incoming query and cached queries. This design enables cache poisoning: an attacker can cache a malicious response under a query with high cosine similarity to benign requests.

Read the full article at arXiv cs.CR →

Source attribution: headline and facts are from arXiv cs.CR (arxiv.org). Summary method: excerpt of the source description. See our source attribution policy.

Related coverage

We tested our own WAF with frontier AI models. Here’s what we found

Hannibal Puts Autonomous Penetration Testing Under Your Control

SelfOp: An Optimization Algorithm for Self-Improving Security Agents

Spectre bug is back, this time to haunt JIT engines

Citrix NetScaler CVE-2026-88772 Exploit Details Show Pre-Auth Path to Shellcode Execution

Says Block, Still Acts: Why LLM Safety Judgments Fail to Govern Action in LLM Agents