https://txtmix.com/tags/cacheattack/Recent content in CacheAttack on Text MatrixHugozh-cnTue, 21 Jul 2026 20:06:14 +0800https://txtmix.com/posts/tech/semantic-cache-collision-attack-llm-supply-chain-2026-06-14/Sun, 14 Jun 2026 20:54:00 +0800https://txtmix.com/posts/tech/semantic-cache-collision-attack-llm-supply-chain-2026-06-14/<h2 id="llm-服务的默认优化正在打开一个隐形后门">LLM 服务的默认优化正在打开一个隐形后门</h2> <p><a href="https://aws.amazon.com/blogs/machine-learning/build-a-read-through-semantic-cache-with-amazon-opensearch-serverless-and-amazon-bedrock/" target="_blank" rel="noopener noreffer ">AWS Bedrock</a>、<a href="https://learn.microsoft.com/en-us/azure/cosmos-db/gen-ai/semantic-cache" target="_blank" rel="noopener noreffer ">Microsoft Azure Cosmos DB</a>、<a href="https://learn.microsoft.com/en-us/azure/api-management/azure-openai-enable-semantic-caching" target="_blank" rel="noopener noreffer ">Azure API Management</a> 这几家大厂,都在 LLM 服务里默认开了 <strong>Semantic Caching</strong>——用 query 的语义 embedding 向量做 cache key,对相似查询直接复用响应,降低延迟和成本。这套机制被开源项目 <a href="https://github.com/zilliztech/GPTCache" target="_blank" rel="noopener noreffer ">GPTCache</a> 和各类 <a href="https://arxiv.org/abs/2509.24832" target="_blank" rel="noopener noreffer ">SemShareKV</a> / <a href="https://arxiv.org/abs/2504.00970" target="_blank" rel="noopener noreffer ">SentenceKV</a> / <a href="https://arxiv.org/abs/2502.03771" target="_blank" rel="noopener noreffer ">vCache</a> 论文推广到整个行业。</p>