# information-leakage
標記為「information-leakage」的 5 篇文章
激活分析與隱藏狀態利用
透過隱藏狀態擷取、logprob 探測、拒絕方向分析與激活導向技術讀取模型內部。
activationshidden-statesprobinginformation-leakagemechanistic-interpretability
實作：GPU 側通道攻擊
Demonstrate information leakage through GPU memory residuals and timing side channels, showing how shared GPU infrastructure can expose sensitive model data.
labgpuside-channelhardwareinformation-leakage
模型快取安全
AI 推論中快取的安全風險——涵蓋 KV 快取攻擊、提示詞快取利用、共享快取資訊洩漏、快取投毒與安全快取設定。
cachingkv-cacheprompt-cachingshared-cacheinformation-leakageperformance
Cross-Modal Information Leakage
Techniques for extracting sensitive information through modality boundary violations, including system prompt leakage via image descriptions, context window dumping through multi-modal queries, and side-channel attacks across modalities.
information-leakagecross-modal
Timing Side-Channel 攻擊 詳解
Extract information from LLM applications through timing differences in response generation and safety filter processing.
walkthroughstimingside-channelinformation-leakage