Similarity Is Not Validity: Defending LLM Semantic Caches Against Poisoning

arXiv:2609.35908 · cs.CR, cs.AI · Submitted 2026-09-28 · Read on arXiv

cs.CR, cs.AI

Submitted: 2026-09-28

Updated: 2026-09-28

Comments: 25 pages, 5 figures, 17 tables. Code: https://github.com/shentoumengxin/deletion-gain

Code: https://github.com/shentoumengxin/deletion-gain

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Deletion Gain — 39× in this paper
• AUC — 29× in this paper · explained in 2 episodes
• Embeddings — 26× in this paper · explained in 2 episodes
• LLM judge — 20× in this paper · explained in 2 episodes
• Perplexity — 20× in this paper · explained in 3 episodes
• Calibration — 15× in this paper · explained in 7 episodes
• Benchmark — 14× in this paper · explained in 2 episodes
• Ablation — 10× in this paper · explained in 1 episode
• Cosine Similarity — 9× in this paper · explained in 3 episodes
• Verification — 6× in this paper · explained in 2 episodes
• Adam — 5× in this paper · explained in 2 episodes
• Empirical Methods in Natural Language Processing — 4× in this paper
• Adversarial Attacks — 3× in this paper · explained in 1 episode
• Trace — 3× in this paper · explained in 2 episodes
• Transformer — 3× in this paper · explained in 2 episodes
• Core — 2× in this paper · explained in 2 episodes
• Distillation — 2× in this paper · explained in 2 episodes
• Robustness — 2× in this paper · explained in 5 episodes
• Semantic Similarity — 2× in this paper · explained in 2 episodes
• Adversarial Examples — 1× in this paper · explained in 2 episodes
• Adversarial Prompting — 1× in this paper · explained in 1 episode
• Agent Memory — 1× in this paper · explained in 1 episode
• Alignment — 1× in this paper · explained in 1 episode
• Attack Success Rate (ASR) — 1× in this paper · explained in 2 episodes
• Commentary — 1× in this paper · explained in 2 episodes
• Constrained Answer Poisoning — 1× in this paper
• Contrastive Learning — 1× in this paper · explained in 5 episodes
• Intrinsic Dimension — 1× in this paper · explained in 2 episodes
• Key Collision Attack — 1× in this paper
• LLM-as-a-Judge — 1× in this paper · explained in 3 episodes
• Large Language Model (LLM) — 1× in this paper · explained in 3 episodes
• Network and Distributed System Security Symposium — 1× in this paper
• RAG — 1× in this paper · explained in 3 episodes
• Regularization — 1× in this paper · explained in 3 episodes
• Steering — 1× in this paper · explained in 2 episodes
• Vector Representations — 1× in this paper · explained in 2 episodes

Sources

• Detecting Language Model Attacks with Perplexity
• Baseline Defenses for Adversarial Attacks Against Aligned Language Models
• Certifying LLM Safety against Adversarial Prompting
• Towards General Text Embeddings with Multi-stage Contrastive Learning
• LaCache: Robust Semantic Caching for LLM Serving
• GPT Semantic Cache: Reducing LLM Costs and Latency via Semantic Embedding Caching
• DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
• vCache: Verified Semantic Prompt Caching
• GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding
• Text Embeddings by Weakly-Supervised Contrastive Pre-training
• Qwen3 Technical Report
• Geometry-Guided Adversarial Prompt Detection via Curvature and Local Intrinsic Dimension
• From Similarity to Vulnerability: Key Collision Attack on LLM Semantic Caching
• Universal and Transferable Adversarial Attacks on Aligned Language Models

Related papers

• SoK: AI-Augmented Binary Reversing
• Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
• Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
• Efficient Fuzzy PSI under One-Sided Assumptions
• Sealing the Audit-Runtime Gap for LLM Skills
• Token Composition: A Graph Based on EVM Logs