mlsanityData to 14 Oct 2026

To Cache or Not to Cache: Understanding and Mitigating Safety Incompatibility in Cross-Model KV Cache Reuse

arXiv 2610.15721 ↗First posted 11 Oct 2026cs.LG cs.CR cs.SEcited by 0

On the map: Efficient LLM inference and serving → KV cache optimization

Abstract (arXiv)

Cross-model KV cache sharing reduces prefill computation by allowing a receiver model to reuse prefix KV states generated by a related sender model. Existing methods optimize such reuse for latency and generation quality, but its safety implications remain unclear. In this work, we systematically evaluate this issue across diverse sender-receiver pairs and reuse configurations. We refer to deviations from the receiver's native safety behavior as safety incompatibility. Our study shows that both the direction and magnitude of safety incompatibility depend on the model pair and reuse configuration, with value reuse causing substantially larger deviations than key reuse across all evaluated pairs. Furthermore, we propose VReparo, a technique that learns per-layer and per-KV-head affine maps offline to align reused sender value tensors with their receiver-native counterparts before decoding. Across two model pairs under full KV reuse and the state-of-the-art selective reuse setting, VReparo reduces the average HarmBench attack success rate (ASR) from 57.28% to 39.21% and the average StrongREJECT score from 0.4215 to 0.2163 upon our benchmarks, while retaining an average Time to First Token (TTFT) reduction of 85.62%.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving. Builds on 24 of its 28 references that are in the map. First posted on arXiv 11 Oct 2026.

Builds on

24 of 28 references.

Closest ancestors · 10

Same direction, newest first

  1. 2026CacheBridge: Efficient Cross-Model KV Cache Transfer1 Sep
  2. 2026Cross-Model KV Cache Transfer in LLM Families: A Closed-Form Linear Mapping for Prefill Reuse4 Aug
  3. 2026HijackKV: New Threat in Position-Independent KV Cache Reuse22 Jul
  4. 2026AnchorKV: Safety-Aware KV Cache Compression via Soft Penalty with a Refusal Anchor16 Jun
  5. 2025Cache-to-Cache: Direct Semantic Communication Between Large Language Models3 Oct
  6. 2025The Pitfalls of KV Cache Compression30 Sep
  7. 2024KVSharer: Efficient Inference via Layer-Wise Dissimilar KV Cache Sharing24 Oct
  8. 2024SnapKV: LLM Knows What You are Looking for Before Generation22 Apr
  9. 2023Efficient Memory Management for Large Language Model Serving with PagedAttention12 Sep
  10. 2023H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models24 Jun

Other references · 14

  1. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,754
  2. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,429
  3. 2023LongBench: A Bilingual, Multitask Benchmark for Long Context Understandingcited by 1,417
  4. 2020Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Stepscited by 1,282
  5. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,154
  6. 2023XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Modelscited by 505
  7. 2024A StrongREJECT for Empty Jailbreakscited by 385
  8. 2024OR-Bench: An Over-Refusal Benchmark for Large Language Modelscited by 244
  9. 2025A Mousetrap: Fooling Large Reasoning Models for Jailbreak with Chain of Iterative Chaoscited by 43
  10. 2025Activated LoRA: Fine-tuned LLMs for Intrinsicscited by 12
  11. 2025Efficient Multi-Adapter LLM Serving via Cross-Model KV-Cache Reuse with Activated LoRAcited by 8
  12. 2026Semantic Cache Distillation: Efficient State Transfer via Reuse and Selective Patchingcited by 5
2 more
  1. 2025MetaCipher: A Time-Persistent and Universal Multi-Agent Framework for Cipher-Based Jailbreak Attacks for LLMscited by 4
  2. 2026Alignment Collapse Under KV Cache Quantization: Diagnosis and Mitigationcited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026CacheLink: Functional KV Transport for Zero-Prefill Cross-Model Context Handoff
  2. 2026A Universal Context-Reuse Layer for Cross-Model KV Sharing
  3. 2026Characterizing Contention-Induced Reliability Collapse in KV-Cache Timing Side Channels for Multi-Tenant LLM Serving
  4. 2026CacheReforge: Bounded Recovery for Stale KV Caches under Evolving Adapters
  5. 2026Governing the KV Cache: Preventing Timing Side-Channel Leakage in Multi-Tenant LLM Inference
  6. 2026Shared-Prefix KV Reuse Across Standard LoRA Adapters: Quality and Serving Tradeoffs
  7. 2026Evaluating the accuracy of KV cache reuse techniques
  8. 2025Selective KV-Cache Sharing to Mitigate Timing Side-Channels in LLM Inference