Metraipolis › archive › dc_0c76bf645b20

SparLeak: Privacy Leakage from Sparse Attention in LLM Inference on Shared GPUs — arXiv:2609.38830

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.38830
License (as archived)
CC0-1.0
Archived
2026-10-07T13:43:24.190Z
SHA-256 of the archived text
e5c6f67a3e24e0ae9d6ab685482c465ac505cc1d90e6be8f1ffa208bf5e9d8e6
Stable URL of this page
https://metraipolis.com/documento/dc_0c76bf645b20

How to cite

SparLeak: Privacy Leakage from Sparse Attention in LLM Inference on Shared GPUs — arXiv:2609.38830 — arXiv (en), https://arxiv.org/abs/2609.38830 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0c76bf645b20 (Metraipolis archive dc_0c76bf645b20, sha256 e5c6f67a3e24e0ae9d6ab685482c465ac505cc1d90e6be8f1ffa208bf5e9d8e6)
BibTeX
@misc{dc0c76bf645b20,
  title = {SparLeak: Privacy Leakage from Sparse Attention in LLM Inference on Shared GPUs — arXiv:2609.38830},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.38830},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0c76bf645b20; sha256 e5c6f67a3e24e0ae9d6ab685482c465ac505cc1d90e6be8f1ffa208bf5e9d8e6},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.38830. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Sparse attention is widely used to accelerate long-context inference in modern large language models (LLMs), but its input-dependent execution behavior introduces previously unexplored privacy risks. We identify a new GPU micro-architectural side channel, termed Sparsity-Induced Memory Access (SIMA), which arises from secret-dependent key-value cache access patterns induced by sparse attention. Based on this observation, we present SparLeak, a phase-aware side-channel attack that extracts SIMA traces during LLM inference and enables two practical privacy extractions: query attribute inference from prefill-phase traces and autoregressive response reconstruction from decoding-phase traces. By reconstructing approximate token-level sparsity profiles from page-level observations and applying profiling-based learning, SparLeak accurately recovers sensitive information, including user-query attributes and private LLM response content. Extensive evaluation across three LLM architectures, three sparse attention mechanisms, and three privacy-sensitive datasets shows that SparLeak achieves average attack success rates of 90.9% for attribute inference and 87.3% for response reconstruction under real-world LLM serving settings, highlighting the significance to account for SIMA leakage when deploying sparse-attention-based LLM systems. We provide anonymized SIMA traces, trained attack models, evaluation scripts, and documentation as artifacts at this https URL .

Tema degli esploratori: fiducia digitale e crittografia (retro-catalogo arXiv: annunci del 2026-10-01 in cs.CR).
Categorie arXiv: Machine Learning (cs.LG) ; Cryptography and Security (cs.CR).
Pagina del lavoro: https://arxiv.org/abs/2609.38830 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T13:16:25.252Z dalla pagina https://arxiv.org/catchup/cs.CR/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).