Metraipolis › archive › dc_9385cfda8ddd
Beyond Vector Hiding: Breaking and Mitigating Shared-Direction Weight Obfuscation in TEE-Offloaded Large Language Models — arXiv:2608.26651
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.26651
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-17T15:54:44.314Z
- SHA-256 of the archived text
2aeeac75d993a23018cd5b5182c1d84de6ad4d9e72a4d19c5203693b29313afc- Stable URL of this page
https://metraipolis.com/documento/dc_9385cfda8ddd
How to cite
Beyond Vector Hiding: Breaking and Mitigating Shared-Direction Weight Obfuscation in TEE-Offloaded Large Language Models — arXiv:2608.26651 — arXiv (en), https://arxiv.org/abs/2608.26651 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_9385cfda8ddd (Metraipolis archive dc_9385cfda8ddd, sha256 2aeeac75d993a23018cd5b5182c1d84de6ad4d9e72a4d19c5203693b29313afc)
BibTeX
@misc{dc9385cfda8ddd,
title = {Beyond Vector Hiding: Breaking and Mitigating Shared-Direction Weight Obfuscation in TEE-Offloaded Large Language Models — arXiv:2608.26651},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.26651},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_9385cfda8ddd; sha256 2aeeac75d993a23018cd5b5182c1d84de6ad4d9e72a4d19c5203693b29313afc},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.26651. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Trusted Execution Environment (TEE)-shielded partitioning of Large Language Models (LLMs) accelerates on-device inference by offloading obfuscated linear layers to an untrusted accelerator while retaining only a small correction inside the TEE. However, earlier lightweight obfuscation schemes preserved weight-vector directions and were broken by ArrowMatch. To defend against this attack, ArrowCloak injects scalar multiples of the same hidden direction into all weight vectors, enabling lightweight trusted correction. We show that this reuse leaves a rank-one relation across the complete accelerator-visible matrix. For the released real-valued scheme, we propose SpectralLeak, which estimates and removes the shared component. Across 12 task settings, its surrogates achieve $87.98\%$ mean accuracy versus $89.85\%$ for the victims. In our defense-favorable mod-$Q$ realization of ArrowCloak's published modular security formulation, mod-$Q$ arithmetic suppresses this spectral signal but retains the algebraic rank-one relation modulo $Q$. We therefore propose LatticeLeak, which exploits the resulting hidden lattice. In our BERT-Base and GPT2-Base experiments, it reconstructs every protected fixed-point parameter exactly; across all evaluated architectures, the reconstructed models retain victim-level task accuracy without victim queries, labels, or fine-tuning. These findings identify shared rank-one reuse as the root cause of the leakage exploited by our attacks. Guided by this insight, we design ButterflyCloak, a keyed maximal-rank butterfly mask that replaces the reused direction with distinct mask rows while retaining fast trusted correction... Tema degli esploratori: fiducia digitale e crittografia (retro-catalogo arXiv: annunci del 2026-08-28 in cs.CR). Categorie arXiv: Cryptography and Security (cs.CR). Pagina del lavoro: https://arxiv.org/abs/2608.26651 — 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 2026-10-17T15:31:38.258Z dalla pagina https://arxiv.org/catchup/cs.CR/2026-08-28?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).