Metraipolis › archive › dc_9acebf4835e6
The Blind Spots of Detection: Adversarial Perturbation Attacks against Watchdogs in Distributed LLM Inference — arXiv:2611.01097
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.01097
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T18:32:59.571Z
- SHA-256 of the archived text
de50c684289b261818a19cc788e89cceb804771e7ec1ec66b81495980560903d- Stable URL of this page
https://metraipolis.com/documento/dc_9acebf4835e6
How to cite
The Blind Spots of Detection: Adversarial Perturbation Attacks against Watchdogs in Distributed LLM Inference — arXiv:2611.01097 — arXiv (en), https://arxiv.org/abs/2611.01097 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_9acebf4835e6 (Metraipolis archive dc_9acebf4835e6, sha256 de50c684289b261818a19cc788e89cceb804771e7ec1ec66b81495980560903d)
BibTeX
@misc{dc9acebf4835e6,
title = {The Blind Spots of Detection: Adversarial Perturbation Attacks against Watchdogs in Distributed LLM Inference — arXiv:2611.01097},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.01097},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_9acebf4835e6; sha256 de50c684289b261818a19cc788e89cceb804771e7ec1ec66b81495980560903d},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.01097. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Decentralizing large language model (LLM) inference across untrusted Web nodes requires robust verification to ensure computational integrity. Existing activation-based verification and detection mechanisms validate intermediate outputs against reference values, permitting bounded numerical deviations to accommodate benign cross-hardware discrepancies. However, a critical security question remains: what can a malicious worker achieve while operating within these tolerance bounds? In this work, we introduce Hush-Up, a stealthy availability attack against distributed LLM inference, where a single malicious worker halts generation while evading attribution. The attacker first profiles its local execution to model an empirical bound of benign hardware errors. Using this error bound as a feasibility constraint, the attacker optimizes an adversarial perturbation on its output activations to maximize the final-layer probability of the end-of-sequence (EOS) token, i.e., to hush up the model. Across diverse models and benchmarks, Hush-Up effectively forces premature generation termination and collapses model capabilities. Furthermore, the attack hushes up the watchdogs themselves: existing defenses often fail to detect the sabotage entirely or falsely attribute the amplified downstream anomalies to honest workers. Our findings expose a fundamental gap between local numerical verification and end-to-end accountability: perturbations too small to indict a worker locally can nevertheless be powerful enough to break distributed inference globally. Tema degli esploratori: fiducia digitale e crittografia (elenco arXiv cs.CR, nuovi lavori del giorno). Categorie arXiv: Cryptography and Security (cs.CR). Pagina del lavoro: https://arxiv.org/abs/2611.01097 — 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 2026-11-03T18:12:23.780Z dalla pagina https://arxiv.org/list/cs.CR/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).