Metraipolis › archive › dc_eead5ec0d545

How Fragile Is On-Device Language Model Safety? Localizing Safety-Critical Parameters for Sparse Fault Analysis — arXiv:2610.09000

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.09000
License (as archived): CC0-1.0
Archived: 2026-10-08T11:06:46.207Z
SHA-256 of the archived text: 895c09de4da623f09fa8b7b3db3cdc460e3cf5bd51c2f04c5ad3da339c12c0e7
Stable URL of this page: https://metraipolis.com/documento/dc_eead5ec0d545

How to cite

How Fragile Is On-Device Language Model Safety? Localizing Safety-Critical Parameters for Sparse Fault Analysis — arXiv:2610.09000 — arXiv (en), https://arxiv.org/abs/2610.09000 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_eead5ec0d545 (Metraipolis archive dc_eead5ec0d545, sha256 895c09de4da623f09fa8b7b3db3cdc460e3cf5bd51c2f04c5ad3da339c12c0e7)

BibTeX

@misc{dceead5ec0d545,
  title = {How Fragile Is On-Device Language Model Safety? Localizing Safety-Critical Parameters for Sparse Fault Analysis — arXiv:2610.09000},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.09000},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_eead5ec0d545; sha256 895c09de4da623f09fa8b7b3db3cdc460e3cf5bd51c2f04c5ad3da339c12c0e7},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.09000. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

As small language models (SLMs) are increasingly deployed on resource-constrained and on-device platforms, including as components of agentic systems, the integrity of locally stored model parameters becomes an important safety concern. We investigate whether safety-sensitive behavior in LLaMA-2-7B-Chat is concentrated within a sparse subset of parameters, creating a reduced fault surface for targeted analysis. We study two complementary localization methods: low-rank safety-associated subspace analysis and parameter-level safety--utility importance filtering. Both approaches reveal highly non-uniform safety sensitivity across the network, with the MLP down_proj consistently emerging as a prominent safety-sensitive component and o_proj providing a smaller contribution. Using parameter-level localization, modifying only 0.19% of model weights in down_proj yields 53% Basic ASR and 56% GCG ASR, while tinyBenchmarks accuracy remains at 51.6% compared with a 52.2% unmodified baseline. These results motivate targeted fault analysis and selective integrity protection for language models deployed in resource-constrained, on-device, and agentic settings.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL); Cryptography and Security (cs.CR); Machine Learning (cs.LG); Software Engineering (cs.SE).
Pagina del lavoro: https://arxiv.org/abs/2610.09000 — 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 2026-10-08T10:14:05.886Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).