Metraipolis › archive › dc_bde7e8ce799d
Fooling LLM Safety Guards into Judging the Wrong Text — arXiv:2610.31695
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.31695
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-26T19:53:44.905Z
- SHA-256 of the archived text
a556f21b1fd5c5a3ffed78cd983e688c942696a8e9d03244c8087a5b472f2a97- Stable URL of this page
https://metraipolis.com/documento/dc_bde7e8ce799d
How to cite
Fooling LLM Safety Guards into Judging the Wrong Text — arXiv:2610.31695 — arXiv (en), https://arxiv.org/abs/2610.31695 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bde7e8ce799d (Metraipolis archive dc_bde7e8ce799d, sha256 a556f21b1fd5c5a3ffed78cd983e688c942696a8e9d03244c8087a5b472f2a97)
BibTeX
@misc{dcbde7e8ce799d,
title = {Fooling LLM Safety Guards into Judging the Wrong Text — arXiv:2610.31695},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.31695},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bde7e8ce799d; sha256 a556f21b1fd5c5a3ffed78cd983e688c942696a8e9d03244c8087a5b472f2a97},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.31695. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Safety guards check content before an LLM application acts on it, and most paste that content into a plain-text prompt. Content that imitates the prompt can append a second task, and the guard answers that one instead. We call this target rebinding. Our explanation is that the guard grades the last complete task in its input, here a harmless decoy, so the harmful request goes through unchanged. In a pre-registered test, an appended forged query and decoy clear all 75 held-out AdvBench requests that Shieldstral-1.0-3B flags. The guard is not persuaded: a directive to ignore the text above clears none of them, and in a follow-up run a harmful decoy stops the attack. On our GuardRebind benchmark, each of five guards has an attack family, chosen per guard, that clears at least 93% of its flagged items. Rebinding explains this best on Shieldstral and gpt-oss-safeguard; on ShieldGemma and Llama Guard 3, generic evasion and trailing text respectively do part of the work, and Qwen3Guard falls mainly to a forgery of its whole task. Training the guard to score content no lower than its most harmful part cuts attack success from 75% to 0.8-1.6% on attack families, decoys and forged queries held out of training. All of these attacks append the forged structure after the request, and we added the held-out-decoy evaluation after a first round of test runs. Clean AUROC falls by at most 0.01, but our test of non-inferiority is inconclusive. A score-guided search still beats this trained guard on 47 of 50 items, usually by placing forged structure before the request. On about half of the items where the search wins, the same search also finds benign text, with no forged structure, that clears the guard. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Cryptography and Security (cs.CR) ; Computation and Language (cs.CL); Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2610.31695 — 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 2026-10-26T19:45:48.378Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).