Metraipolis › archive › dc_36dcb30867f6

CASCADE Against Jailbreaks: Combination Across Stages with Controlled Attack-Defense Evaluation — arXiv:2609.21793

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.21793
License (as archived)
CC0-1.0
Archived
2026-10-10T01:27:54.520Z
SHA-256 of the archived text
df4adbc89a7777470df37adf24db1cb8b74d65d5a41417266e678e72f616bba7
Stable URL of this page
https://metraipolis.com/documento/dc_36dcb30867f6

How to cite

CASCADE Against Jailbreaks: Combination Across Stages with Controlled Attack-Defense Evaluation — arXiv:2609.21793 — arXiv (en), https://arxiv.org/abs/2609.21793 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_36dcb30867f6 (Metraipolis archive dc_36dcb30867f6, sha256 df4adbc89a7777470df37adf24db1cb8b74d65d5a41417266e678e72f616bba7)
BibTeX
@misc{dc36dcb30867f6,
  title = {CASCADE Against Jailbreaks: Combination Across Stages with Controlled Attack-Defense Evaluation — arXiv:2609.21793},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.21793},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_36dcb30867f6; sha256 df4adbc89a7777470df37adf24db1cb8b74d65d5a41417266e678e72f616bba7},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.21793. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Defenses against jailbreak attacks on Large Language Models (LLMs) operate at different pipeline stages, such as input modification or output guard, but it remains unclear which defenses to deploy at each stage and how to combine them. Prior empirical studies, fragmented by inconsistent attack-success-rate definitions and experimental settings, have evaluated defenses largely in isolation. Here we present the first systematic study, to our knowledge, of defense combinations both within and across pipeline stages, under a consistent threat model of direct, black-box, single-turn attacks. Our decision framework standardizes evaluation through a principled attack-success-rate formulation with controlled query budgets, together with explicit fairness rules. Across 19 attacks and 15 defenses, we find that no single defense is universally best, but well-chosen combinations achieve substantial safety with minimal utility degradation, yielding practical recommendations for layered defense pipelines.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-21 in cs.CL).
Categorie arXiv: Cryptography and Security (cs.CR) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.21793 — 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 2026-10-10T01:00:30.329Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-21?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).