Metraipolis › archive › dc_136298cec362
SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency Measurement — arXiv:2609.05850
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.05850
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T13:42:41.154Z
- SHA-256 of the archived text
e0f843c592cda36fdd90993e21885f50b5404436d4c019ff3d5cb70c8cb67191- Stable URL of this page
https://metraipolis.com/documento/dc_136298cec362
How to cite
SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency Measurement — arXiv:2609.05850 — arXiv (en), https://arxiv.org/abs/2609.05850 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_136298cec362 (Metraipolis archive dc_136298cec362, sha256 e0f843c592cda36fdd90993e21885f50b5404436d4c019ff3d5cb70c8cb67191)
BibTeX
@misc{dc136298cec362,
title = {SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency Measurement — arXiv:2609.05850},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.05850},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_136298cec362; sha256 e0f843c592cda36fdd90993e21885f50b5404436d4c019ff3d5cb70c8cb67191},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.05850. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Despite the significant efforts devoted to aligning large language models (LLMs) with human values and ensuring safe deployment, recent work has revealed that LLMs remain vulnerable to adversarial jailbreak attacks that can bypass safety guardrails and elicit harmful responses. Many defense methods are proposed to detect jailbreaks but they are limited in their effectiveness to counter wide-range optimization-based jailbreak mechanisms that can yield highly fluency-optimized or harmful semantic obfuscated prompts. To tackle this challenge, we propose a unified detection framework SAFEGuard which incorporates a hybrid fluency measurement based on cross-layer distribution distance and perplexity, and the analysis of harmful semantics through gradient matching. Our method is grounded in a paramount observation: high fluency prompts maintain their malicious intention close to harmful prompts while harmful semantic obfuscated prompts often inject gibberish token sequences. Our evaluation demonstrates that SAFEGuard consistently outperforms state-of-the-art baselines and achieves significant improvement in accuracy across different optimization-based jailbreaks. This underscores the effectiveness of SAFEGuard against evolving jailbreak attacks. Tema degli esploratori: fiducia digitale e crittografia (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CR). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR). Pagina del lavoro: https://arxiv.org/abs/2609.05850 — 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 2026-10-16T13:09:48.519Z dalla pagina https://arxiv.org/catchup/cs.CR/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).