Evolving Jailbreaks: Automated Multi-Objective Long-Tail Attacks on Large Language Models
Identification
- Clé publique
- radar:r-daa1fd02ba5f3aadd17373b25508aaa2
- Auteurs
- Wenjing Hong · Zhonghua Rong · Li Wang · Feng Chang · Jian Zhu · Ke Tang · Zexuan Zhu · Yew-Soon Ong
- Année déclarée
- 2026
- Langue déclarée
- en
- Type source
- article
- Première détection
- 2026-11-14T02:45:52.452452+00:00
- Dernière observation des métadonnées
- 2026-11-14T02:45:52.452452+00:00
- Dernière vérification HTTP
- Non effectuée
Institutions déclarées par la source
Shenzhen University · University of Science and Technology of China · Southern University of Science and Technology · Agency for Science, Technology and Research
Pays d’affiliation déclarés
CN · SG
Résumé documentaire — langue originale
Large Language Models (LLMs) have been widely deployed, especially through free Web-based applications that expose them to diverse user-generated inputs, including those from long-tail distributions such as low-resource languages and encrypted private data. This open-ended exposure increases the risk of jailbreak attacks that undermine model safety alignment. While recent studies have shown that leveraging long-tail distributions can facilitate such jailbreaks, existing approaches largely rely on handcrafted rules, limiting the systematic evaluation of these security and privacy vulnerabilities. In this work, we present EvoJail, an automated framework for discovering long-tail distribution attacks via multi-objective evolutionary search. EvoJail formulates long-tail attack prompt generation as a multi-objective optimization problem that jointly maximizes attack effectiveness and minimizes output perplexity, and introduces a semantic-algorithmic solution representation to capture both high-level semantic intent and low-level structural transformations of encryption-decryption logic. Building upon this representation, EvoJail integrates LLM-assisted operators into a multi-objective evolutionary framework, enabling adaptive and semantically informed mutation and crossover for efficiently exploring a highly structured and open-ended search space. Extensive experiments demonstrate that EvoJail consistently discovers diverse and effective long-tail jailbreak strategies, achieving competitive performance with existing methods in both individual and ensemble level.
Identifiants
- OPENALEX
- https://openalex.org/W7140140067
- DOI
- 10.1109/tevc.2026.3726531
Repères documentaires
- Famille documentaire
- article
- Types déclarés par les sources
- article
- Éditeurs
- Institute of Electrical and Electronics Engineers
Les mentions d’entités ne constituent pas une preuve de reconnaissance juridique. La juridiction étudiée est distincte du pays d’affiliation des auteurs.
Sources
OpenAlex · Document 1
Citer cette référence
Wenjing Hong; Zhonghua Rong; Li Wang; Feng Chang; Jian Zhu; Ke Tang; Zexuan Zhu; Yew-Soon Ong. Evolving Jailbreaks: Automated Multi-Objective Long-Tail Attacks on Large Language Models. 2026. https://doi.org/10.1109/tevc.2026.3726531
JSON · CSL-JSON · BibTeX · RIS · CSV · JSON-LD ·
APA, Chicago, ISO 690 : formats bibliographiques fondés sur les champs disponibles. OSCOLA / Bluebook : données structurées uniquement ; aucune conformité juridique complète n’est revendiquée sans vérification des éléments nécessaires.
- APA
- Wenjing Hong; Zhonghua Rong; Li Wang; Feng Chang; Jian Zhu; Ke Tang; Zexuan Zhu; Yew-Soon Ong. (2026). Evolving Jailbreaks: Automated Multi-Objective Long-Tail Attacks on Large Language Models. https://doi.org/10.1109/tevc.2026.3726531
- Chicago
- Wenjing Hong; Zhonghua Rong; Li Wang; Feng Chang; Jian Zhu; Ke Tang; Zexuan Zhu; Yew-Soon Ong. Evolving Jailbreaks: Automated Multi-Objective Long-Tail Attacks on Large Language Models. 2026. https://doi.org/10.1109/tevc.2026.3726531
- ISO 690
- Wenjing Hong; Zhonghua Rong; Li Wang; Feng Chang; Jian Zhu; Ke Tang; Zexuan Zhu; Yew-Soon Ong. Evolving Jailbreaks: Automated Multi-Objective Long-Tail Attacks on Large Language Models. 2026. https://doi.org/10.1109/tevc.2026.3726531
Warning: Undefined array key "last_link_verified" in /home/clients/e27d81e12f04d2bfd747aa264540ac56/sites/_UDTR_APP/legal_library_reference.php on line 31
UDTR — Bibliothèque juridique mondiale des droits de la Nature. Evolving Jailbreaks: Automated Multi-Objective Long-Tail Attacks on Large Language Models. Notice documentaire: https://www.declarationuniverselledesdroitsdelarbre.org/global-rights-of-nature-radar/reference/r-daa1fd02ba5f3aadd17373b25508aaa2/ · Dernière vérification HTTP : non effectuée
Données de cette version figée
Dossiers et relations documentaires
Rôles des sources
Source bibliographique / publication d’origine · https://doi.org/10.1109/tevc.2026.3726531
Un lien vers une autorité ne prouve pas à lui seul une affirmation juridique. Les résumés et les textes intégraux restent soumis aux conditions de leur source.
Statut documentaire
Pertinence à examiner
Bibliothèque
Cette identité n’est pas encore intégrée à la Bibliothèque juridique.