{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,30]],"date-time":"2026-07-30T17:08:24Z","timestamp":1785431304902,"version":"3.56.0"},"reference-count":33,"publisher":"MDPI AG","issue":"6","license":[{"start":{"date-parts":[[2026,5,24]],"date-time":"2026-05-24T00:00:00Z","timestamp":1779580800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["62176045"],"award-info":[{"award-number":["62176045"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"award":["62176045"],"award-info":[{"award-number":["62176045"]}],"id":[{"id":"https:\/\/ror.org\/01h0zpd94","id-type":"ROR","asserted-by":"publisher"}]},{"name":"Sichuan Science and Technology Program","award":["2025ZNSFSC0453"],"award-info":[{"award-number":["2025ZNSFSC0453"]}]},{"name":"Sichuan Science and Technology Program","award":["2022014"],"award-info":[{"award-number":["2022014"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Entropy"],"abstract":"<jats:p>Large language models (LLMs) remain vulnerable to jailbreak attacks, especially in black-box settings where target-model gradients and internal tokenization are inaccessible. Recent information bottleneck-based defenses cast prompt protection as a compression problem, but existing methods still rely heavily on white-box optimization and the intrinsic alignment strength of the protected model. To address these limitations, we propose RIB-Guard, a safety-aware information bottleneck defense for black-box LLMs. RIB-Guard learns a token-level masking policy that extracts a minimally safety-sufficient prompt via reinforcement learning using only black-box feedback. In addition, it introduces an independent lightweight safety head to estimate residual jailbreak risk and provide model-agnostic safety guidance during training. The proposed framework jointly balances prompt compactness, benign utility preservation, and residual risk suppression within a unified objective. Experimental results on direct single-turn harmful and benign prompt settings show that RIB-Guard improves jailbreak robustness while maintaining competitive benign utility. By extending information bottleneck-based prompt protection from white-box to black-box settings, RIB-Guard provides a step toward safety-aware information-theoretic front-end defense for black-box LLMs.<\/jats:p>","DOI":"10.3390\/e28060585","type":"journal-article","created":{"date-parts":[[2026,5,25]],"date-time":"2026-05-25T02:32:28Z","timestamp":1779676348000},"page":"585","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["RIB-Guard: A Risk-Aware Information Bottleneck Defense for Black-Box Large Language Models"],"prefix":"10.3390","volume":"28","author":[{"given":"Muen","family":"Cai","sequence":"first","affiliation":[{"name":"School of Computer Science and Engineering, University of Electronic Science and Technology of China, No. 2006, Xiyuan Ave, Chengdu 611731, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0001-0795-0848","authenticated-orcid":false,"given":"Yuan","family":"Shen","sequence":"additional","affiliation":[{"name":"Meta Platforms Inc., Menlo Park, CA 94025, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0004-9559-3538","authenticated-orcid":false,"given":"Xiong","family":"Luo","sequence":"additional","affiliation":[{"name":"Department of Information Technology, Uppsala University, 752 37 Uppsala, Sweden"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jian","family":"Hu","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, University of Electronic Science and Technology of China, No. 2006, Xiyuan Ave, Chengdu 611731, China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1968","published-online":{"date-parts":[[2026,5,24]]},"reference":[{"key":"ref_1","doi-asserted-by":"crossref","first-page":"61478","DOI":"10.52202\/075280-2687","article-title":"Are aligned neural networks adversarially aligned?","volume":"36","author":"Carlini","year":"2023","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_2","unstructured":"Zou, A., Wang, Z., Carlini, N., Nasr, M., Kolter, J.Z., and Fredrikson, M. (2023). Universal and transferable adversarial attacks on aligned language models. arXiv."},{"key":"ref_3","doi-asserted-by":"crossref","unstructured":"Chao, P., Robey, A., Dobriban, E., Hassani, H., Pappas, G.J., and Wong, E. (2025). Jailbreaking black box large language models in twenty queries. Proceedings of the 2025 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML), IEEE.","DOI":"10.1109\/SaTML64287.2025.00010"},{"key":"ref_4","unstructured":"Liu, X., Xu, N., Chen, M., and Xiao, C. (2024, January 7\u201311). AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models. Proceedings of the The Twelfth International Conference on Learning Representations, Vienna, Austria."},{"key":"ref_5","doi-asserted-by":"crossref","unstructured":"Ding, P., Kuang, J., Ma, D., Cao, X., Xian, Y., Chen, J., and Huang, S. (2024, January 16\u201321). A wolf in sheep\u2019s clothing: Generalized nested jailbreak prompts can fool large language models easily. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), Mexico City, Mexico.","DOI":"10.18653\/v1\/2024.naacl-long.118"},{"key":"ref_6","unstructured":"Yu, J., Lin, X., Yu, Z., and Xing, X. (2023). Gptfuzzer: Red teaming large language models with auto-generated jailbreak prompts. arXiv."},{"key":"ref_7","first-page":"80079","article-title":"Jailbroken: How does llm safety training fail?","volume":"36","author":"Wei","year":"2023","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_8","unstructured":"Robey, A., Wong, E., Hassani, H., and Pappas, G.J. (2023). SmoothLLM: Defending Large Language Models Against Jailbreaking Attacks. arXiv."},{"key":"ref_9","doi-asserted-by":"crossref","unstructured":"Ji, J., Hou, B., Robey, A., Pappas, G.J., Hassani, H., Zhang, Y., Wong, E., and Chang, S. (2025, January 20\u201324). Defending large language models against jailbreak attacks via semantic smoothing. Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics, Mumbai, India.","DOI":"10.18653\/v1\/2025.ijcnlp-short.2"},{"key":"ref_10","doi-asserted-by":"crossref","unstructured":"Cao, B., Cao, Y., Lin, L., and Chen, J. (2024, January 11\u201316). Defending against alignment-breaking attacks via robustly aligned llm. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Bangkok, Thailand.","DOI":"10.18653\/v1\/2024.acl-long.568"},{"key":"ref_11","unstructured":"Wang, X., Wu, D., Ji, Z., Li, Z., Ma, P., Wang, S., Li, Y., Liu, Y., Liu, N., and Rahmel, J. (2025, January 13\u201315). {SelfDefend}:{LLMs} can defend themselves against jailbreaking in a practical manner. Proceedings of the 34th USENIX Security Symposium (USENIX Security 25), Seattle, WA, USA."},{"key":"ref_12","first-page":"55005","article-title":"Jailbreakbench: An open robustness benchmark for jailbreaking large language models","volume":"37","author":"Chao","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_13","unstructured":"Zhou, W., Wang, X., Xiong, L., Xia, H., Gu, Y., Chai, M., Zhu, F., Huang, C., Dou, S., and Xi, Z. (2024). Easyjailbreak: A unified framework for jailbreaking large language models. arXiv."},{"key":"ref_14","doi-asserted-by":"crossref","first-page":"1","DOI":"10.1145\/3605943","article-title":"Recent advances in natural language processing via large pre-trained language models: A survey","volume":"56","author":"Min","year":"2023","journal-title":"Acm Comput. Surv."},{"key":"ref_15","first-page":"53728","article-title":"Direct preference optimization: Your language model is secretly a reward model","volume":"36","author":"Rafailov","year":"2023","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_16","unstructured":"Tishby, N., Pereira, F.C., and Bialek, W. (2000). The information bottleneck method. arXiv."},{"key":"ref_17","unstructured":"Alemi, A.A., Fischer, I., Dillon, J.V., and Murphy, K. (2017, January 24\u201326). Deep Variational Information Bottleneck. Proceedings of the International Conference on Learning Representations, Toulon, France."},{"key":"ref_18","first-page":"29723","article-title":"Protecting your llms with information bottleneck","volume":"37","author":"Liu","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_19","first-page":"61065","article-title":"Tree of attacks: Jailbreaking black-box llms automatically","volume":"37","author":"Mehrotra","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_20","unstructured":"Greshake, K., Abdelnabi, S., Mishra, S., Endres, C., Holz, T., and Fritz, M. (2023). More than you\u2019ve asked for: A comprehensive analysis of novel prompt injection threats to application-integrated large language models. arXiv."},{"key":"ref_21","unstructured":"Liu, Y., Deng, G., Li, Y., Wang, K., Wang, Z., Wang, X., Zhang, T., Liu, Y., Wang, H., and Zheng, Y. (2023). Prompt injection attack against llm-integrated applications. arXiv."},{"key":"ref_22","unstructured":"Russinovich, M., Salem, A., and Eldan, R. (2025, January 13\u201315). Great, now write an article about that: The crescendo {Multi-Turn}{LLM} jailbreak attack. Proceedings of the 34th USENIX Security Symposium (USENIX Security 25), Seattle, WA, USA."},{"key":"ref_23","unstructured":"Sun, X., Zhang, D., Yang, D., Zou, Q., and Li, H. (2024). Multi-turn context jailbreak attack on large language models from first principles. arXiv."},{"key":"ref_24","unstructured":"Mazeika, M., Phan, L., Yin, X., Zou, A., Wang, Z., Mu, N., Sakhaee, E., Li, N., Basart, S., and Li, B. (2024, January 21\u201327). HarmBench: A standardized evaluation framework for automated red teaming and robust refusal. Proceedings of the 41st International Conference on Machine Learning, Vienna, Austria."},{"key":"ref_25","first-page":"125416","article-title":"A strongreject for empty jailbreaks","volume":"37","author":"Souly","year":"2024","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"ref_26","doi-asserted-by":"crossref","unstructured":"Paranjape, B., Joshi, M., Thickstun, J., Hajishirzi, H., and Zettlemoyer, L. (2020, January 16\u201320). An information bottleneck approach for controlling conciseness in rationale extraction. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), Online.","DOI":"10.18653\/v1\/2020.emnlp-main.153"},{"key":"ref_27","doi-asserted-by":"crossref","unstructured":"Li, Q., Wu, Z., Kong, L., and Bi, W. (2023, January 9\u201314). Explanation regeneration via information bottleneck. Proceedings of the Findings of the Association for Computational Linguistics: ACL 2023, Toronto, ON, Canada.","DOI":"10.18653\/v1\/2023.findings-acl.765"},{"key":"ref_28","doi-asserted-by":"crossref","unstructured":"Hu, X., Hong, Z., Zhang, C., King, I., and Yu, P. (2023, January 23\u201327). Think rationally about what you see: Continuous rationale extraction for relation extraction. Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval, Taipei, Taiwan.","DOI":"10.1145\/3539618.3592072"},{"key":"ref_29","unstructured":"Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bashlykov, N., Batra, S., Bhargava, P., and Bhosale, S. (2023). Llama 2: Open foundation and fine-tuned chat models. arXiv."},{"key":"ref_30","unstructured":"Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F.L., Almeida, D., Altenschmidt, J., Altman, S., and Anadkat, S. (2023). Gpt-4 technical report. arXiv."},{"key":"ref_31","unstructured":"Zhao, X., Yang, X., Pang, T., Du, C., Li, L., Wang, Y.X., and Wang, W.Y. (2025, January 13\u201319). Weak-to-Strong Jailbreaking on Large Language Models. Proceedings of the International Conference on Machine Learning, Vancouver, BC, Canada."},{"key":"ref_32","unstructured":"Joshi, M., Choi, E., Weld, D.S., and Zettlemoyer, L. (August, January 30). Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Vancouver, BC, Canada."},{"key":"ref_33","unstructured":"Miao, S., Liu, M., and Li, P. (2022, January 17\u201323). Interpretable and generalizable graph learning via stochastic attention mechanism. Proceedings of the International conference on machine learning, Baltimore, MD, USA."}],"container-title":["Entropy"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/1099-4300\/28\/6\/585\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,3]],"date-time":"2026-06-03T04:21:13Z","timestamp":1780460473000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/1099-4300\/28\/6\/585"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,5,24]]},"references-count":33,"journal-issue":{"issue":"6","published-online":{"date-parts":[[2026,6]]}},"alternative-id":["e28060585"],"URL":"https:\/\/doi.org\/10.3390\/e28060585","relation":{},"ISSN":["1099-4300"],"issn-type":[{"value":"1099-4300","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,5,24]]}}}