{"type":"ScholarlyArticle","title":"Towards Comprehensive Post Safety Alignment of Large Language Models via Safety Patching","authors":[{"name":"Weixiang Zhao","slug":"weixiang-zhao","given":"Weixiang","family":"Zhao","affiliation":""},{"name":"Yulin Hu","slug":"yulin-hu","given":"Yulin","family":"Hu","affiliation":""},{"name":"Yang Deng","slug":"yang-deng","given":"Yang","family":"Deng","affiliation":""},{"name":"Jiahe Guo","slug":"jiahe-guo","given":"Jiahe","family":"Guo","affiliation":""},{"name":"Xingyu Sui","slug":"xingyu-sui","given":"Xingyu","family":"Sui","affiliation":""},{"name":"Yanyan Zhao","slug":"yanyan-zhao","given":"Yanyan","family":"Zhao","affiliation":""},{"name":"Bing Qin","slug":"bing-qin","given":"Bing","family":"Qin","affiliation":""},{"name":"Ting Liu","slug":"ting-liu","given":"Ting","family":"Liu","affiliation":""}],"abstract":"Safety alignment of large language models (LLMs) has been gaining increasing attention. However, current safety-aligned LLMs suffer from the fragile and imbalanced safety mechanisms, which can still be induced to generate unsafe responses, exhibit over-safety by rejecting safe user inputs, and fail to preserve general utility after safety alignment. To this end, we propose a novel post safety alignment (PSA) method to address these inherent and emerging safety challenges, including safety enhancement, over-safety mitigation, and utility preservation. In specific, we introduce SafePatching, a novel framework for comprehensive PSA, where two distinct safety patches are developed on the harmful data to enhance safety and mitigate over-safety concerns, and then seamlessly integrated into the target LLM backbone without compromising its utility.  Extensive experiments on four representative aligned LLMs, including LLaMA-2/3, Gemma and Mistral, show that SafePatching achieves a more comprehensive PSA than baseline methods, further optimizing the balance between being helpful and harmless in current aligned LLMs. Also, SafePatching demonstrates its superiority in continual PSA scenarios.","doi":"10.24963/ijcai.2026/99","url":"https://ijcai.org/proceedings/2026/99","pdf":"https://ijcai.org/proceedings/2026/0099.pdf","markdown":"https://ijcai.org/proceedings/2026/99.md","bibtex":"https://ijcai.org/proceedings/2026/bibtex/99","volume":{"slug":"2026","title":"Proceedings of the Thirty-Fifth International Joint Conference on Artificial Intelligence","short_title":"IJCAI 2026","url":"https://ijcai.org/proceedings/2026/","year":2026,"location":"Bremen, Germany","dates":"15-21 August 2026","editors":"Diego Calvanese","isbn":"978-1-956792-09-6","publisher":"International Joint Conferences on Artificial Intelligence Organization","doi_prefix":"10.24963/ijcai.2026"},"section":"AI Ethics, Trust, Fairnes","track":"Main Track","number":99,"pages":{"first":880,"last":888},"keywords":["AI Ethics, Trust, Fairnes: Safety and robustness","Natural Language Processing: Language models"],"videos":[],"references":[],"license_note":"Copyright International Joint Conferences on Artificial Intelligence."}