Large language models (LLMs) exhibit impressive language capabilities but remain vulnerable to malicious prompts and jailbreaking attacks. Existing knowledge editing methods for LLM detoxification face two major challenges. First, they often rely on entity-specific localization, making them ineffective against adversarial inputs without explicit entities. Second, these methods suffer from over-editing, where detoxified models reject legitimate queries, compromising overall performance. In this paper, we propose ToxEdit, a toxicity-aware knowledge editing approach that dynamically detects toxic activation patterns during forward propagation. It then routes computations through adaptive inter-layer pathways to mitigate toxicity effectively. This design ensures precise toxicity mitigation while preserving LLMs’ general capabilities. To more accurately assess over-editing, we also enhance the SafeEdit benchmark by incorporating instruction-following evaluation tasks. Experimental results on multiple LLMs demonstrate that our ToxEdit outperforms previous state-of-the-art methods in both detoxification performance and safeguarding general capabilities of LLMs.

Recent research shows that supplementing Large Language Models (LLMs) with knowledge graphs can enhance their performance. However, existing methods often introduce noise in the retrieval and reasoning pipeline, hindering LLMs’ ability to effectively integrate external knowledge for complex multi-hop question answering. To address this, we propose RefKG, a novel framework designed to enhance the reasoning capabilities of LLMs through reflective engagement with knowledge graphs. RefKG autonomously conduct retrieval and reflection on knowledge graphs. It consists of three modules: Query Decoupling, LLM-Driven Knowledge Graph Exploration, and Inference with Knowledge Reconstruction. We also introduce a multi-task tuning strategy that not only integrates external knowledge into LLMs but also trains them to leverage this knowledge for answering questions. This significantly improves their performance on knowledge-intensive tasks. Experiments on fact verification and knowledge graph question answering demonstrate RefKG’s effectiveness.