Adaptive Detoxification: Safeguarding General Capabilities of LLMs through Toxicity-Aware Knowledge Editing

TL;DR

This paper introduces ToxEdit, a toxicity-aware knowledge editing method that dynamically detects and mitigates toxic responses in large language models while preserving their overall capabilities.

Contribution

ToxEdit is a novel approach that detects toxicity during forward pass and adaptively routes computations, overcoming entity-specific limitations and over-editing issues in LLM detoxification.

Findings

ToxEdit outperforms previous methods in detoxification effectiveness.

ToxEdit better preserves LLMs' general capabilities.

Enhanced benchmark shows improved evaluation of over-editing.

Abstract

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.