Eraser: Jailbreaking Defense in Large Language Models via Unlearning Harmful Knowledge

arXiv:2404.05880v1 Announce Type: new
Abstract: Jailbreaking attacks can enable Large Language Models (LLMs) to bypass the safeguard and generate harmful content. Existing jailbreaking defense methods have failed to address the fundamental issue that harmful knowledge resides within the model, leading to potential jailbreak risks for LLMs. In this paper, we propose a novel defense method called Eraser, which mainly includes three goals: unlearning harmful knowledge, retaining general knowledge, and maintaining safety alignment. The intuition is that if an LLM forgets …

abstract arxiv attacks cs.cl defense eraser generate issue jailbreak jailbreaking knowledge language language models large language large language models llms paper risks type unlearning via