What's in Your "Safe" Data?: Identifying Benign Data that Breaks Safety

arXiv:2404.01099v1 Announce Type: new
Abstract: Current Large Language Models (LLMs), even those tuned for safety and alignment, are susceptible to jailbreaking. Some have found that just further fine-tuning an aligned model with benign data (i.e., data without harmful content) surprisingly leads to substantial degradation in safety. We delve into the data-centric aspects of why benign fine-tuning inadvertently contributes to jailbreaking. First, we represent fine-tuning data through two lenses: representation and gradient spaces. Furthermore, we propose a bi-directional anchoring method that …

abstract alignment arxiv cs.ai cs.cl cs.cr cs.lg current data fine-tuning found jailbreaking language language models large language large language models leads llms safe safety type