Open the Pandora's Box of LLMs: Jailbreaking LLMs through Representation Engineering

arXiv:2401.06824v2 Announce Type: replace
Abstract: Jailbreaking techniques aim to probe the boundaries of safety in large language models (LLMs) by inducing them to generate toxic responses to malicious queries, a significant concern within the LLM community. While existing jailbreaking methods primarily rely on prompt engineering, altering inputs to evade LLM safety mechanisms, they suffer from low attack success rates and significant time overheads, rendering them inflexible. To overcome these limitations, we propose a novel jailbreaking approach, named Jailbreaking LLMs through …

abstract aim arxiv box community cs.ai cs.cl engineering generate inputs jailbreaking language language models large language large language models llm llms pandora pandora's box probe prompt queries representation responses safety them through type