PRP: Propagating Universal Perturbations to Attack Large Language Model Guard-Rails

arXiv:2402.15911v1 Announce Type: new
Abstract: Large language models (LLMs) are typically aligned to be harmless to humans. Unfortunately, recent work has shown that such models are susceptible to automated jailbreak attacks that induce them to generate harmful content. More recent LLMs often incorporate an additional layer of defense, a Guard Model, which is a second LLM that is designed to check and moderate the output response of the primary LLM. Our key contribution is to show a novel attack strategy, …

arxiv attack attacks automated cs.cl cs.cr defense guard humans jailbreak language language models large large language model llms prp rails work