Open Sesame! Universal Black Box Jailbreaking of Large Language Models
This addresses security vulnerabilities in LLMs for AI developers and researchers, though it is incremental as it builds on existing jailbreak techniques with a novel automated approach.
The paper tackles the problem of exploiting alignment in large language models (LLMs) to produce harmful outputs, introducing a genetic algorithm that optimizes a universal adversarial prompt to disrupt model alignment, achieving automated jailbreaking as demonstrated through extensive experiments.
Large language models (LLMs), designed to provide helpful and safe responses, often rely on alignment techniques to align with user intent and social guidelines. Unfortunately, this alignment can be exploited by malicious actors seeking to manipulate an LLM's outputs for unintended purposes. In this paper we introduce a novel approach that employs a genetic algorithm (GA) to manipulate LLMs when model architecture and parameters are inaccessible. The GA attack works by optimizing a universal adversarial prompt that -- when combined with a user's query -- disrupts the attacked model's alignment, resulting in unintended and potentially harmful outputs. Our novel approach systematically reveals a model's limitations and vulnerabilities by uncovering instances where its responses deviate from expected behavior. Through extensive experiments we demonstrate the efficacy of our technique, thus contributing to the ongoing discussion on responsible AI development by providing a diagnostic tool for evaluating and enhancing alignment of LLMs with human intent. To our knowledge this is the first automated universal black box jailbreak attack.