Bypassing the Safety Training of Open-Source LLMs with Priming Attacks

Autor:	Vega, Jason, Chaudhary, Isha, Xu, Changming, Singh, Gagandeep
Rok vydání:	2023
Předmět:	Computer Science - Cryptography and Security Computer Science - Artificial Intelligence Computer Science - Computation and Language Computer Science - Machine Learning
Druh dokumentu:	Working Paper
Popis:	With the recent surge in popularity of LLMs has come an ever-increasing need for LLM safety training. In this paper, we investigate the fragility of SOTA open-source LLMs under simple, optimization-free attacks we refer to as $\textit{priming attacks}$, which are easy to execute and effectively bypass alignment from safety training. Our proposed attack improves the Attack Success Rate on Harmful Behaviors, as measured by Llama Guard, by up to $3.3\times$ compared to baselines. Source code and data are available at https://github.com/uiuc-focal-lab/llm-priming-attacks. Comment: ICLR Tiny Paper camera ready version
Databáze:	arXiv
Externí odkaz:	http://arxiv.org/abs/2312.12321 Zobrazit plný text záznamu View this record from Arxiv