Vocabulary Attack to Hijack Large Language Model Applications

arXiv:2404.02637v1 Announce Type: new
Abstract: The fast advancements in Large Language Models (LLMs) are driving an increasing number of applications. Together with the growing number of users, we also see an increasing number of attackers who try to outsmart these systems. They want the model to reveal confidential information, specific false information, or offensive behavior. To this end, they manipulate their instructions for the LLM by inserting separators or rephrasing them systematically until they reach their goal. Our approach is …

applications arxiv attack attackers confidential cs.ai cs.cr cs.dc driving fast hijack information language language models large large language model llms reveal systems try vocabulary