Understanding the Adversarial Landscape of Large Language Models Through the Lens of Attack Objectives

Abstract

Large language models have revolutionized artificial intelligence but face growing adversarial threats. This article presents the adversarial landscape from a novel perspective by categorizing threats into four key objectives: privacy breaches, integrity compromises, adversarial misuse, and availability disruptions.