AI RESEARCH
Beyond Suffixes: Token Position in GCG Adversarial Attacks on Large Language Models
arXiv CS.LG
•
ArXi:2602.03265v2 Announce Type: replace Large Language Models (LLMs) have seen widespread adoption across multiple domains, creating an urgent need for robust safety alignment mechanisms. However, robustness remains challenging due to jailbreak attacks that bypass alignment via adversarial prompts. In this work, we focus on the prevalent Greedy Coordinate Gradient (GCG) attack and identify a previously underexplored attack axis in jailbreak attacks typically framed as suffix-based: the placement of adversarial tokens within the prompt.