Abstract: In this paper, we propose a novel classification method that utilizes syntax trees and perplexity to identify jailbreak attacks that use hostile suffixes to make large language models (LLMs) ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results