DUALLM, a dual-method pipeline that integrates two approaches based on a Large Language Model (LLM) and a fine-tuned small language model, achieves 87.4% accuracy and an F1-score of 0.875, significantly outperforming prior solutions.
Xing-Yu Li, Jue-Fei Pu, Yifan Wu et al.· Network and Distributed Syst...· 1 citation
It is found that although the new protection measures aimed at stopping popular heap exploit techniques or restricting the exploit strategy space can effectively defend against the exploitation of most vulnerabilities, the attack strategies proposed by this work can still make these vulnerabilities exploitable again.
The AI Companion Vulnerability-Response Taxonomy is introduced, a grounded, paired taxonomy of user vulnerability and chatbot response designed for analyzing extended companion chatbot interactions and reveals distinct response profiles of AI companions in conversations with vulnerable users.