1 paper · 1 filter
Zhengyang Shan, Jiayun Xin, Yue Zhang +1
Code agents powered by large language models can execute shell commands on behalf of users, introducing severe security vulnerabilities. This paper presents a two-phase security an…