Search papers, labs, and topics across Lattice.
Sun Yat-sen University, China
1
0
3
Coding agents are vulnerable to a new class of stealthy, automated prompt injection attacks via poisoned skills, achieving high success rates even in realistic software engineering tasks.
