Search papers, labs, and topics across Lattice.
Harbin Institute of Technology, Peng Cheng Laboratory
1
0
2
Malicious content can be injected into web agents' memories, achieving up to 99.15% attack success without affecting their benign performance.
