Search papers, labs, and topics across Lattice.
1
0
3
6
Web agents trained with adaptive adversarial simulation achieve 33.6% higher completion rates against unseen prompt injection attacks, even outperforming frontier adversaries.
