Search papers, labs, and topics across Lattice.
1
0
2
A new seven-level harm scale reveals hidden vulnerabilities in AI agent defenses that binary metrics overlook, exposing risks even when attack-success rates appear low.
