Search papers, labs, and topics across Lattice.
UC Santa Cruz
1
0
3
11
Poisoning a personal AI agent's Capability, Identity, or Knowledge triples its vulnerability to real-world attacks, even in the most robust models.