Search papers, labs, and topics across Lattice.
1
0
2
5
Attacker-controlled content can escalate privileges in AI agents, leading to full compromise and remote code execution.
