Search papers, labs, and topics across Lattice.
1
0
2
Current benchmarks mislead on AI agent security; robust defenses against indirect prompt injection require dynamic replanning, constrained LLM usage, and human oversight.
