Search papers, labs, and topics across Lattice.
1
0
2
2
Current benchmarks mislead on the security of AI agents against indirect prompt injection; robust defenses require dynamic replanning, constrained LLM-based security checks, and human interaction.
