Search papers, labs, and topics across Lattice.
3
0
3
2
Security benchmarks for LLM agents are fundamentally flawed: changing a tool's name can drastically alter reported attack success rates, invalidating cross-model comparisons.
Traditional cyber deception is breaking down: LLM-driven attack agents can now spot honeypots mid-reconnaissance and dynamically reroute their compute budgets toward real, high-value targets.
Forged reasoning attacks can completely compromise LLM agents' memory integrity, achieving up to 100% success against existing defenses.
