Search papers, labs, and topics across Lattice.
5
0
6
7
Securing autonomous AI agents demands a lifecycle-oriented approach, and AgentWard provides a blueprint for defense-in-depth across initialization, input processing, memory, decision-making, and execution.
A multi-agent system can dramatically improve the accuracy of LLM-based housing consultations, jumping from 75% to 95% correctness in real-world scenarios.
Autonomous LLM agents are riddled with vulnerabilities, as point defenses fail to address cross-temporal and multi-stage systemic risks like memory poisoning and intent drift.
Human-AI cybersecurity teams are held back by poor prompting, as autonomous AI agents that self-direct their tool use outperformed most human teams in a live CTF competition.
Humans are surprisingly vulnerable to deception by compromised LLM agents, with less than 10% detecting attacks even in high-stakes scenarios.
