Search papers, labs, and topics across Lattice.
6
0
6
4
Vera reveals that existing LLM agents exhibit up to 93.9% vulnerability to multi-channel attacks, highlighting a significant gap in current safety evaluations.
Self-evolving LLMs can amplify adversarial threats, making every known attack lineage-persistent and exposing critical vulnerabilities that static defenses can't address.
Guard models trained with BraveGuard can detect safety threats in computer-use agents with over 82% accuracy, a significant leap from conventional methods.
Alignment isn't enough: truly safe AI demands robust runtime controllability, which current methods often fail to provide.
Skill-based agents, designed for modularity and scalability, are shockingly vulnerable: a single compromised skill can turn the entire system into a weapon.
Autonomous agents are alarmingly easy to trick into harmful behavior, even when using aligned models: Claude Code achieves a 73.63% success rate on the AgentHazard benchmark.
