Agent Security
-
Unified Misuse Monitoring Benchmark for LLM Agents
A 6,200-trace benchmark scores agent monitors on when they flag harm. Action-framed monitors catch injection; content-framed ones drop to AUC 0.52.
-
Physical Danger Probes for Embodied LLM Agents
The analysis explains how hidden-state probes can flag unsafe embodied-agent plans, with benchmark results, deployment limits, and layered controls.
-
AI Agent Security Tools for Runtime Protection
A comparison of AI agent security tools for runtime guardrails, tool-use sandboxing, scoped identity and authorization, and behavioral observability.