guardscope: Measuring Which Actions Runtime Controls Check in LLM Agent Systems
Summary
- Researchers introduce guardscope, a framework to measure which action paths runtime controls check in LLM agent systems.
- guardscope uses a static stage to predict checkable action paths and an empirical stage comparing decisions against kernel-level ground truth.
- The framework classifies executed actions as checked, bypassed, or prevented across various deployment topologies.
- Experiments evaluated two agent harnesses and real controls including Microsoft's governance toolkit and agentgateway.
Why it matters
As enterprises deploy LLM agents that execute actions through diverse channels, verifying the actual coverage of runtime security controls is critical for robust agent governance.