PROVISO Controller for Provenance-Informed Secure LLM Agent Actions
October 21, 2026
PROVISO is a security controller that selects agent actions by analyzing the provenance of information from users versus external tools. By comparing proposed actions against alternatives using provenance-typed memory, it improves both security and utility over existing baseline defenses.
HOW THIS AFFECTS YOU
●
builderYou can integrate this controller to prevent agents from executing harmful actions triggered by untrusted third-party tool outputs.
●
policyThis offers a technical mechanism to enforce provenance-based safety boundaries in autonomous systems.