Deontic Policies for Runtime Governance of Agentic AI Systems
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
Researchers propose AgenticRei, a runtime governance system for agentic AI that uses deontic policies to manage permissions, prohibitions, obligations, waivers and conflict resolution. The approach is designed to constrain LLM-driven agents’ tool use and agent-to-agent messages through an external logic engine, addressing security, privacy and compliance needs beyond current policy engines such as XACML, Rego and Cedar.
Researchers have developed a new framework called AgenticRei to govern autonomous AI systems in real time, addressing gaps in existing policy engines by incorporating obligations, conflict resolution, and waivers alongside traditional permit/prohibit rules. The approach uses a deontic policy language to enforce enterprise-level governance, including security and privacy constraints, without relying on the AI model itself. The system aims to better manage complex AI agent behaviors across organizational boundaries.