Deontic Policies for Runtime Governance of Agentic AI Systems

2026-06-17 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Cybersecurity & Data Privacy · Depth: Expert, quick

Summary

A new paper introduces AgenticRei, a system designed to provide comprehensive runtime governance for autonomous agentic AI systems powered by Large Language Models (LLMs). These systems, capable of invoking tools, manipulating data, and coordinating with other agents, present complex security, privacy, and compliance challenges that exceed the capabilities of existing policy engines like XACML, Rego, and Cedar. Current engines primarily handle "permit/prohibit" rules but lack features such as obligation lifecycle management, meta-policy conflict resolution, dispensations for waiving obligations, and ontological reasoning over domain hierarchies. AgenticRei addresses these gaps by implementing a deontic policy language based on the Rei framework, expressed in OWL (Web Ontology Language), and evaluated by a high-performance logic engine external to the LLM. This approach governs both agent tool invocations and inter-agent messages, demonstrating its ability to capture critical security and privacy governance constraints that are not expressible in current production systems, while also composing with industry-standard frameworks like A2AS.

Key takeaway

For AI Architects or AI Security Engineers designing governance for autonomous LLM-driven agents, you must move beyond basic permit/prohibit access controls. Your systems require a deontic policy framework like AgenticRei to manage obligations, resolve policy conflicts, and handle dispensations. This ensures comprehensive security, privacy, and compliance, especially when agents invoke tools or communicate across organizational boundaries. Consider integrating external logic engines for robust runtime policy evaluation.

Key insights

AgenticRei extends AI governance beyond permit/prohibit rules to include obligations, dispensations, and conflict resolution for LLM agents.

Principles

• LLM agent governance needs deontic policies.
• Current policy engines are insufficient for agentic AI.
• External logic engines enhance LLM governance.

Method

AgenticRei uses a deontic policy language built on the Rei framework, expressed as OWL, and evaluated at runtime by a high-performance logic engine outside the LLM to govern tool invocations and agent-to-agent messages.

In practice

• Implement obligation lifecycle management.
• Integrate meta-policy conflict resolution.
• Apply ontological reasoning for domain hierarchies.

Topics

• Agentic AI Governance
• Deontic Policy
• Large Language Models
• Runtime Policy Enforcement
• OWL Ontology
• Multi-agent Systems

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Scientist, AI Architect, AI Security Engineer

Related on AIssential

• Securing AI agents with agent governance — Applying operating system and distributed systems security principles is crucial for governing autonomous AI agents.
• Governance by Construction for Generalist Agents — The CUGA policy system embeds continuous, multi-stage governance into generalist LLM agent execution without fine-tuning, ensuring auditable and compliant behavior.
• Algorithmic Constitutionalism — Algorithmic constitutionalism offers a layered, self-monitoring framework for AI governance, moving beyond ethical engineering to protect core principles.
• The Containment Gap: How Deployed Agentic AI Frameworks Fail Public-Facing Safety Requirements — Deployed agentic LLM frameworks lack native architectural safety, making them vulnerable to persistent, targeted corruption.
• A Five-Plane Reference Architecture for Runtime Governance of Production AI Agents — A five-plane architecture provides runtime governance for production AI agents by mediating delegated actions.
• Governance Is Not a Prompt — Existing model risk management frameworks are inadequate for agentic AI, necessitating a new governance paradigm focused on external, enforceable controls.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.