This article defines agentic governance as runtime enforcement infrastructure controlling what autonomous AI agents are permitted to do, distinguishing it from principle-based frameworks.
•Two competing definitions exist: governance as principles/accountability structures vs. governance as runtime enforcement independent of the agent's own reasoning
•Four properties make AI systems agentic and amplify governance risk: autonomy, tool access, persistence, and delegation across multi-agent hierarchies
•Runtime governance spans five enforcement domains: access (tool/data permissions), cost (token budget limits per session), content (PII and sensitive data filtering), quality (output confidence thresholds), and operational (behavioral pattern escalation)
•Governance is distinct from observability (records actions), prompting (instructs agents), testing (reduces probability), and compliance documentation (none of these enforce at runtime)
•The gap between framework governance and runtim
This summary was automatically generated by AI based on the original article and may not be fully accurate.
Google Cloud
Databricks
Stripe