What is AI Agent Decision Authorization?

When an AI system only answers questions, evaluation and monitoring can be enough. When an agent can move money, update a record, send a customer message, or invoke a tool, a different question appears: may this exact action proceed?

Decision Authorization is that question, answered at runtime. OntoGuard evaluates the proposed action against intent continuity, governing basis, evidence, and authority, and returns ALLOW, BLOCK, or ESCALATE. A score is not authorization. A policy document is not authorization. A prior ALLOW does not bind a mutated action.

This is not the same problem as model quality, prompt safety, or identity. Those matter. They do not decide whether a specific state transition may become an enterprise consequence.

Related: Authorization vs IAM/OAuth, ALLOW / BLOCK / ESCALATE, vs guardrails.