01 / Source and scope
Who acts, and what changes?
Principal and authority origin
Organization deploying agentic AI. System owner and risk-based deployment decision.
Consequential action
Connect an agent to privileged tools.
02 / Action-time control
Make the requested effect testable.
Australian cyber guidance calls for cautious adoption and mitigations; it is not a product approval. Inventory tools and privileges, limit reach, monitor actions and hold high-impact effects for review.
OriginSystem owner and risk-based deployment decision
RequestConnect an agent to privileged tools
DecisionEvaluate and enforce before effect
03 / Evidence and responsibility
Permission, execution, and outcome are separate.
Tool inventory, permission set, runtime decision and incident test.
ProofGrid can support runtime control, while the operator owns secure deployment and monitoring.
Primary source