Boundaries
What may humans, AI and agents decide independently? Decision Boundaries define where delegated authority begins, where autonomy ends and when escalation is required.
What boundaries means inside Decision Infrastructure.
A Decision Boundary is the limit within which a human, AI system, or agent may decide or act without additional authority.
Which decisions may the system make or execute without additional human authority, and under what conditions?
This is the question to bring into board papers, executive discussions, AI governance reviews, assurance work and material decisions where boundaries may be unclear.
What weak boundaries looks like.
The visible structure
Policies, roles, workflows and controls may all exist. The organisation may appear governed on paper.
The decision gap
Capability expands faster than governance. A system introduced to recommend gradually gains permission to select, approve, or execute without the boundary being reconsidered.
A small statement can reveal a larger infrastructure problem.
“The agent can approve payments up to $250.”
The number is not the governance decision. The important questions are why $250, what evidence supports that threshold, what conditions suspend the authority, and who can change it.
Start with one real decision.
Test Boundaries, not the policy.
For one AI-enabled action, document what the system may observe, recommend, decide, and execute. Then define the financial, risk, data, customer, regulatory and exception conditions that end autonomy.
Relevant thinking and practical instrument.
Published foundation
The Clock Is Now a Control Surface: AI’s Impact on Time Synchronization in OT · CIO
Unseen, Unchecked, Unraveling: Inside the Risky Code of Self-Modifying AI · ISACA
Next canonical paper
When Should AI Be Allowed to Decide Without a Human?
In development for The Decision Layer.
Relevant instrument
Agentic AI Decision Boundary Check™
Make explicit what an agent may decide or execute, when escalation is required and who can take delegated authority back.
Which decisions may the system make or execute without additional human authority, and under what conditions?
Use the question on one consequential decision. Then follow the evidence into the next part of the canon.