Control areas · Human oversight
Human Approval Controls for AI Agent Actions
Route consequential actions to someone who can understand and authorise the proposed change.
Design and verification
Show the decision
Display the target, action, impact and relevant arguments before approval.
Bind the approval
Bind approval to the proposed action and scope; invalidate it when the request changes or expires.
Provide a stop path
Define rejection, emergency stop and escalation, including behaviour when an approver is unavailable.
Example
An approval for a small refund should not authorise a larger payment with altered arguments. Check the approval against the exact action at execution.
Engineering scope
These are control design requirements. Implementation, supported versions and evidence mechanisms are selected and verified for the customer environment. See Guardrail Engineering.
Start with a clear scope
Tell us which systems, actions and review requirements are in scope. We will discuss the work, responsibilities and deliverables before you commit.