What are AI approval controls?
AI approval controls define which actions an AI team may carry out, which require human review, and when it must pause or return to an accountable owner. They connect the customer's policy to the actual execution path.
Nodes works within configured permissions. Actions requiring approval remain gated by the customer's policy.
Where the human gate applies
Named people control consequential decisions and new production scope. A reviewer receives the evidence, uncertainty, and proposed action, then can question, edit, approve, delay, or decline it. A workflow can require a second signer when the customer designates that control.
Routine steps already authorized by policy do not need a new human click merely because another agent performs them. They remain inside the agreed tools, data access, and action boundaries. A new opportunity or proposed process change does not silently extend those permissions.
What remains in the record
The Decision Trace preserves the evidence, recommendation, human input, disposition, executed action, and later outcome. That record supports review of what happened and why.
Approval controls are one part of a deployment's governance. Their presence alone does not establish regulatory compliance or validate the quality of a recommendation.