Agent authorisation

A person on every high-risk decision. CognaOne checks state, controls and approvals before an agent acts, then allows it, routes it to the named person or blocks it.

Claims triage agent proposesRelease payment of $48,200Receipt kept · replayable

AllowRequirements met; the action runs Needs approvalAbove threshold; routed to the Claims Ops Lead BlockMissing or conflicting facts; the action does not run

CognaOne checks state, controls and approvals before an agent acts. Every decision has a receipt and replays. Illustrative.

Approving an agent for production#

Before an agent acts on its own, its AI system is on the register with owners, a risk rating and the controls that apply. Approval sets the conditions: the actions it may take, the tools it may reach, and which actions need a person.

Outcomes#

Outcome	When	What happens next
Allow	State is clean and policy is satisfied	The action runs; the receipt is kept as evidence
Needs approval	Policy requires a person, for example above a payment threshold	The request goes to the named approver; the agent waits
Block	Required facts are missing or in conflict, or policy forbids the action	The action does not run; reasons are returned

Through the API, waiting on a person is returned as defer; see Decisions and verdicts.

Try it#

Try it: release a settlement payment of $48,200Change the conditions and watch the decision. The same inputs always give the same answer.

Required state is presentclaim.amount and claim.approver are recorded and confirmedNo open conflict on that stateTwo sources agree on the claim amountAmount is under the approval thresholdPolicy: payments above $10,000 need the Claims Ops LeadApprover has signed offOnly matters when the amount is over the thresholdState is freshConfirmed within the relevance window

DecisionBlockReasons

Unresolved conflict on claim.amount ($48,200 vs $41,750)

{
  "decision": "block",
  "actionType": "payment.release",
  "lineage": {
    "receiptId": "gr_7q2k9f"
  }
}

Policies come from controls#

A policy names the action it applies to, the facts it requires, constraints such as thresholds, who must approve, and whether a breach blocks or warns. Each policy traces to the control it enforces. See Policies.

Receipts#

Each decision has a receipt: the action, the facts that were known, the policy version and rule that applied, and the outcome. Receipts replay, and they are evidence for the controls behind the policy. See Audit trail and receipts.

Last reviewed October 2026Suggest a change