Risk managers approve the agent's drafts and register updates. Changes to limits that bound other agents need the risk owner and a second approver, with those agents' owners told first. Accepting a risk outside appetite stays with the senior person accountable for it. Closing a significant incident needs the risk manager plus the business owner. Rights lapse with certificates.
What does a risk team's approver rule cover, and who owns it?
Risk teams in the second line set appetite, track indicators, record incidents and challenge the business. Agents help by monitoring key risk indicators, drafting assessments, updating the register and pulling evidence from other agents' verdict records. The risk team also owns something unusual: some of the limits other agents must obey. When a limit changes, every agent bound by it behaves differently from the next action on.
The chief risk officer owns the rule. The board risk committee is told of it. The agent owner configures it. Our guide to AI agent skills by role covers what risk managers need to know.
What are the steps?
- Chief risk officer: lists the agent's actions by depth. Depth 1: reads the risk register, indicators, loss events and other agents' verdict records. Depth 2: drafts risk assessments, indicator reports, incident reports and risk acceptance papers. Depth 3: updates the register, raises indicator breach alerts and assigns actions to owners. Depth 4: changes a limit that bounds other agents, records a risk acceptance, closes a significant incident.
- Chief risk officer: assigns drafts to risk managers covering each area.
- Chief risk officer: puts limit changes under two people: the risk owner and a second approver who did not request the change. Owners of every agent bound by that limit are told before it takes effect.
- Chief risk officer: keeps risk acceptance with the senior person accountable for the risk, as your governance requires.
- Chief risk officer: assigns significant incident closure to the risk manager plus the business owner.
- Chief risk officer: ties each role to a current certificate. Agent owner configures and files.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Monitor indicators | Reads data and records | 1 | Nothing |
| Breach alert | Raises and assigns | 3 | Owner responds |
| Draft assessment | Drafts with evidence | 2 | Risk manager approves |
| Change a limit | Prepares, lists affected agents | 4 | Risk owner plus second approver |
| Risk acceptance | Drafts paper | 2 | Accountable senior person |
| Close incident | Prepares closure note | 4 | Risk manager plus business owner |
Why does a limit change need the affected agents' owners told?
Each agent's actions meet several bounds at once: its role scope, its certificate, its sources, the customer's own limits and the platform guardrails. The lowest one wins. When the risk team changes a limit, it changes one of those bounds for every agent that reads it. A higher refund limit set by risk means a support agent may approve refunds it held yesterday.
That is often the intended effect. But the owners of those agents should know, because they will see the change in their own verdict records and may need to brief reviewers. If a limit is loosened, the owner may want the agent re-certified at the new level before it acts on it.
The second approver for a limit change checks the reason, the evidence and the list of affected agents. As our explainer on what a second approver is says, their job is a specific independent check, not a second opinion on everything.
Which controls make the rule work, and what evidence do they leave?
Two approvers on limit changes. Evidence: both names on each change, as set out in the approver authority matrix for the risk team.
Affected owners told. Evidence: the notice and acknowledgements before the change took effect.
Risk acceptance by the accountable person. Evidence: their name and the paper on each acceptance.
Incident closure by two people. Evidence: risk and business names on each closure.
Rights lapse with certificates. Evidence: certificate status on each approval date.
What should you measure?
- Limit changes without affected owners told. The target is none.
- Indicator breaches without an owner response within your target.
- Risk acceptances past their review date.
- Incidents reopened after closure.
- Approver coverage, compared with the approver authority matrix for the operations team for the business side.
Where does it usually go wrong?
Risk becomes first line. Risk managers approve business actions because they understand the limits. That removes their independence.
Risk acceptances never expire. A risk accepted for six months while a fix is built is still accepted two years later. Give every acceptance a review date, and have the agent raise it when the date arrives.
Limits change in a spreadsheet. Someone updates a limit in a document that agents do not read. Keep limits in one place, as a source the agents read.
Risk and compliance approve each other. Compare the approver rule for the compliance team: the two should challenge each other, not sign for each other.
Payments limits change at the wrong time. A limit change during a busy settlement window can affect thousands of payments. Coordinate with the approver rule for the payments operations team.
Where the Academy fits
Pryme Intelligence Academy's Foundations track explains the decision chain and how bounds combine, which risk managers need before approving limit changes. The Architect track is for people leading regulated, multi-agent deployments, where one limit may bind many agents. Architect certification is valid for 24 months and includes a board review with two Pryme assessors.
Questions people ask
Can the agent recommend a limit change?
Yes, as a draft with evidence. The risk owner decides whether to propose it, and two people approve it.
Who owns limits set by the business rather than risk?
The business process owner, through their own approver rule. Risk should see those changes and may challenge them.
Should risk managers read other teams' verdict records?
Yes, as part of second-line monitoring. The agent can sample records against approver rules, with risk managers reviewing exceptions.