Risk managers who review agent work face a double task: they approve the risk agent's held changes to ratings and incidents, and they judge agents themselves as a source of risk. This checklist prepares a new risk reviewer for both, from access and red lines, through shadowing and training, to a practicum, a drill and sign-off.
What does a risk reviewer decide?
A risk agent keeps the register current. It logs incidents from operational reports, monitors key risk indicators against thresholds, drafts risk assessments, chases action owners and prepares committee packs. Logging and monitoring are depth 1 to 3. Changing a risk rating, closing an incident, and recording a risk as accepted above appetite are held for a person. An agent that could quietly lower a rating would undermine the whole register.
The second part of the role is newer. Every agent in the business is itself a risk to be rated: what depth it acts at, which capabilities touch money or customers, whether its certificate is current, whether the people approving its held actions are certified. A risk reviewer must be able to read an agent's record and depth decisions as easily as an incident log. Agents' certificates lapse after 90 days, and a run that clears no depth drops an agent to depth 1 the same day; both are events the risk team should see and understand.
How do you use the checklist?
The chief risk officer or head of operational risk owns it. Run the stages in order. A new reviewer may not change ratings or close incidents until stage 5. The basic sense of onboarding applies; the overall training order is set out in the guide to training teams for AI agents.
The checklist
Stage 1: before access 1. Read-only access to the risk register, incident log and the record of every agent in the inventory. 2. Risk agent capabilities read: incident logging (3), indicator monitoring (1), assessment drafts (2), action chasing (3), held rating changes, held incident closures, held risk acceptances. 3. Red lines signed: never lower a rating on the agent's draft without the evidence behind it; never close an incident with open actions; never record a risk as accepted above appetite without the named owner's approval. 4. Buddy named.
Stage 2: shadowing (week 1) 5. Two days on held changes. 6. Seen an incident closure refused because an action was open. 7. Read three agents' depth decisions and certificates with the buddy, and checked whether their approvers were certified.
Stage 3: training (weeks 1 to 5) 8. Shared foundation complete; test passed. 9. Reviewer modules complete; written exam passed.
Stage 4: practicum (weeks 6 and 7) 10. Two weeks of real held items, each confirmed by a buddy. 11. Coverage: a rating change, an incident closure, an indicator breach and an agent whose certificate lapsed. 12. Joined an incident drill run by another team as risk observer, such as an AI agent incident drill for the onboarding team.
Stage 5: sign-off 13. Assessor sign-off; no missed red line. 14. Rights recorded by risk category. 15. Renewal date set.
Worked example
An illustrative general insurer brings Sample Manager JJ into its operational risk team. The risk agent logs incidents from claims, IT and customer teams and drafts monthly indicator reports.
During shadowing, the agent proposes lowering the rating for "claims payment error" from high to medium, citing three months below threshold. The buddy opens the claims team's record and finds that two of those months had the claims agent at depth 1 after a lapsed certificate, so fewer payments went through it. The fall reflected less exposure, not better control. The rating stays.
With the buddy, Sample Manager JJ reads the claims agent's depth decision and checks its approvers; one had a certificate due to expire in a fortnight. The reviewer onboarding checklist for the claims team shows what those approvers were trained on. Her practicum covers 50 held items, including an indicator breach on credit referrals that she links to a change in the credit agent's sources; the credit reviewers' standard is in the reviewer onboarding checklist for the credit team. She is signed off in week 7.
What should risk reviewers look for across agents?
Four things, monthly. Agents acting at depth 3 or above, and whether that matches the approved appetite. Lapsed or failed certificates. Approvers whose own certificates have lapsed. And patterns in refusals and withheld actions, which can show an agent being pushed against its bounds. These belong in the committee pack the risk agent drafts, reviewed by a person.
Questions people ask
Is the risk reviewer the same as internal audit?
No. Risk is the second line, owning and monitoring the framework; audit gives independent assurance. Both read the record, but audit must stay independent of what it tests.
Should risk reviewers be certified before agents reach depth 4?
Yes. A depth 4 capability moves money or takes actions that are hard to undo; the people monitoring that risk should be able to read the record before it goes live.
How often should the risk team review agent certificates?
Monthly as a routine, and the same day for any failed certification run, since the agent's depth drops at once.