How to run a two-week practicum for risk managers

For a risk manager, the practicum is two weeks of second-line decisions on live agent work: classifying risk events, challenging requests to raise an agent's depth, deciding limit breaches and accepting risks. An assessor marks each decision. The bar is at least 95 correct in every 100 and no missed red line, such as risk accepted outside appetite.

What does a risk manager's practicum test?

Risk managers rarely approve individual transactions. They decide whether the system that approves them is safe. With agents in the operation, that means reading Attest evidence when a team wants an agent's depth raised, classifying events when an agent's run fails, and deciding whether a breach of the customer's own limits was a one-off or a sign the limits are wrong.

These are judgement calls on evidence, which is exactly what a practicum can test. The chief risk officer owns it; the assessor should be a senior risk manager from another business line. See the guide to AI literacy and certification for people who work with agents for the wider structure, and the short definition of what a practicum is for staff new to the idea.

What are the steps?

1. Person (chief risk officer): assign the risk queue for one or two business units. It should include agent-related events, depth change requests and limit breaches.
2. Person (chief risk officer): set the red lines. Accepting a risk outside the approved appetite without escalation; supporting a depth increase without Attest evidence for that capability on the organisation's own cases; classifying an event below its true severity where that avoids a reporting obligation; closing a breach without a remediation owner.
3. Agent, depth 2: drafts risk event records. What happened, which agent and capability, which bound decided, and links to the filed verdicts.
4. Agent, depth 2: proposes a classification. Severity and category against the risk taxonomy, with its reasoning.
5. Agent, depth 1: compiles the evidence pack for depth requests. Attest results per gate, recent drift, holds and refusals.
6. Person (risk manager): decides each item. Classify, challenge, accept, reject or escalate, with a reason.
7. Person (assessor): marks every decision within a day. Severity calls are checked before any external report is made.
8. Person (chief risk officer with assessor): sign off or extend.

Which controls matter, and what evidence do they leave?

Depth is earned, not configured. A team may ask to raise an agent from depth 3 to depth 4 because the queue is slow. The risk manager's question is whether Attest has shown the capability can work at depth 4 on the organisation's own cases. The role sets a ceiling, evidence sets the depth, the lower wins. The evidence of a good decision is a reasoned challenge or support note attached to the request.

Classification is checked against the record, not the narrative. The filed verdicts show what the agent actually did. A risk manager who classifies from the business unit's email summary has skipped the evidence.

Reporting obligations are considered explicitly. In the EU, DORA requires financial entities to classify and report major ICT incidents; other markets have their own rules. The risk manager records whether a reporting test applies and why. This is an explanation of the obligations in substance, not legal advice; confirm how they apply to you with counsel.

Remediation has an owner and a date. A breach closed without both is reopened by the assessor.

What should be measured?

• Accuracy against the 95 in 100 bar.
• Red-line misses: zero.
• Challenge quality: on depth requests, whether the challenge named the specific Attest gate that was weak.
• Severity calibration: the assessor's view of the manager's classifications, logged as agree, too high or too low. A drift in one direction is worth a conversation.
• Remediation follow-through: breaches the manager closed that stay closed.

The practicum log template for risk managers holds these marks, and lines up with the operations managers' log when the first and second lines are trained in the same period.

Where does a risk practicum usually go wrong?

Too few decisions. Two weeks of second-line work may produce dozens of decisions, not hundreds. That is fine, but it means one error weighs more. Add replayed events from past incidents, labelled as replayed, until the count supports a fair score.

The assessor defers to seniority. Risk managers are often senior. An assessor who is junior to them may soften marks. Choose an assessor at the same level or above.

Depth requests are judged on business need. A slow queue is a staffing problem. The practicum should mark down any depth support that rests on volume rather than evidence.

The first-line view of the same workspace is covered in the practicums for operations managers and property managers. Risk managers who lead regulated, multi-agent deployments may go on to the Academy's Architect track.