Bring compliance and risk in at design, not at sign-off. The second line should set the red lines trainees must never miss, check that tests use your own regulated cases, sample review decisions during practice, and sign off the rule that ties approval rights to current certificates. They stay independent: they examine the training, they do not run it.
What is the second line's part, and who owns the training?
The first line runs agents: reviewers, approvers, team leads and agent owners. The second line, usually compliance and risk, sets standards and checks that the first line meets them. Our explainer on the second line of defence covers the model. Training for agent operations sits with the first line, normally a learning and development lead working with the process owner. The second line has a defined part in it, and needs enough of the skills in our guide to AI agent skills by role to play it.
That part matters because the training decides who may approve what an agent does. If reviewers learn the wrong red lines, every held action they release carries that error. The case for having compliance in the room is made in why the second line should sit in your AI agent training; this playbook is about how.
What are the steps?
- Learning lead: maps training to the agent's capabilities. For each capability and depth, which role reviews or approves, and what they must know.
- Second line: writes the red lines. The actions a reviewer must never release: payments on an external claim alone, advice outside permissions, decisions with legal effects made without a person. These become the "no missed red line" test in practice.
- Second line: reviews test content for regulated accuracy. Questions should be set in the team's own industry and role, and answers must match current policy.
- Trainees: complete practice on real review decisions. The agent drafts and holds; the trainee decides. The agent's depth does not change while a trainee is learning.
- Second line: samples practice decisions independently, focusing on held actions near the red lines and on refusals the trainee overrode or questioned.
- Process owner: drafts the approver rule: who may approve each depth, and which certificate must be current.
- Second line: signs off the approver rule and the red-line list, and sets how often they will sample the live record afterwards.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Map training | Its capabilities define scope | per capability | Learning lead |
| Write red lines | Refuses or holds at those lines | 1 to 4 | Second line |
| Review tests | Nothing | none | Second line |
| Practice | Drafts and holds real work | 2 to 4 | Trainee decides; assessor checks |
| Sample practice | Its verdicts are the evidence | none | Second line |
| Approver rule | Holds actions for named approvers | 3 and 4 | Process owner drafts, second line signs |
How do you keep the second line independent?
Independence is the reason the second line is useful. If a compliance officer becomes an everyday approver of the agent's held actions, they are first line for that work and cannot also check it. Our explainer on what a second approver is is clear that the second approver on a payment is a first-line control, not a second-line one.
So the second line contributes content and checks outcomes. They write red lines, they review tests, they sample decisions. They do not mark the practicum, approve live actions or own the agent. Where a small firm has one person doing both, write down which hat they wear for each task, and get someone else to check their first-line approvals.
The second-line separation checklist for compliance agents covers the same question when the agent itself does compliance work.
Which controls show the second line is involved, and what evidence do they leave?
A signed red-line list. Evidence: the list, versioned, with the second-line signatory and date of approval filed with the training pack.
Test content reviewed. Evidence: a review note on each test bank recording what was changed and why.
Independent sampling. The second line chooses its own sample from the practice decisions and later from the live verdict record. Evidence: the sample, the findings and the first line's response.
Approver rule signed off. Evidence: the approver rule with the second-line sign-off and the date it takes effect.
Re-sampling after change. When the agent's depth, sources or limits change, the second line samples again. Evidence: the change record linked to the new sample.
What should you measure?
- Red-line misses in practice, which should be none before anyone approves live work.
- Second-line findings per sample and how quickly the first line closes them.
- Approvers acting without a current certificate. The target is none.
- Refusals overridden by changing a limit, which the second line should see every time.
Where does it usually go wrong?
Sign-off at the end. Compliance is shown a finished course and asked to approve it. They can only say yes or no. Bring them in when red lines are written.
The second line becomes the approver. Busy teams route hard held actions to compliance because they "know the rules". It feels safe and removes the check.
Red lines in the course do not match the agent's bounds. Trainees learn one list; the agent refuses on another. Keep them as one list owned in one place.
Sampling stops after launch. The second line reviews training once, then never looks at live decisions.
Where the Academy fits
In Pryme Intelligence Academy, the Certified Pryme Operator practicum is two weeks of real review decisions, with at least 95 of every 100 correct and no missed red line, signed off by a Pryme assessor. Your second line can add its own red lines on top and sample the same decisions. Module tests are set in the learner's own industry and role, which gives compliance something concrete to review.
Questions people ask
Should compliance staff take the training themselves?
Yes, at least Foundations, so they understand depth, held actions and the decision chain. Some will take the Operator track to review it properly, without becoming first-line approvers.
Who owns the red-line list?
The second line owns its content; the agent owner makes sure the agent's bounds match it. One list, one owner, with changes recorded.
How often should the second line sample?
Set it by risk. Sample more where the agent acts at depth 4 or where decisions affect customers' rights, and again after any change to the agent.