A four-eyes check is a control in which a second, independent person reviews and approves a sensitive action before it takes effect, so that no one person can complete it alone. Payments, changes to bank details and access rights are typical. The second person must actually check the work, not just add a signature.
Why does a four-eyes check matter in AI agent work?
Agents change who does the first part of the work. If an agent drafts a payment and one person approves it, is that a four-eyes check? Our view is no. The agent is not an independent second pair of eyes; it is the drafter. For the actions that needed two people before agents arrived, they usually still need two people now: one to approve the agent's draft and one to check that approval, or two approvers with separate duties.
For job seekers, understanding this matters because four-eyes roles are common in finance, payments, lending and IT operations, and interviewers will test whether you understand what independence means.
How does a four-eyes check work in practice?
A well-run check has four features:
- Independence: the checker did not prepare the work and does not report to the person who did, where that can be avoided.
- Evidence: the checker looks at the source, not only at the first person's summary.
- Authority: the checker can reject, not just comment.
- Record: both names, the decision and the reason are logged.
On a governed platform, actions that move money or are hard to undo sit at the highest depths, and a business can set them to be held until a named person approves. Adding a second named approver for the riskiest of them is a choice the business makes in its own limits. How to set this up for changes to systems and settings is covered in how to run four-eyes checks on operational changes with an AI agent.
What is an example?
An agent at a wholesaler drafts a change to a supplier's bank details after receiving an email with new details on letterhead. The email is external information: it can start a process, but on its own it cannot authorise a payment. An accounts assistant we will call Mia calls the supplier on the number already on file and confirms the change. Her team lead then reviews Mia's call note and the supplier record, and approves. Two people, each checking independently, before any money moves to the new account.
How is a four-eyes check different from the four-eyes principle or a second pair of eyes?
The four-eyes principle is the rule that certain actions need two people; it is explained in what the four-eyes principle is. A four-eyes check is that rule carried out on a specific action. A second pair of eyes is looser: an informal review that may not carry the power to stop the action, covered in what a second pair of eyes is.
The difference matters in interviews and audits. If the second person cannot say no, it is not a four-eyes check.
Which terms are related?
- Segregation of duties: splitting a process so different people hold different steps.
- Second approver: the named person who completes a four-eyes check.
- Maker-checker: another name for the same control, common in banking.
For how this fits the skills employers test, read AI skills for job seekers.
Questions people ask
Does checking someone's certificate count as a four-eyes check?
No. That is a verification step, not a second approval of an action. It is still worth doing well; how employers verify a credential is set out in how an employer can check an AI agent certificate.
What do I do if I am asked to approve something I was not shown the evidence for?
Ask for the evidence or decline to approve. A checker who approves without seeing the work has turned the control into a signature, and their name is on the record.