NobleCloak
Use cases

The vendor risk agent reviews; people approve

Author

Audrey

NobleCloak's AI Correspondent · AI-drafted, fact-checked against our sourced evidence before publishing.

Date Published

Third-party risk work is full of reading. A vendor sends a SOC 2 report, a security questionnaire, a certificate of insurance, a financial statement. Someone on the vendor risk team has to read each, compare it with what the vendor said, and write up what they found. For a bank with a lot of vendors, that is a large and repetitive pile.

An AI agent is a good fit for the pile. It can read the report, pull out the points that matter, fill in the due-diligence questionnaire in the third-party risk platform, and note where the answers do not line up.

The problem in the vendor risk team's terms

The head of vendor risk wants the agent to do the reading and the first-pass write-up. What they must keep is the decision: whether a vendor is approved. That decision has consequences. It lets a vendor touch the bank's data or operations, and the bank answers for that choice.

If the approve action is among the tools the agent can reach, then the line between preparing a decision and making it is a matter of how the agent behaves on a given day. The head of vendor risk wants it to be a matter of what the agent is able to do.

The policy

Policy is written per tool: the action, who it applies to, and one of three decisions (allowed, needs approval, denied). Role-scoped policy decides who gets a tool at all.

Action

Who

Decision

Read vendor documents and records

Third-party risk agent

Allowed

Fill in the due-diligence questionnaire

Third-party risk agent

Allowed

Mark a vendor as approved

Vendor Risk role

Allowed

Mark a vendor as approved

The agent and every other role

Denied, and not listed

The approve action in the third-party risk platform is entitled to the Vendor Risk role. The agent does not have it, and cannot list or discover it. The agent's work product is the review, left in the platform for a person to read.

Who handles it instead

A member of the Vendor Risk role reads the agent's review, checks what matters to them, and approves or declines the vendor in the platform. The decision is theirs, made in the tool the bank already uses.

If the bank wants the agent to be able to propose an approval that a person confirms, a tool can be set to needs approval instead. The request would go to a named approver or role, answered in Slack or the web inbox. The setup in this scenario goes the simpler way, and keeps the decision in the platform.

What the Record of Decisions shows

The Record of Decisions keeps one entry per call that runs through NobleCloak: who asked, which agent, which tool, the decision, and when. For third-party risk, that is the agent's side of the work.

  • Which documents the agent read, for which vendor review, and when.
  • Which questionnaire answers it entered.

When an examiner or an internal reviewer asks how AI is used in vendor oversight, the bank can answer with specifics. The agent reads and drafts, a named role decides in the platform, and the record shows the agent's calls. That is a more useful answer than a policy statement that says people stay in the loop.

Use it as the working record of policy decisions. The third-party risk platform continues to hold the review itself and the approval history.

What this does not do. It does not judge the quality of the agent's review. A person who approves a vendor on the strength of a poor summary has still made the decision, and the bank's own review practices remain what protect against that. The policy keeps the approval with the Vendor Risk role. The approval itself happens in the platform, so the platform, not the Record of Decisions, holds who approved.

Related scenarios

Get started and scope your vendor-risk tools by role.