Governance agents, and where a person still decides
Nine agents run or prepare the platform's own governance. Every one of them has a point where a named person decides.
For: Model Risk & Governance, Platform Administrator, Marketplace Operator, Builder / Publisher · 1 min read
Three levels of agent involvement
- Agent runs it: rule-based work with no discretion, such as granting Internal access or scanning for PII.
- Agent prepares, a person decides: the agent assembles the evidence; a named person approves.
- A person decides: the agent stays out of the decision entirely.
The agents
| Agent | Level | Where a person decides |
|---|---|---|
| Use-Case Matching Agent | Agent runs it | Group AI Steering Committee ranks and funds the use-case pipeline each month |
| Use-Case Sizing Agent | Agent prepares, a person decides | Group AI Steering Committee confirms the size and commits budget |
| Intake Triage Agent | Agent runs it | A named approver decides every Confidential request the agent routes |
| Assurance Agent | Agent runs it | The Model Risk validator reads the check report before any validation opinion |
| Validation Case Agent | Agent prepares, a person decides | A Model Risk Management validator signs every model approval |
| Drift Watch Agent | Agent runs it | The business owner accepts the retrain cost before any job starts |
| Data Steward Agent | Agent runs it | The data owner signs off when the PII scan flags anything |
| Chargeback Agent | Agent prepares, a person decides | Group Finance clears variances and posts the chargeback journal |
| Compliance Evidence Agent | Agent runs it | The control owner reviews the pack before it goes to Internal Audit or the regulator |
How agents are controlled
Governance agents are catalogue assets under the same validation as any other. Each declares what it may read and write, returns to a safe state when it leaves its operating envelope, and reports how often a person overturned its recommendations.
Read next
Was this helpful?