A regulator asks who changed an agent's grounding data and when. What should your design include?
Why C is the answer
An audit trail for changes to models and data
An audit trail records who changed models and data, and when. Access controls on grounding data support it.
In detail
- Only an audit trail records who changed what, and when.
- Restrict who can change grounding data.
Use case
An illustrative scenario, not a real incident.
A bank must show who changed a policy document that an agent used. An audit trail records the person, the change and the time.
See it in different ways
flowchart TD A["AI governance"] --> B["Microsoft Purview: data protection"] A --> C["Managed environments: Power Platform control"] A --> D["Audit logs: who changed what"]
sequenceDiagram participant D as Dev participant T as Test participant P as Production D->>D: Unmanaged solution D->>T: Export as managed T->>P: Import after approval P->>P: Audit trail of the change
sequenceDiagram participant P0 as Editor participant P1 as Grounding data participant P2 as Audit trail participant P3 as Regulator P0->>P1: Change document P1->>P2: Record who and when P3-->>P2: Request history
Key point: Audit changes to models and data.
Why the other options are wrong
- A. A more detailed system prompt that describes each change
- A prompt is not a record.
- B. Faster model inference on the grounding pipeline
- Speed has no bearing on audit.
- D. Larger Dataverse capacity to hold more history
- More capacity does not record changes.