An agent can delete customer accounts. What is the right oversight?
Why B is the answer
Require human approval before the action runs
Irreversible, high-impact actions need a person in the loop before they execute.
In detail
- Deleting an account cannot be undone, so a person must approve it.
- Reversible actions need less oversight.
Use case
An illustrative scenario, not a real incident.
An agent proposes deleting an account, and a person must approve first.
See it in different ways
sequenceDiagram participant G as Agent participant H as Human participant S as System G->>H: Proposed action and reason H-->>G: Approve or reject G->>S: Run if approved
sequenceDiagram
participant A as Agent
participant H as Reviewer
participant S as System
A->>H: Action, reason and likely effect
alt Approved
H->>S: Run
else Rejected
H-->>A: Stop
endflowchart TD
Q{"Can it be undone?"}
Q -->|"No"| A0["Require human approval"]
Q -->|"Yes"| A1["Lighter checks"]Key point: Irreversible actions need human approval.
Why the other options are wrong
- A. Allow the action only during business hours when staff are around
- Business hours do not add review.
- C. Automatic execution, with a log that is reviewed each week
- A weekly log is too late.
- D. A clear warning in the system prompt about deleting accounts
- A warning cannot guarantee anything.