Question 540
A configuration review considers this situation: A proposed AI agent would autonomously approve high-value refunds even though policy requires a human finance signoff. Which shortcut would create the most directly relevant risk?
Explanation
Correct Answer
Replacing the required finance approval control with unchecked generative output. The test to require is: A large refund request routes to authorized human approval before execution.
Incorrect Answers
Running every caller's query under a global unrestricted system identity. Replacing the required finance approval control with unchecked generative output.
Routing every simple numeric policy decision through unreviewed generative output. Replacing the required finance approval control with unchecked generative output.
Granting Modify All Data to every service employee to enable one agent. It would not address the vulnerability presented here: Replacing the required finance approval control with unchecked generative output.