Question

Your team is deploying an Azure AI Foundry agent that can call a tool to issue refunds through an internal payments API. Compliance requires that any refund above $500 must be reviewed by a human before it executes, while smaller refunds can proceed automatically. You need to configure the agent so that qualifying tool calls pause for explicit human sign-off before invocation. Which approach should you implement?

Accepted Answer

Configure a human-in-the-loop approval workflow (oversight mode) that requires manual approval for the refund tool invocation when the amount exceeds the threshold

Answer

Enable a stricter content safety filter with a higher severity threshold on the agent's model deployment

Answer

Increase the groundedness evaluator sensitivity so low-confidence refund responses are blocked

Answer

Restrict the agent by assigning it a role that removes access to the payments API entirely