Expert level quiz

Question 1 of 7The online shop support agent must keep inference data in one specific Azure region. Which deployment data-processing scope should the team choose?
Global, with cross-region Azure-managed routingData zone, within a defined boundary such as the US or EURegional, limited to a single Azure regionA co-located Foundry resource with multiple projects only
Question 2 of 7Production code still sends the catalog model name on each request and calls fail. What is the correct identifier to use in SDK calls?
The deployment name created when the model was deployedThe display name shown in the Foundry portal catalog onlyThe subscription ID paired with the resource group nameThe model-router SKU name without deploying model-router
Question 3 of 7A support agent was published for production and must reach Azure Storage with its own audit trail. Roles were granted to the shared project identity during experiments. What should the team do now?
Keep the shared project identity; publishing does not change which identity runs tool callsAssign the same Storage roles to the published agent's distinct agent identityUse creator identity on a recurring routine so Storage inherits the developer's accessMove all agents back to unpublished mode to reuse the shared identity
Question 4 of 7Several teams need the same MCP tools, shared credentials on project connections, and tool search for a growing catalog. What pattern fits best?
Attach every MCP server separately to each prompt agent without a toolboxPut function calling tools inside a toolbox so all agents share one endpointStore API keys in agent instructions and rotate them per agent versionBuild a toolbox version, verify it, promote the default, and point agents at the consumer MCP endpoint
Question 5 of 7Engineers ran the shop support agent several times but see no traces in the portal. Tracing was never configured after the project was created. What is the best first step?
Enable client-side OpenTelemetry in the storefront app before any portal changeTurn on content recording for every span in productionConnect an Application Insights resource to the Foundry projectAssign Subscription Contributor so builders can open the Traces tab
Question 6 of 7Red teaming and evaluations flagged jailbreak-style user prompts and indirect injection hidden in Azure AI Search snippets returned to the shop support agent. What is the sensible next production step?
Ship the agent and rely only on the model deployment default guardrail without an agent assignmentCreate a guardrail on the agent with User prompt attacks at user input and Indirect attack at tool response, using Annotate and blockTrack Attack Success Rate (ASR) but wait until live users report issues before changing guardrailsAdd only output harm filters on the model deployment and skip tool response scanning
Question 7 of 7The platform team wants multiregion availability for production agents. What does Microsoft Foundry documentation say about automatic disaster recovery?
Foundry fails over agent runtimes to a paired region without application changesModel Router deployments replicate state across regions automaticallyFoundry does not provide automatic failover or disaster recovery; you plan regions, routing, and backupsEnabling Foundry Control Plane Operate replaces the need for secondary projects

Answer every question to check your answers.