Microsoft AI-102 (Designing and Implementing a Microsoft Azure AI Solution) Practice Test 9

Question 80

A release for a production inference rollout is blocked until the application can execute offline inference on a large queued dataset rather than respond to each request interactively. The tester will verify that submitted work finishes later with retrievable batch results. What should be implemented?

  1. Use a batch-processing deployment or API
  2. Select a standard online model deployment
  3. Fine-tune a supported model before deployment
  4. Deploy a supported Azure AI container

Explanation

Correct Answer

A:

Batch inference is designed for asynchronous processing of submitted input collections when the selected model and API support it. For this production inference rollout, the intended evidence is that submitted work finishes later with retrievable batch results. It is unsuitable for a synchronous low-latency user chat turn.

Incorrect Answers

B:

Standard online deployments are used for request-driven inference within supported model and service limits; it may be used to serve ordinary interactive inference traffic without a pre-purchased dedicated capacity reservation in a production inference rollout, but the current requirement is instead to execute offline inference on a large queued dataset rather than respond to each request interactively. That separate capability does not prove that submitted work finishes later with retrievable batch results; They do not promise reserved throughput when a workload requires predictable provisioned capacity, so selecting that alternative would not meet the stated acceptance check.

C:

Fine-tuning modifies model behavior through a supported training process and produces an adapted model artifact; it may be used to adapt a pretrained model to a specialized labeled example set where that model supports tuning in a production inference rollout, but the current requirement is instead to execute offline inference on a large queued dataset rather than respond to each request interactively. That separate capability does not prove that submitted work finishes later with retrievable batch results; It is not a substitute for deploying a model or retrieving fresh private documents at query time, so selecting that alternative would not meet the stated acceptance check.

D:

Containers can bring supported Azure AI inference features to an environment that meets their licensing and connectivity conditions; it may be used to run a supported AI capability close to data in a controlled container host in a production inference rollout, but the current requirement is instead to execute offline inference on a large queued dataset rather than respond to each request interactively. That separate capability does not prove that submitted work finishes later with retrievable batch results; Not every model or Foundry service is available as a container; compatibility must be checked, so selecting that alternative would not meet the stated acceptance check.

Question 81

A private endpoint works from one virtual network but not from a peered network. What should the team investigate?

  1. Scope the role assignment to that individual resource where feasible
  2. Grant the caller permission to read the secret using the vault's chosen authorization model
  3. Check peering, routes, private DNS reachability, and applicable network policies
  4. Acquire short-lived tokens at runtime through the managed-identity credential chain

Explanation

Correct Answer

C:

Successful private connectivity depends on actual network paths and name resolution from each client network. In the regulated Azure AI application, confirm the fix by retesting the affected behavior after checking peering, routes, private DNS reachability, and applicable network policies.

Incorrect Answers

A:

Although useful elsewhere in the regulated Azure AI application, rBAC inheritance makes broad assignments effective on multiple resources; narrower scope reduces unnecessary exposure; the present case is different and calls for the team to check peering, routes, private DNS reachability, and applicable network policies.

B:

Knowing a secret's URI does not grant access; RBAC or vault access policy controls the read operation in its own workflow, but that is a different correction from the present requirement to check peering, routes, private DNS reachability, and applicable network policies. The reported issue would remain.

D:

That change addresses another failure mode: access tokens expire and are not durable application secrets; supported credentials renew them when needed; it does not address this case, for which the relevant intervention is to check peering, routes, private DNS reachability, and applicable network policies.

Question 82

An Azure Function currently embeds an AI service key in its configuration. The service supports Entra authorization. How should it authenticate without shared secrets?

  1. Enable the Function's managed identity and request a resource-scoped Entra token
  2. Assign the least-privileged supported data-plane inference role at the narrowest useful scope
  3. Switch clients to the unused key, validate traffic, then regenerate the old key
  4. Store the service key in Key Vault and retrieve it using a controlled application identity

Explanation

Correct Answer

A:

Managed identity avoids a long-lived shared key, but the identity still needs the correct resource permissions. In the regulated Azure AI application, confirm the fix by retesting the affected behavior after enabling the Function's managed identity and request a resource-scoped Entra token.

Incorrect Answers

B:

That change addresses another failure mode: administrative Contributor permissions are broader than inference access; role scope and allowed actions should match the runtime task; it does not address this case, for which the relevant intervention is to enable the Function's managed identity and request a resource-scoped Entra token.

C:

Dual-key rotation limits downtime because applications can continue using a valid key while the compromised key is revoked in its own workflow, but that is a different correction from the present requirement to enable the Function's managed identity and request a resource-scoped Entra token. The reported issue would remain.

D:

Although useful elsewhere in the regulated Azure AI application, key Vault separates secret storage from source code and supports access policies and auditing; the present case is different and calls for the team to enable the Function's managed identity and request a resource-scoped Entra token.

Question 83

In a regulated Azure AI application, engineers must protect and rotate service credentials away from application source code. The expected evidence is that application code loads a controlled secret from the vault. Which implementation is the best match?

  1. Store API keys in Azure Key Vault
  2. Use Microsoft Entra managed-identity authentication
  3. Assign appropriate Azure RBAC roles
  4. Apply key rotation and credential rollback procedures

Explanation

Correct Answer

A:

Key Vault centralizes secret storage and access policies or RBAC for clients allowed to retrieve credentials. For this regulated Azure AI application, the intended evidence is that application code loads a controlled secret from the vault. Secret storage does not convert an API-key authorization flow into Entra authorization.

Incorrect Answers

B:

A managed identity obtains Entra tokens for a workload and requires appropriate resource RBAC permissions to authorize access; it may be used to allow Azure-hosted code to call a compatible AI resource without storing a shared API key in a regulated Azure AI application, but the current requirement is instead to protect and rotate service credentials away from application source code. That separate capability does not prove that application code loads a controlled secret from the vault; Identity by itself does not remove public network exposure or grant every data action, so selecting that alternative would not meet the stated acceptance check.

C:

Azure RBAC assignments combine a principal, role definition, and scope to authorize supported management or data operations; it may be used to grant a principal the minimum resource permissions required for its AI operations in a regulated Azure AI application, but the current requirement is instead to protect and rotate service credentials away from application source code. That separate capability does not prove that application code loads a controlled secret from the vault; Assigning a role does not establish an encrypted private network path by itself, so selecting that alternative would not meet the stated acceptance check.

D:

Coordinated key rotation updates consumers and changes active credentials, commonly using alternate keys where supported; it may be used to replace compromised shared service keys without prolonged application interruption in a regulated Azure AI application, but the current requirement is instead to protect and rotate service credentials away from application source code. That separate capability does not prove that application code loads a controlled secret from the vault; It does not fix a missing model deployment or an unauthorized Entra principal, so selecting that alternative would not meet the stated acceptance check.

Question 84

A regulated Azure AI application must grant a principal the minimum resource permissions required for its AI operations. Reviewers will accept the implementation when the calling identity can perform only the authorized AI resource operations. Which approach best fulfills that requirement?

  1. Assign appropriate Azure RBAC roles
  2. Configure a private endpoint
  3. Apply key rotation and credential rollback procedures
  4. Use Microsoft Entra managed-identity authentication

Explanation

Correct Answer

A:

Azure RBAC assignments combine a principal, role definition, and scope to authorize supported management or data operations. For this regulated Azure AI application, the intended evidence is that the calling identity can perform only the authorized AI resource operations. Assigning a role does not establish an encrypted private network path by itself.

Incorrect Answers

B:

Private endpoints map supported service access to a private IP in a VNet, alongside required DNS and network controls; it may be used to provide private-network connectivity to an AI resource from an approved virtual network in a regulated Azure AI application, but the current requirement is instead to grant a principal the minimum resource permissions required for its AI operations. That separate capability does not prove that the calling identity can perform only the authorized AI resource operations; Private Link does not replace authentication or decide a user's RBAC privileges, so selecting that alternative would not meet the stated acceptance check.

C:

Coordinated key rotation updates consumers and changes active credentials, commonly using alternate keys where supported; it may be used to replace compromised shared service keys without prolonged application interruption in a regulated Azure AI application, but the current requirement is instead to grant a principal the minimum resource permissions required for its AI operations. That separate capability does not prove that the calling identity can perform only the authorized AI resource operations; It does not fix a missing model deployment or an unauthorized Entra principal, so selecting that alternative would not meet the stated acceptance check.

D:

A managed identity obtains Entra tokens for a workload and requires appropriate resource RBAC permissions to authorize access; it may be used to allow Azure-hosted code to call a compatible AI resource without storing a shared API key in a regulated Azure AI application, but the current requirement is instead to grant a principal the minimum resource permissions required for its AI operations. That separate capability does not prove that the calling identity can perform only the authorized AI resource operations; Identity by itself does not remove public network exposure or grant every data action, so selecting that alternative would not meet the stated acceptance check.

Question 85

A principal has an Azure management Reader role yet receives data-plane authorization failures. What explains the mismatch?

  1. Redact secret values from application and deployment logs
  2. Restrict public network access according to the resource's supported networking controls
  3. Inspect the deployed application's cached key or secret reference and refresh it
  4. Use a role that grants the required AI service data actions, not only management read access

Explanation

Correct Answer

D:

Management-plane visibility and data-plane inference permissions are separate authorization concerns. In the regulated Azure AI application, confirm the fix by retesting the affected behavior after using a role that grants the required AI service data actions, not only management read access.

Incorrect Answers

A:

Secure storage is undermined if retrieved credentials are written into accessible telemetry in its own workflow, but that is a different correction from the present requirement to use a role that grants the required AI service data actions, not only management read access. The reported issue would remain.

B:

Although useful elsewhere in the regulated Azure AI application, creating a private endpoint does not necessarily disable the public access path; both must be governed deliberately; the present case is different and calls for the team to use a role that grants the required AI service data actions, not only management read access.

C:

That change addresses another failure mode: rotating the resource credential is insufficient when the client still sends the prior revoked value; it does not address this case, for which the relevant intervention is to use a role that grants the required AI service data actions, not only management read access.

Question 86

A release for a regulated Azure AI application is blocked until the application can allow Azure-hosted code to call a compatible AI resource without storing a shared API key. The tester will verify that outbound requests use short-lived identity tokens instead of embedded keys. What should be implemented?

  1. Store API keys in Azure Key Vault
  2. Configure a private endpoint
  3. Use Microsoft Entra managed-identity authentication
  4. Apply key rotation and credential rollback procedures

Explanation

Correct Answer

C:

A managed identity obtains Entra tokens for a workload and requires appropriate resource RBAC permissions to authorize access. For this regulated Azure AI application, the intended evidence is that outbound requests use short-lived identity tokens instead of embedded keys. Identity by itself does not remove public network exposure or grant every data action.

Incorrect Answers

A:

Key Vault centralizes secret storage and access policies or RBAC for clients allowed to retrieve credentials; it may be used to protect and rotate service credentials away from application source code in a regulated Azure AI application, but the current requirement is instead to allow Azure-hosted code to call a compatible AI resource without storing a shared API key. That separate capability does not prove that outbound requests use short-lived identity tokens instead of embedded keys; Secret storage does not convert an API-key authorization flow into Entra authorization, so selecting that alternative would not meet the stated acceptance check.

B:

Private endpoints map supported service access to a private IP in a VNet, alongside required DNS and network controls; it may be used to provide private-network connectivity to an AI resource from an approved virtual network in a regulated Azure AI application, but the current requirement is instead to allow Azure-hosted code to call a compatible AI resource without storing a shared API key. That separate capability does not prove that outbound requests use short-lived identity tokens instead of embedded keys; Private Link does not replace authentication or decide a user's RBAC privileges, so selecting that alternative would not meet the stated acceptance check.

D:

Coordinated key rotation updates consumers and changes active credentials, commonly using alternate keys where supported; it may be used to replace compromised shared service keys without prolonged application interruption in a regulated Azure AI application, but the current requirement is instead to allow Azure-hosted code to call a compatible AI resource without storing a shared API key. That separate capability does not prove that outbound requests use short-lived identity tokens instead of embedded keys; It does not fix a missing model deployment or an unauthorized Entra principal, so selecting that alternative would not meet the stated acceptance check.

Question 87

A developer copies an Entra access token into a secret vault as a permanent credential. What should change?

  1. Scope the role assignment to that individual resource where feasible
  2. Acquire short-lived tokens at runtime through the managed-identity credential chain
  3. Keep a verified alternate credential and audit cutover before revoking the exposed key
  4. Grant the caller permission to read the secret using the vault's chosen authorization model

Explanation

Correct Answer

B:

Access tokens expire and are not durable application secrets; supported credentials renew them when needed. In the regulated Azure AI application, confirm the fix by retesting the affected behavior after acquiring short-lived tokens at runtime through the managed-identity credential chain.

Incorrect Answers

A:

Although useful elsewhere in the regulated Azure AI application, rBAC inheritance makes broad assignments effective on multiple resources; narrower scope reduces unnecessary exposure; the present case is different and calls for the team to acquire short-lived tokens at runtime through the managed-identity credential chain.

C:

Rollback should restore authorized connectivity through a known safe credential, not re-enable a compromised secret in its own workflow, but that is a different correction from the present requirement to acquire short-lived tokens at runtime through the managed-identity credential chain. The reported issue would remain.

D:

That change addresses another failure mode: knowing a secret's URI does not grant access; RBAC or vault access policy controls the read operation; it does not address this case, for which the relevant intervention is to acquire short-lived tokens at runtime through the managed-identity credential chain.

Question 88

Users outside a corporate network can still reach the service's public endpoint after a private endpoint was created. Which setting needs evaluation?

  1. Inspect the deployed application's cached key or secret reference and refresh it
  2. Restrict public network access according to the resource's supported networking controls
  3. Check that the identity has the required data-plane role on the AI resource
  4. Redact secret values from application and deployment logs

Explanation

Correct Answer

B:

Creating a private endpoint does not necessarily disable the public access path; both must be governed deliberately. In the regulated Azure AI application, confirm the fix by retesting the affected behavior after restricting public network access according to the resource's supported networking controls.

Incorrect Answers

A:

That change addresses another failure mode: rotating the resource credential is insufficient when the client still sends the prior revoked value; it does not address this case, for which the relevant intervention is to restrict public network access according to the resource's supported networking controls.

C:

Although useful elsewhere in the regulated Azure AI application, obtaining a token proves authentication; authorization can still fail when the principal lacks a suitable role or scope; the present case is different and calls for the team to restrict public network access according to the resource's supported networking controls.

D:

Secure storage is undermined if retrieved credentials are written into accessible telemetry in its own workflow, but that is a different correction from the present requirement to restrict public network access according to the resource's supported networking controls. The reported issue would remain.

Question 89

An Azure-hosted AI resource already has public network access disabled and must be reachable over an approved virtual network; client lookups must resolve the resource to its private address. Which TWO network changes are required? (Choose TWO.)

  1. Switch clients to the unused key, validate traffic, then regenerate the old key
  2. Enable the Function's managed identity and request a resource-scoped Entra token
  3. Store the service key in Key Vault and retrieve it using a controlled application identity
  4. Create a supported private endpoint for the AI resource in the approved network
  5. Configure the associated private DNS resolution for the resource hostname

Explanation

Correct Answers

D:

A private endpoint provides a private-network entry point for the service, so approved clients need not traverse a public destination to reach it.

E:

Correct private DNS mapping ensures clients resolve the standard service name to the intended private endpoint IP rather than the public address.

Incorrect Answers

A:

Dual-key rotation limits downtime because applications can continue using a valid key while the compromised key is revoked in its own workflow, but that is a different correction from the present requirement to create a supported private endpoint for the AI resource in the approved network and configure the associated private DNS resolution for the resource hostname. The reported issue would remain.

B:

Although useful elsewhere in the regulated Azure AI application, managed identity avoids a long-lived shared key, but the identity still needs the correct resource permissions; the present case is different and calls for the team to create a supported private endpoint for the AI resource in the approved network and configure the associated private DNS resolution for the resource hostname.

C:

That change addresses another failure mode: key Vault separates secret storage from source code and supports access policies and auditing; it does not address this case, for which the relevant intervention is to create a supported private endpoint for the AI resource in the approved network and configure the associated private DNS resolution for the resource hostname.

Question 90

A security scan flags resource-group-wide AI role assignments that were only needed for one resource. What should change?

  1. Keep a verified alternate credential and audit cutover before revoking the exposed key
  2. Acquire short-lived tokens at runtime through the managed-identity credential chain
  3. Grant the caller permission to read the secret using the vault's chosen authorization model
  4. Scope the role assignment to that individual resource where feasible

Explanation

Correct Answer

D:

RBAC inheritance makes broad assignments effective on multiple resources; narrower scope reduces unnecessary exposure. In the regulated Azure AI application, confirm the fix by retesting the affected behavior after scoping the role assignment to that individual resource where feasible.

Incorrect Answers

A:

Rollback should restore authorized connectivity through a known safe credential, not re-enable a compromised secret in its own workflow, but that is a different correction from the present requirement to scope the role assignment to that individual resource where feasible. The reported issue would remain.

B:

That change addresses another failure mode: access tokens expire and are not durable application secrets; supported credentials renew them when needed; it does not address this case, for which the relevant intervention is to scope the role assignment to that individual resource where feasible.

C:

Although useful elsewhere in the regulated Azure AI application, knowing a secret's URI does not grant access; RBAC or vault access policy controls the read operation; the present case is different and calls for the team to scope the role assignment to that individual resource where feasible.

Designing and Implementing a Microsoft Azure AI Solution (AI-102) Practice Tests