McAfee Secure

Exam Code: AI-300

Exam Name: Operationalizing Machine Learning and Generative AI Solutions

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Machine Learning Operations Engineer Associate

Microsoft AI-300 Questions & Answers

Study with Up-To-Date REAL Exam Questions and Answers from the ACTUAL Test

213 Questions & Answers with Testing Engine

"Operationalizing Machine Learning and Generative AI Solutions Exam", also known as AI-300 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-300 Exam Engine. Your AI-300 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-300 Practice Test 2

Question 10

A shared Azure Machine Learning workspace has multiple dependent Azure resources but no accountable operational steward. The team must establish ownership and limit who can administer those dependencies. What is the best response?

  1. Enable experiment metrics collection without recording any resource owner or administrator.
  2. Apply only a generic cost tag to the workspace while leaving the dependent-resource ownership and access assignments unresolved.
  3. Assign an accountable steward in managed resource metadata and grant approved scoped Azure roles for the workspace and dependencies.
  4. Grant all development users subscription Owner so one of them can respond to incidents.

Explanation

Correct Answer

C:

Ownership metadata identifies an operational contact while least-privilege RBAC constrains administration of the separate workspace and supporting resources. Both are needed for accountable governance.

Incorrect Answers

A:

Model metrics help evaluate runs but do not establish who is responsible for the workspace’s linked storage, vault, registry, or management permissions.

B:

A cost tag can support chargeback but cannot by itself give a steward responsibility or constrain who can change related resources.

D:

Broad subscription Owner rights violate the least-privilege requirement and still do not identify an accountable resource steward for each dependency.

Question 11

A team is migrating a training workflow to a newly created Azure Machine Learning workspace. The old workspace’s internal storage endpoints must not be assumed valid in the new resource. What must be done for the migrated jobs to use the intended dependencies?

  1. Reuse every old workspace datastore URI without checking which storage accounts and compute targets the new workspace references.
  2. Provision or link the approved backing resources for the new workspace, explicitly rebind required data and compute references, then run a smoke test.
  3. Change only the old workspace’s display name and assume that its dependent resource IDs transfer automatically.
  4. Copy the source resource-group tags to the new workspace without revisiting the training data connections.

Explanation

Correct Answer

B:

The new workspace has its own resource identity and associated dependencies. Explicitly resolving its storage, datastore, asset, and compute links establishes a valid migration target.

Incorrect Answers

A:

Workspace identity and associated storage settings can differ across environments. Blindly reusing internal paths risks reading stale data or failing authentication.

C:

Changing a name is not an approved asset migration mechanism and cannot make a newly created workspace inherit all separate resource IDs from the old one.

D:

Tags are governance metadata, not credentials or bindings for data and compute. This leaves the new job’s runtime paths unresolved.

Question 12

A compliance service evaluates Azure Policy against a specific Azure Machine Learning workspace. It needs the exact ARM resource identifier rather than a portal hyperlink or friendly label. Which value should the service retrieve?

  1. The portal navigation URL copied from the workspace overview page.
  2. Only the short workspace name without its subscription, resource group or provider path.
  3. The project team’s informal display label from its tracking spreadsheet.
  4. The full workspace ARM resource ID containing subscription, resource group, `Microsoft.MachineLearningServices/workspaces`, and workspace name.

Explanation

Correct Answer

D:

This structured ARM path uniquely addresses the actual resource at its Azure scope, making it suitable for policy, inventory, and governance records.

Incorrect Answers

A:

A portal deep link is a user-interface route and is not the canonical Azure Resource Manager identity used for resource-scoped policy evaluation.

B:

Workspace names alone do not describe the entire Azure resource scope; compliance policy needs a unique resource ID.

C:

A project label does not identify the Azure subscription, resource group and resource provider that own the deployed workspace.

Question 13

A production Azure Machine Learning v2 job must read a team-approved Blob container by a reusable datastore name. The job already uses a managed identity with Storage Blob Data Reader access, but security prohibits account keys in notebooks or YAML. Which configuration satisfies the required named connection and passwordless access?

  1. Grant the interactive developer subscription Contributor access and keep the storage key in the notebook.
  2. Register the approved container as an Azure ML datastore and configure the job to use its identity-based datastore URI.
  3. Configure a public inbound endpoint for the storage account and continue using the notebook account key.
  4. Assign another Storage Blob Data Reader grant to the current managed identity but keep a raw Blob URI instead of creating a datastore.

Explanation

Correct Answer

B:

A datastore supplies a reusable name for the Blob location; the existing compute managed identity can authenticate to the data plane without a stored storage key. A test job should resolve the named datastore and read a permitted blob.

Incorrect Answers

A:

Subscription Contributor is a control-plane role, does not itself supply a safe named data connection, and does not justify storing an account key in source.

C:

Opening public network exposure does not meet the passwordless requirement, and there is no indication that storage routing—not credentials or naming—is the problem.

D:

The identity already has the needed Blob data role. A redundant grant alone cannot create the reusable datastore name explicitly required in the question.

Question 14

A training job can authenticate to Azure Storage and list its storage account, but a registered Azure ML datastore specifies the nonexistent container `trainng-2026`. The correct container is `training-2026`; Blob requests return ContainerNotFound. What is the most targeted repair?

  1. Grant Storage Blob Data Owner at the subscription root.
  2. Correct the datastore connection to `training-2026`, then read a known blob using the job identity.
  3. Rotate the storage account key before every retry.
  4. Increase request retries while leaving the datastore definition unchanged.

Explanation

Correct Answer

B:

A datastore resolves the storage location named in its connection. Correcting that location removes the 404 ContainerNotFound cause; a read probe confirms both the intended path and existing access.

Incorrect Answers

A:

The documented failure is a missing container name, not insufficient access. A subscription-wide data role would expand privilege yet still send requests to `trainng-2026`.

C:

The job is already authenticating, and a new key cannot create the misspelled container. Rotating credentials here adds disruption without correcting the datastore target.

D:

A deterministic 404 for a nonexistent container does not become a valid path through retries. The reference itself must identify the existing container.

Question 15

An Azure ML evaluation job reads a Blob input successfully using its compute managed identity. It then tries to upload a predictions file to the same approved container, and Storage logs report insufficient *write* permission. IAM shows only Storage Blob Data Reader for that identity. What is the least-privilege fix?

  1. Assign the Azure Machine Learning workspace Reader role to the job identity.
  2. Give the job identity a scoped role that includes Blob write, such as Storage Blob Data Contributor on the approved output container.
  3. Re-register the input as `uri_folder` instead of a `uri_file`.
  4. Turn on anonymous public Blob access to bypass the identity error.

Explanation

Correct Answer

B:

Blob Data Reader permits download but not Put Blob. Container-scoped Data Contributor supplies the missing write data action without subscription-wide administration.

Incorrect Answers

A:

Workspace Reader controls access to AML management metadata, not the Blob write data action required by the failed upload.

C:

Input type controls how the training job receives data, but it cannot grant a Blob write data action to the identity attempting to upload an output.

D:

Public anonymous read does not authorize the protected write operation. It would also violate the documented least-privilege identity approach.

Question 16

A scheduled AML command job still includes a shared storage key literal in its checked-in YAML. The pipeline owner requires passwordless execution and an auditable identity per job. The compute cluster supports managed identity and the approved storage path is known. What should be changed?

  1. Move the storage key into a second YAML file committed to the same repository.
  2. Give every developer Owner permissions on the storage account so the key is unnecessary.
  3. Enable anonymous Blob access and delete the key from YAML.
  4. Remove the committed credential, revoke or rotate the exposed key, and run against a credential-free datastore using an approved managed identity with scoped storage data permissions.

Explanation

Correct Answer

D:

The job can use the compute managed identity for data access, while the datastore locates the target without embedding a key. Revoking the previously exposed key closes the existing credential leak.

Incorrect Answers

A:

Changing which tracked file holds the shared secret does not remove credential exposure or provide a per-job identity.

B:

Control-plane Owner is excessive and is not a substitute for configuring the job execution identity and its specific storage data permissions.

C:

Anonymous access may eliminate a credential in code but it removes the intended access boundary; it does not provide the required auditable workload identity.

Question 17

An AML scoring component reads Parquet files through an ADLS Gen2 datastore. The execution identity has access to the correct storage account, but the URI names filesystem `archive` and directory `/2025` while the approved partition is in filesystem `curated` under `/2026`. What should the engineer change?

  1. Raise GPU quota because a missing Parquet file is a compute allocation problem.
  2. Grant the identity a workspace-level Contributor role without changing the URI.
  3. Declare the source as a new MLflow model asset.
  4. Update the datastore/data URI to filesystem `curated` and directory `/2026`, then list and read a known Parquet file.

Explanation

Correct Answer

D:

ADLS Gen2 paths are tied to a specified filesystem and directory. The corrected URI targets the approved partition and a read probe verifies that it resolves as intended.

Incorrect Answers

A:

The job has passed compute allocation and is looking in the wrong filesystem. GPU capacity cannot change an ADLS Gen2 path.

B:

An AML management role does not redirect a datastore path from `archive/2025` to `curated/2026`; the location error persists.

C:

An MLflow model describes a trained artifact; it is not a method for correcting a filesystem and directory reference to training data.

Question 18

Workspace `aml-dev` and workspace `aml-prod` each register a datastore named `raw`. A deployment script assumes that name maps to the same storage account everywhere; inspection shows dev points to `raw-dev` while prod must use `raw-prod`. Which control prevents the wrong source from silently being used?

  1. Rename both workspaces `raw` so their datastore names match.
  2. Grant the application Owner in both subscriptions and keep the hard-coded datastore name.
  3. Inspect and version the datastore definitions per workspace, assert the expected storage account in deployment tests, and bind jobs to the intended environment configuration.
  4. Change every job from `download` to `ro_mount` but do not inspect datastore metadata.

Explanation

Correct Answer

C:

Datastore names are workspace-scoped. An environment-specific mapping and effective-account validation prevent identical logical names from resolving to the wrong physical data source.

Incorrect Answers

A:

Workspace names are separate resource identifiers; a matching workspace label does not synchronize datastore connection definitions.

B:

Broader privileges do not prove which storage account `raw` resolves to in each workspace and increase cross-environment exposure.

D:

Mount and download are data transfer modes. Both would still consume whichever storage account the datastore name actually references.

Question 19

An Azure ML v2 command job succeeds when submitted interactively but fails in unattended automation when reading a private Blob. The YAML specifies `identity: {type: user_identity}`; the developer has Blob read access, but the unattended execution identity should be the compute cluster managed identity. What is the correct job configuration?

  1. Set the job step to use `identity: {type: managed_identity}` and ensure that compute identity has scoped Blob data-read permission.
  2. Leave `user_identity` in YAML and grant the submitting developer subscription Owner.
  3. Set the input mode to `direct` and expect AML to mount the file using managed identity automatically.
  4. Store the developer account key in a datastore and omit the job identity.

Explanation

Correct Answer

A:

The YAML `user_identity` mode passes through the submitting user. Managed-identity mode makes the compute identity the data-access principal for unattended jobs, so authorization can be scoped and tested independently of a developer login.

Incorrect Answers

B:

The workflow still depends on an interactive human identity, and Owner is unnecessarily broad. It does not establish the intended unattended compute principal.

C:

Direct mode passes a storage URL for the code to handle. It does not itself mount or download data with the desired managed identity.

D:

Credential caching may permit a read but violates the explicit unattended identity requirement and bypasses the intended managed-identity audit boundary.

Question 20

An AML compute subnet connects privately to a hierarchical-namespace Azure Storage account. Blob reads succeed over `account.blob.core.windows.net`, but an ADLS Gen2 operation that creates a directory fails because `account.dfs.core.windows.net` resolves publicly. Public network access is disabled. What network configuration fixes this *DFS-specific* failure?

  1. Create/verify the DFS private endpoint and the `privatelink.dfs.core.windows.net` DNS resolution from compute, retaining the Blob private endpoint.
  2. Remove the Blob private endpoint and use only the storage account public hostname.
  3. Assign Storage Blob Data Owner to all workspace users.
  4. Edit an Azure ML environment Conda file to include the latest Python.

Explanation

Correct Answer

A:

Data Lake `dfs` APIs use a separate storage subresource from Blob APIs. A Blob private endpoint cannot serve every DFS operation; the DFS endpoint and private DNS must be available from compute.

Incorrect Answers

B:

The Blob path already works and public access is deliberately blocked. Removing private connectivity worsens access and does not provide a DFS private endpoint.

C:

The observed failure is name resolution and routing to DFS, not an ACL or Azure RBAC denial. More data rights cannot make the DFS hostname resolve privately.

D:

Application dependencies do not create a DNS zone or a Private Link endpoint for the DFS service subresource.

Operationalizing Machine Learning and Generative AI Solutions (AI-300) Practice Tests