cert
cert-1
cert-2

Pass Microsoft AI-103 Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers to Pass the Actual Exam!
30 Days Free Updates, Instant Download!

cert-5
cert-6
AI-103 Exam - Verified By Experts
AI-103 Premium File

AI-103 Premium File

$89.99
$98.99
  • Premium File 187 Questions & Answers. Last Update: Nov 15, 2026

Whats Included:

  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates
 
$98.99
$89.99
accept 397 downloads in the last 7 days
block-screenshots
Microsoft AI-103 Practice Test 19

Question 180

A customer asks a model to update an order, but the user is authenticated only for read-only support. What should block the update?

  1. Have the model reinterpret the user's read-only role as write permission.
  2. Execute the update first and check the customer's permissions afterward.
  3. Enforce the caller's server-side tool permission scope before accepting the mutation.
  4. Hide the update confirmation in the final reply and skip backend role checks.

Explanation

Correct Answer

C:

A natural-language request cannot expand the caller's role; the API must reject unsupported side effects even when the model asks.

Incorrect Answers

A:

Model text cannot confer the authorization required by the order-update backend.

B:

Checking permission after executing an order update allows an unauthorized mutation to occur. The tool must validate the authenticated caller's write scope against the target order before performing any side effect.

D:

Hiding the response to an unauthorized order update conceals activity rather than enforcing access control. The backend needs caller-specific authorization checks before modifying the order, independently of the model's message.

Question 181

A test model uses different performance settings than production. How should clients select the correct deployment?

  1. Version the agent tool contract and provide a compatibility migration window
  2. Version model identifiers, deployment settings, tool schemas and prompt artifacts together
  3. Expose separate versioned deployments and promote through controlled environments
  4. Make the deployment identifier an explicit environment configuration value

Explanation

Correct Answer

D:

Model deployment names route requests to intended serving configurations and should not be hard-coded incorrectly. The practical requirement is to route each environment to its own model deployment; a solution built to roll out breaking agent interface changes safely would not meet that same need.

Incorrect Answers

A:

Contract versioning allows clients and services to transition without abruptly invalidating old requests. That is relevant to roll out breaking agent interface changes safely, but the present scenario requires an ability to route each environment to its own model deployment.

B:

Generation behavior depends on more than code; recording AI configuration helps reproducible rollbacks. This helps when a team needs to reproduce the deployed AI configuration; it does not establish how to route each environment to its own model deployment.

C:

Keeping production and candidate deployments separate supports comparison and rollback. The function serves to stage model replacements independently of production, whereas the actual requirement is to route each environment to its own model deployment.

Question 182

A project SDK client succeeds locally but fails in CI because it cannot find tenant credentials. What should be configured?

  1. Suppress missing-credential errors in CI and allow the SDK client to continue without an authenticated identity.
  2. Install the SDK in CI and assume it can reuse a developer's interactive local sign-in without workload credentials.
  3. Provide an explicit supported noninteractive workload-identity credential chain in CI
  4. Normalize connector outputs to a common typed evidence schema

Explanation

Correct Answer

C:

Developer sign-in state is not automatically available to a pipeline; CI needs a supported identity. In this scenario the team must authenticate Foundry SDK within CI. A feature used to remain compatible with optional connector fields does not substitute for it.

Incorrect Answers

A:

The failure occurs before the requested SDK call can be authorized. Configure a supported noninteractive identity and grant it the required scope; ignoring credential-discovery errors cannot authenticate the build job.

B:

Importing a client library does not provision CI identity. Configure a supported noninteractive credential source and the permissions required by the deployed resource.

D:

A typed normalization layer lets downstream prompts and citations treat distinct connectors consistently. This helps when a team needs to harmonize evidence from multiple connectors; it does not establish how to authenticate Foundry SDK within CI.

Question 183

A foundation model lacks enough regional capacity for expected production traffic. Which pre-deployment check prevents surprises?

  1. Reserve additional disk storage while leaving the target model's regional serving quota and deployment availability unchecked.
  2. Expose separate versioned deployments and promote through controlled environments
  3. Define least-privilege per-department service identities and access scopes
  4. Check model availability, supported region and assigned quota before provisioning

Explanation

Correct Answer

D:

Not all models or capacity classes are available in all locations or subscriptions. Here the relevant capability must let the team verify regional model capacity before rollout. An alternative targeting route each environment to its own model deployment solves another problem.

Incorrect Answers

A:

Disk capacity is not model inference throughput. Confirm the selected model's available region, deployment type and quota for the expected workload before release.

B:

Keeping production and candidate deployments separate supports comparison and rollback. The function serves to stage model replacements independently of production, whereas the actual requirement is to verify regional model capacity before rollout.

C:

Shared serving infrastructure does not imply shared authorization to every model, file or tool. That capability is appropriate for separate departmental serving permissions, not this separate task of finding how to verify regional model capacity before rollout.

Question 184

An application opens a new authenticated SDK client for every token of streamed output. What efficiency correction is warranted?

  1. Validate tool-result types and treat returned content as untrusted evidence
  2. Recreate and authenticate the SDK client separately for every incoming stream token.
  3. Open another authenticated SDK client for each stream token so every chunk receives fresh credential and parser initialization.
  4. Reuse appropriately managed clients and connections across requests

Explanation

Correct Answer

D:

Connection/client reuse reduces repeated initialization overhead without weakening authentication. This resolves the specific need to avoid repeated per-token client setup, while a measure that helps enforce typed untrusted tool-result handling addresses a different requirement.

Incorrect Answers

A:

Even a trusted API can return malformed or attacker-controlled content; validation precedes prompting. Its primary purpose is to enforce typed untrusted tool-result handling; the stated constraint instead demands a way to avoid repeated per-token client setup.

B:

Per-token client setup repeats connection and authentication costs without adding information. Reuse an appropriately managed client while consuming events from the existing stream.

C:

Per-token client construction adds connection, credential and allocation overhead. Reuse a properly configured SDK client for the response stream and process incremental events as they arrive.

Question 185

An agent must use a new tool schema, but older mobile clients still send the earlier contract. What deployment strategy reduces breakage?

  1. Keep a tested previous deployment and a controlled traffic-switch procedure
  2. Define least-privilege per-department service identities and access scopes
  3. Version the agent tool contract and provide a compatibility migration window
  4. Check only regional TPM quota and deploy an incompatible new tool schema directly over existing mobile clients.

Explanation

Correct Answer

C:

Contract versioning allows clients and services to transition without abruptly invalidating old requests. This directly supports the need to roll out breaking agent interface changes safely, unlike a mechanism primarily intended to restore serving behavior after a model regression.

Incorrect Answers

A:

A known-good model deployment with a tested switch limits recovery time when regressions appear. The function serves to restore serving behavior after a model regression, whereas the actual requirement is to roll out breaking agent interface changes safely.

B:

Shared serving infrastructure does not imply shared authorization to every model, file or tool. That capability is appropriate for separate departmental serving permissions, not this separate task of finding how to roll out breaking agent interface changes safely.

D:

Quota limits serving capacity, not schema compatibility. Use versioned tool contracts and backward-compatible rollout or explicit client migration.

Question 186

Two agent tools return results concurrently and the final response associates a price with the wrong product. What execution metadata must be preserved?

  1. Assign received results to products based only on callback completion order.
  2. Use the most recent price result for every product in the batch.
  3. Discard tool-call arguments after dispatch to reduce metadata volume.
  4. Correlate each asynchronous result to its original tool-call ID and requested arguments.

Explanation

Correct Answer

D:

Explicit call identifiers prevent mixing results from different concurrent operations and preserve auditability.

Incorrect Answers

A:

Concurrent operations may finish in any order, so positional pairing can attach the wrong price.

B:

One response is not proof of the value for other product identifiers.

C:

Without the initiating product IDs and call IDs the controller cannot safely correlate replies.

Question 187

A product needs reproducible model and prompt configuration between staging and production. What should be recorded?

  1. Save only the human-readable model display name and leave the prompt and tool schema outside release version control.
  2. Keep a tested previous deployment and a controlled traffic-switch procedure
  3. Define least-privilege per-department service identities and access scopes
  4. Version model identifiers, deployment settings, tool schemas and prompt artifacts together

Explanation

Correct Answer

D:

Generation behavior depends on more than code; recording AI configuration helps reproducible rollbacks. The decisive difference is whether the design can reproduce the deployed AI configuration; the competing function serves to route each environment to its own model deployment.

Incorrect Answers

A:

A display name cannot reproduce the exact runtime behavior. Reproducible staging-to-production deployment requires recorded model/version, configuration, prompt, tools and dependency artifacts.

B:

A known-good model deployment with a tested switch limits recovery time when regressions appear. Here the needed outcome is to reproduce the deployed AI configuration; an option designed to restore serving behavior after a model regression would not directly provide it.

C:

Shared serving infrastructure does not imply shared authorization to every model, file or tool. Its primary purpose is to separate departmental serving permissions; the stated constraint instead demands a way to reproduce the deployed AI configuration.

Question 188

A streaming text UI must render partial model output without waiting for the entire completion. Which API usage is appropriate?

  1. Consume a supported streaming inference response and assemble events in order
  2. Buffer every streamed event until completion, then parse the final aggregate message for optional fields.
  3. Use a nonstreaming SDK request and update the UI only after the full completion has been returned.
  4. Reuse appropriately managed clients and connections across requests

Explanation

Correct Answer

A:

Streaming emits incremental content that a client can progressively display while handling completion and errors. The practical requirement is to render model generation incrementally; a solution built to remain compatible with optional connector fields would not meet that same need.

Incorrect Answers

B:

Waiting for completion discards the low-latency benefit of streaming. The UI must iterate or subscribe to incremental response events and render each supported text delta as it arrives.

C:

A nonstreaming completion cannot provide incremental model tokens to the UI. Use a supported streaming API and process response events in arrival order.

D:

Connection/client reuse reduces repeated initialization overhead without weakening authentication. That capability is appropriate for avoid repeated per-token client setup, not this separate task of finding how to render model generation incrementally.

Question 189

A Python service currently stores a Foundry API key in source control. Which credential design removes the static secret?

  1. Keep the API key in an encrypted repository file loaded by the same application.
  2. Put the static key in a container image layer instead of source code.
  3. Use the workload's managed identity with the appropriate Foundry role and supported token-based SDK authentication.
  4. Grant public unauthenticated access to the model endpoint.

Explanation

Correct Answer

C:

Managed identity avoids distributing a persistent key, but the identity must still have the correct project/model invocation authorization.

Incorrect Answers

A:

Encrypting a persistent source-controlled key does not achieve the requested keyless identity architecture.

B:

An image layer may expose the secret and still requires shared-key rotation.

D:

Anonymous access is not a secure alternative to scoped token-based authentication.

Question 190

A tool connector returns an unexpected object type to the agent and it treats the data as trusted instructions. What integration check is needed?

  1. Provide an explicit supported noninteractive workload-identity credential chain in CI
  2. Accept every syntactically valid connector object as new agent instructions if its JSON parser tolerates extra fields.
  3. Treat every connector payload as trusted instructions merely because a supported SDK successfully deserializes it.
  4. Validate tool-result types and treat returned content as untrusted evidence

Explanation

Correct Answer

D:

Even a trusted API can return malformed or attacker-controlled content; validation precedes prompting. Here the relevant capability must let the team enforce typed untrusted tool-result handling. An alternative targeting authenticate Foundry SDK within CI solves another problem.

Incorrect Answers

A:

Developer sign-in state is not automatically available to a pipeline; CI needs a supported identity. This helps when a team needs to authenticate Foundry SDK within CI; it does not establish how to enforce typed untrusted tool-result handling.

B:

Schema acceptance is not an instruction trust boundary. Validate the tool response against its declared contract and keep external content as untrusted data rather than treating its text as agent directives.

C:

Syntactic parsing is not authorization or trust validation. Check expected types, sanitize untrusted content and prevent tool responses from becoming privileged instructions.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests