Microsoft AI-102 (Designing and Implementing a Microsoft Azure AI Solution) Practice Test 25

Question 240

The review pass doubles latency on every request, even trivial greetings. What optimization fits?

  1. Evaluate with held-out queries and documents that represent production retrieval failures
  2. Retrieve the current authoritative price catalog at inference time
  3. Break down quality metrics by relevant subgroups and inspect high-risk failures
  4. Trigger review selectively for outputs whose risk or complexity warrants its cost

Explanation

Correct Answer

D:

Conditional verification can control latency and spend while retaining checks on high-impact content. In the deployed enterprise chat and retrieval product, confirm the fix by retesting the affected behavior after triggering review selectively for outputs whose risk or complexity warrants its cost.

Incorrect Answers

A:

Realistic independent examples better measure generalization and avoid inflated retrieval metrics in its own workflow, but that is a different correction from the present requirement to trigger review selectively for outputs whose risk or complexity warrants its cost. The reported issue would remain.

B:

That change addresses another failure mode: parameter updates are inefficient and unreliable for frequently changing source facts compared with live grounding; it does not address this case, for which the relevant intervention is to trigger review selectively for outputs whose risk or complexity warrants its cost.

C:

Although useful elsewhere in the deployed enterprise chat and retrieval product, averages conceal concentrated error patterns that matter for production suitability; the present case is different and calls for the team to trigger review selectively for outputs whose risk or complexity warrants its cost.

Question 241

Session data from one customer appears in another customer's answer. Which design fault must be addressed?

  1. Apply tenant or document-level access restrictions when retrieving source files
  2. Keep untrusted retrieved content subordinate to the agent's governing instructions
  3. Isolate conversation state using authenticated user and session identifiers
  4. Authorize the application-side function executor to call the external service

Explanation

Correct Answer

C:

A shared conversation store without proper partitioning risks cross-user information leakage. In the Foundry customer-service agent, confirm the fix by retesting the affected behavior after isolating conversation state using authenticated user and session identifiers.

Incorrect Answers

A:

Although useful elsewhere in the Foundry customer-service agent, a model cannot reliably protect data if an upstream search tool supplies unauthorized content; the present case is different and calls for the team to isolate conversation state using authenticated user and session identifiers.

B:

Reference content is data, not a higher-priority policy authorizer in its own workflow, but that is a different correction from the present requirement to isolate conversation state using authenticated user and session identifiers. The reported issue would remain.

D:

That change addresses another failure mode: a function schema declares an action; it does not supply network access or external service authorization; it does not address this case, for which the relevant intervention is to isolate conversation state using authenticated user and session identifiers.

Question 242

A user approves an agent to refund $30, but the agent's next tool request attempts to refund $300. Which TWO controls prevent this approval from being misused? (Choose TWO.)

  1. Bind the approval to the exact operation and approved amount
  2. Authorize the application-side function executor to call the external service
  3. Require renewed approval if the tool arguments differ materially from the authorized payload
  4. Apply tenant or document-level access restrictions when retrieving source files
  5. Isolate conversation state using authenticated user and session identifiers

Explanation

Correct Answers

A:

Human consent is for a specific action payload; it must not be treated as a general authorization for related financial transactions.

C:

Reconfirmation stops a later agent revision from turning an originally acceptable refund into a larger unapproved transaction.

Incorrect Answers

B:

That change addresses another failure mode: a function schema declares an action; it does not supply network access or external service authorization; it does not address this case, for which the relevant intervention is to bind the approval to the exact operation and approved amount and require renewed approval if the tool arguments differ materially from the authorized payload.

D:

Although useful elsewhere in the Foundry customer-service agent, a model cannot reliably protect data if an upstream search tool supplies unauthorized content; the present case is different and calls for the team to bind the approval to the exact operation and approved amount and require renewed approval if the tool arguments differ materially from the authorized payload.

E:

A shared conversation store without proper partitioning risks cross-user information leakage in its own workflow, but that is a different correction from the present requirement to bind the approval to the exact operation and approved amount and require renewed approval if the tool arguments differ materially from the authorized payload. The reported issue would remain.

Question 243

A conversation has grown beyond the model's usable context limit. What should the agent do?

  1. Require a fresh valid approval instead of replaying an expired authorization
  2. Summarize or prune nonessential history while retaining verified active task state
  3. Version the agent instructions and regression-test representative tool workflows
  4. Enforce ownership and permission checks before applying the cancellation

Explanation

Correct Answer

B:

Selective history management respects token limits without losing the essential context for the current workflow. In the Foundry customer-service agent, confirm the fix by retesting the affected behavior after summarizing or prune nonessential history while retaining verified active task state.

Incorrect Answers

A:

Although useful elsewhere in the Foundry customer-service agent, time-bounded approval protects against stale consent for actions that may have changed context; the present case is different and calls for the team to summarize or prune nonessential history while retaining verified active task state.

C:

Instruction changes can affect tool selection and interaction outcomes, so controlled comparison is needed in its own workflow, but that is a different correction from the present requirement to summarize or prune nonessential history while retaining verified active task state. The reported issue would remain.

D:

That change addresses another failure mode: the host must authorize side effects independently of model reasoning or user-provided identifiers; it does not address this case, for which the relevant intervention is to summarize or prune nonessential history while retaining verified active task state.

Question 244

An approval request expires while an agent is offline. How should execution proceed?

  1. Version the agent instructions and regression-test representative tool workflows
  2. Require a fresh valid approval instead of replaying an expired authorization
  3. Require references to map to actual file-search results and their identifiers
  4. Enforce ownership and permission checks before applying the cancellation

Explanation

Correct Answer

B:

Time-bounded approval protects against stale consent for actions that may have changed context. In the Foundry customer-service agent, confirm the fix by retesting the affected behavior after requiring a fresh valid approval instead of replaying an expired authorization.

Incorrect Answers

A:

Instruction changes can affect tool selection and interaction outcomes, so controlled comparison is needed in its own workflow, but that is a different correction from the present requirement to require a fresh valid approval instead of replaying an expired authorization. The reported issue would remain.

C:

That change addresses another failure mode: grounded attribution should come from retrieval evidence, not invented model text; it does not address this case, for which the relevant intervention is to require a fresh valid approval instead of replaying an expired authorization.

D:

Although useful elsewhere in the Foundry customer-service agent, the host must authorize side effects independently of model reasoning or user-provided identifiers; the present case is different and calls for the team to require a fresh valid approval instead of replaying an expired authorization.

Question 245

A user asks the agent to cancel another person's order. How should the function executor respond?

  1. Require a fresh valid approval instead of replaying an expired authorization
  2. Summarize or prune nonessential history while retaining verified active task state
  3. Enforce ownership and permission checks before applying the cancellation
  4. Require references to map to actual file-search results and their identifiers

Explanation

Correct Answer

C:

The host must authorize side effects independently of model reasoning or user-provided identifiers. In the Foundry customer-service agent, confirm the fix by retesting the affected behavior after enforcing ownership and permission checks before applying the cancellation.

Incorrect Answers

A:

That change addresses another failure mode: time-bounded approval protects against stale consent for actions that may have changed context; it does not address this case, for which the relevant intervention is to enforce ownership and permission checks before applying the cancellation.

B:

Selective history management respects token limits without losing the essential context for the current workflow in its own workflow, but that is a different correction from the present requirement to enforce ownership and permission checks before applying the cancellation. The reported issue would remain.

D:

Although useful elsewhere in the Foundry customer-service agent, grounded attribution should come from retrieval evidence, not invented model text; the present case is different and calls for the team to enforce ownership and permission checks before applying the cancellation.

Question 246

A cited file title is not in the tool's retrieval results. What should the application check?

  1. Version the agent instructions and regression-test representative tool workflows
  2. Require a fresh valid approval instead of replaying an expired authorization
  3. Enforce ownership and permission checks before applying the cancellation
  4. Require references to map to actual file-search results and their identifiers

Explanation

Correct Answer

D:

Grounded attribution should come from retrieval evidence, not invented model text. In the Foundry customer-service agent, confirm the fix by retesting the affected behavior after requiring references to map to actual file-search results and their identifiers.

Incorrect Answers

A:

Instruction changes can affect tool selection and interaction outcomes, so controlled comparison is needed in its own workflow, but that is a different correction from the present requirement to require references to map to actual file-search results and their identifiers. The reported issue would remain.

B:

Although useful elsewhere in the Foundry customer-service agent, time-bounded approval protects against stale consent for actions that may have changed context; the present case is different and calls for the team to require references to map to actual file-search results and their identifiers.

C:

That change addresses another failure mode: the host must authorize side effects independently of model reasoning or user-provided identifiers; it does not address this case, for which the relevant intervention is to require references to map to actual file-search results and their identifiers.

Question 247

The inventory service requires OAuth and the agent's tool configuration contains no credential. What integration work is needed?

  1. Apply tenant or document-level access restrictions when retrieving source files
  2. Authorize the application-side function executor to call the external service
  3. Require the approved action, arguments, and amount to match the actual executed tool call
  4. Isolate conversation state using authenticated user and session identifiers

Explanation

Correct Answer

B:

A function schema declares an action; it does not supply network access or external service authorization. In the Foundry customer-service agent, confirm the fix by retesting the affected behavior after authorizing the application-side function executor to call the external service.

Incorrect Answers

A:

A model cannot reliably protect data if an upstream search tool supplies unauthorized content in its own workflow, but that is a different correction from the present requirement to authorize the application-side function executor to call the external service. The reported issue would remain.

C:

Although useful elsewhere in the Foundry customer-service agent, approval of one payload does not authorize a materially different transaction; the present case is different and calls for the team to authorize the application-side function executor to call the external service.

D:

That change addresses another failure mode: a shared conversation store without proper partitioning risks cross-user information leakage; it does not address this case, for which the relevant intervention is to authorize the application-side function executor to call the external service.

Question 248

A file-search tool indexes policies from several departments, but users must see only their own division. What must the retrieval layer enforce?

  1. Require the approved action, arguments, and amount to match the actual executed tool call
  2. Isolate conversation state using authenticated user and session identifiers
  3. Authorize the application-side function executor to call the external service
  4. Apply tenant or document-level access restrictions when retrieving source files

Explanation

Correct Answer

D:

A model cannot reliably protect data if an upstream search tool supplies unauthorized content. In the Foundry customer-service agent, confirm the fix by retesting the affected behavior after applying tenant or document-level access restrictions when retrieving source files.

Incorrect Answers

A:

Although useful elsewhere in the Foundry customer-service agent, approval of one payload does not authorize a materially different transaction; the present case is different and calls for the team to apply tenant or document-level access restrictions when retrieving source files.

B:

A shared conversation store without proper partitioning risks cross-user information leakage in its own workflow, but that is a different correction from the present requirement to apply tenant or document-level access restrictions when retrieving source files. The reported issue would remain.

C:

That change addresses another failure mode: a function schema declares an action; it does not supply network access or external service authorization; it does not address this case, for which the relevant intervention is to apply tenant or document-level access restrictions when retrieving source files.

Question 249

An agent produces a `get_order` function call with an invalid order ID type. What should the host enforce?

  1. Persist and pass the appropriate conversation context or managed session state
  2. Refresh the file-search knowledge store with the approved policy and verify retrieval of the current version
  3. Validate generated tool arguments against the declared parameter schema
  4. Pause execution and request explicit approval before invoking the transfer tool

Explanation

Correct Answer

C:

Function-calling structure reduces ambiguity but does not guarantee every proposed value is valid for the backend. In the Foundry customer-service agent, confirm the fix by retesting the affected behavior after validating generated tool arguments against the declared parameter schema.

Incorrect Answers

A:

That change addresses another failure mode: multi-turn follow-ups need relevant prior messages or state; separate independent calls do not remember context automatically; it does not address this case, for which the relevant intervention is to validate generated tool arguments against the declared parameter schema.

B:

Although useful elsewhere in the Foundry customer-service agent, attaching a tool is not sufficient when its backing content is stale; updating the approved knowledge source and confirming retrieval distinguishes current policy evidence from archived material; the present case is different and calls for the team to validate generated tool arguments against the declared parameter schema.

D:

Human approval gates high-impact side effects; an agent's plan is not sufficient authorization in its own workflow, but that is a different correction from the present requirement to validate generated tool arguments against the declared parameter schema. The reported issue would remain.

Question 250

A team responsible for a Foundry customer-service agent needs to preserve relevant context across multiple turns of the same support interaction; the acceptance check is that a follow-up pronoun can be resolved using prior conversation context. Which capability should be selected?

  1. Require approval before a sensitive tool action
  2. Register an application function tool
  3. Configure the agent's instructions
  4. Maintain conversation state or a session

Explanation

Correct Answer

D:

Agent session or conversation management connects messages so later turns can use earlier user constraints. For this Foundry customer-service agent, the intended evidence is that a follow-up pronoun can be resolved using prior conversation context. State alone does not guarantee long-term factual memory or authorize external tools.

Incorrect Answers

A:

An approval checkpoint lets the workflow inspect intended tool arguments and obtain human authorization before acting; it may be used to block account changes until an authorized reviewer approves the proposed operation in a Foundry customer-service agent, but the current requirement is instead to preserve relevant context across multiple turns of the same support interaction. That separate capability does not prove that a follow-up pronoun can be resolved using prior conversation context; Recording model confidence is not an approval or an access-control decision, so selecting that alternative would not meet the stated acceptance check.

B:

A function tool exposes a named operation and argument schema; the application or runtime executes permitted calls and returns results; it may be used to let the agent request an approved order-status lookup in an existing business API in a Foundry customer-service agent, but the current requirement is instead to preserve relevant context across multiple turns of the same support interaction. That separate capability does not prove that a follow-up pronoun can be resolved using prior conversation context; An instruction to imagine a database lookup is not an authenticated API invocation, so selecting that alternative would not meet the stated acceptance check.

C:

Agent instructions establish task-level behavior that a model follows while working with supplied context and tools; it may be used to define the agent's scope, task priorities, and rules for interacting with users in a Foundry customer-service agent, but the current requirement is instead to preserve relevant context across multiple turns of the same support interaction. That separate capability does not prove that a follow-up pronoun can be resolved using prior conversation context; Instructions alone cannot run an external transaction or override the executing application's permissions, so selecting that alternative would not meet the stated acceptance check.

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