Microsoft AI-103 Practice Test 22 - Prepaway
Practice Exams:

Pass Microsoft AI-103 Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers!

30 Days Free Updates, Instant Download!

Verified By Experts
Premium File
$89.99 $98.99

AI-103 Premium ETE File


  • Premium File 187 Questions & Answers. Last Update: Nov 15, 2026
Whats Included:
  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates

Microsoft AI-103 Practice Test 22

Question 210

Production requirement: A bank assistant receives malicious instructions embedded in a retrieved customer-support article and must not execute external actions. Which TWO decisions would undermine the required system boundary? Select TWO answers.

  1. Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds
  2. Approve release using only general task accuracy without abuse-case tests
  3. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates
  4. Turn off safety filtering because ordinary users report occasional refusals
  5. Inspect records that explain the affected process, then check the result described in this case: A bank assistant receives malicious instructions embedded in a retrieved customer-support article and must not execute external actions

Explanation

Correct Answer

B:

“Approve release using only general task accuracy without abuse-case tests” would violate the requirement.

D:

“Turn off safety filtering because ordinary users report occasional refusals” would violate the requirement.

Incorrect Answers

A:

“Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds” is the controlled implementation, not an unsafe workaround. Confirm that measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

C:

“Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates” is the required outcome of a working control, not an action that weakens it.

E:

“Inspect records that explain the affected process, then check the result described in this case: A bank assistant receives malicious instructions embedded in a retrieved customer-support article” adds a relevant safeguard or verification step rather than weakening the solution.

Question 211

A team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. Which design decision is justified by the failure and the required outcome?

  1. Assume every blocked request is inherently an attack
  2. Discard challenge prompts after a single policy change
  3. Apply the proposed setting to all connected workloads before testing the case-specific boundary: a public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered
  4. Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds

Explanation

Correct Answer

D:

A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Confirm that measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

Incorrect Answers

A:

Because a team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered, “Assume every blocked request is inherently an attack” would not meet the requirement. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Instead, run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

B:

Because a team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered, “Discard challenge prompts after a single policy change” does not implement the control that the facts call for. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Instead, run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

C:

“Apply the proposed setting to all connected workloads before testing the case-specific boundary: a public agency must demonstrate that legitimate language-translation requests are not overblocked while” does not implement the control that the facts call for. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Instead, run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

Question 212

Situation: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. Which underlying technical principle explains the recommended implementation?

  1. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses
  2. The most powerful or least restricted model is always the safest choice regardless of task conditions
  3. Successful provisioning by itself proves all security, output, and operational constraints are satisfied
  4. The fact that the team started work is enough to prove correctness for this exact case: a public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered

Explanation

Correct Answer

A:

Because a team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered, a green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds. The decisive evidence is that measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

Incorrect Answers

B:

A team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. “The most powerful or least restricted model is always the safest choice regardless of task conditions” does not fully explain the decisive technical boundary in this case. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

C:

A team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. “Successful provisioning by itself proves all security, output, and operational constraints are satisfied” does not fully explain the decisive technical boundary in this case. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. The implementation must run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

D:

“The fact that the team started work is enough to prove correctness for this exact case: a public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe” does not fully explain the decisive technical boundary in this case. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. The implementation must run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

Question 213

Architecture case: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. What observable acceptance condition should be required before closing the task?

  1. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates
  2. The operator reports success without reproducing the observed behavior: a public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered
  3. The team closes a change ticket without replaying a representative positive and negative test
  4. A project dashboard remains green, although the affected request, output and permissions were not checked

Explanation

Correct Answer

A:

A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates. If this test fails, review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions.

Incorrect Answers

B:

“The operator reports success without reproducing the observed behavior: a public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are” alone does not establish all the required behavior. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

C:

Because a team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered, “The team closes a change ticket without replaying a representative positive and negative test” alone does not establish all the required behavior. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

D:

“A project dashboard remains green, although the affected request, output and permissions were not checked” would not meet the requirement. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

Question 214

Production requirement: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. Which first investigation is more useful than an unrelated health indicator?

  1. Assume the general service is at fault without checking the affected operation in this case: a public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered
  2. Restart every affected component before collecting request or tool traces
  3. Review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions
  4. Read only the broad service-health summary and skip actual workload evidence

Explanation

Correct Answer

C:

Because a team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered, review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

Incorrect Answers

A:

“Assume the general service is at fault without checking the affected operation in this case: a public agency must demonstrate that legitimate language-translation requests are not overblocked while” would not meet the requirement. Review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

B:

Because a team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered, “Restart every affected component before collecting request or tool traces” would not be the most direct first check for this failure. Review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

D:

A team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. “Read only the broad service-health summary and skip actual workload evidence” would not meet the requirement. Review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

Question 215

A team is investigating this task: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. Which TWO actions could conceal a serious failure while appearing faster? Select TWO answers.

  1. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates
  2. Approve release using only general task accuracy without abuse-case tests
  3. Inspect records that explain the affected process, then check the result described in this case: A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered
  4. Turn off safety filtering because ordinary users report occasional refusals
  5. Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds

Explanation

Correct Answer

B:

“Approve release using only general task accuracy without abuse-case tests” would violate the requirement. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

D:

A public agency must demonstrate that legitimate language-translation requests are not overblocked while unsafe instructions are filtered. “Turn off safety filtering because ordinary users report occasional refusals” would violate the requirement. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

Incorrect Answers

A:

“Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates” is the required outcome of a working control, not an action that weakens it. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

C:

“Inspect records that explain the affected process, then check the result described in this case: A public agency must demonstrate that legitimate language-translation requests are not overblocked while” adds a relevant safeguard or verification step rather than weakening the solution. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Confirm that measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

E:

“Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds” is the controlled implementation, not an unsafe workaround. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Confirm that measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

Question 216

Situation: A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. Which implementation best meets the requirement while honoring the stated constraints?

  1. Discard challenge prompts after a single policy change
  2. Apply the proposed setting to all connected workloads before testing the case-specific boundary: a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet
  3. Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds
  4. Assume every blocked request is inherently an attack

Explanation

Correct Answer

C:

Because a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet, a green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

Incorrect Answers

A:

A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. “Discard challenge prompts after a single policy change” does not implement the control that the facts call for. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Instead, run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

B:

“Apply the proposed setting to all connected workloads before testing the case-specific boundary: a shopping agent was updated” does not implement the control that the facts call for. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Instead, run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

D:

“Assume every blocked request is inherently an attack” would not meet the requirement. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Instead, run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

Question 217

Architecture case: A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. Which design boundary is most relevant to this decision?

  1. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses
  2. Successful provisioning by itself proves all security, output, and operational constraints are satisfied
  3. The most powerful or least restricted model is always the safest choice regardless of task conditions
  4. The fact that the team started work is enough to prove correctness for this exact case: a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet

Explanation

Correct Answer

A:

A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

Incorrect Answers

B:

A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. “Successful provisioning by itself proves all security, output, and operational constraints are satisfied” does not fully explain the decisive technical boundary in this case. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. The implementation must run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

C:

“The most powerful or least restricted model is always the safest choice regardless of task conditions” does not fully explain the decisive technical boundary in this case. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. The implementation must run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds.

D:

“The fact that the team started work is enough to prove correctness for this exact case: a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet” does not fully explain the decisive technical boundary in this case. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

Question 218

Production requirement: A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. Which evidence is strongest for proving this requirement was met?

  1. A project dashboard remains green, although the affected request, output and permissions were not checked
  2. The operator reports success without reproducing the observed behavior: a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet
  3. The team closes a change ticket without replaying a representative positive and negative test
  4. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates

Explanation

Correct Answer

D:

A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

Incorrect Answers

A:

Because a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet, “A project dashboard remains green, although the affected request, output and permissions were not checked” would not meet the requirement. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

B:

“The operator reports success without reproducing the observed behavior: a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet” alone does not establish all the required behavior. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

C:

A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. “The team closes a change ticket without replaying a representative positive and negative test” alone does not establish all the required behavior. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

Question 219

A team is investigating this task: A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. What targeted records should the responder inspect to distinguish configuration from runtime behavior?

  1. Assume the general service is at fault without checking the affected operation in this case: a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet
  2. Review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions
  3. Restart every affected component before collecting request or tool traces
  4. Read only the broad service-health summary and skip actual workload evidence

Explanation

Correct Answer

B:

Because a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet, review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

Incorrect Answers

A:

“Assume the general service is at fault without checking the affected operation in this case: a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet” would not meet the requirement. Review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions.

C:

“Restart every affected component before collecting request or tool traces” would not be the most direct first check for this failure. Review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

D:

A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. “Read only the broad service-health summary and skip actual workload evidence” would not meet the requirement. Review harmful and benign evaluation slices, moderation category severity, blocked-request logs, appeals and versioned policy decisions. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

Question 220

Situation: A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. Which TWO proposed practices create material safety, correctness, or reliability risks? Select TWO answers.

  1. Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates
  2. Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds
  3. Turn off safety filtering because ordinary users report occasional refusals
  4. Inspect records that explain the affected process, then check the result described in this case: A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet
  5. Approve release using only general task accuracy without abuse-case tests

Explanation

Correct Answer

C:

A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. “Turn off safety filtering because ordinary users report occasional refusals” would violate the requirement. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

E:

Because a shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet, “Approve release using only general task accuracy without abuse-case tests” would violate the requirement. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

Incorrect Answers

A:

A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. “Measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates” is the required outcome of a working control, not an action that weakens it. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses.

B:

A shopping agent was updated and now reveals internal tool names when prompted with a crafted retrieval snippet. “Run a documented abuse-case evaluation set, tune filters and prompt protections, and gate release on measurable false-positive and false-negative thresholds” is the controlled implementation, not an unsafe workaround. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Confirm that measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

D:

“Inspect records that explain the affected process, then check the result described in this case: A shopping agent was updated” adds a relevant safeguard or verification step rather than weakening the solution. A green ordinary-task benchmark does not establish safety against harmful inputs, indirect prompt injection or domain-specific prohibited responses. Confirm that measured safety violations fall below agreed thresholds while legitimate requests retain acceptable completion and escalation rates.

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