Amazon AWS Certified AI Practitioner AIF-C01 Practice Test 45
Question 440
A regulated AI application must access a supported AWS service privately rather than over the public internet. A colleague suggests: Use a publicly shared access key in a document to achieve private networking. Which finding BEST explains why the suggestion should be rejected?
- Exposing credentials does not create private network connectivity
- The shortcut is safe because its terminology is familiar to the team
- Access policy labels do not classify sensitive object content
- External content does not gain instruction authority from its wording
Explanation
Correct Answer
A:Exposing credentials does not create private network connectivity. This is the direct defect in the proposed action: Use a publicly shared access key in a document to achieve private networking.
Incorrect Answers
B:Familiar wording is not proof that an action is appropriate. The actual defect is: Exposing credentials does not create private network connectivity.
C:This finding is associated with a different problem: A data lake used for model grounding may contain customer identification numbers inside unstructured S3 objects.
D:This concern belongs to another situation: A knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets.
Question 441
A regulated AI application must access a supported AWS service privately rather than over the public internet. After implementing Evaluate AWS PrivateLink connectivity where supported, which observation BEST verifies the intended result?
- The team approves Evaluate AWS PrivateLink connectivity where supported but does not run an operational test
- Findings locate example sensitive objects for authorized remediation
- The team schedules a future test of whether network tests confirm traffic follows the intended private endpoint path, but has no result
- Network tests confirm traffic follows the intended private endpoint path
Explanation
Correct Answer
D:The observed result, Network tests confirm traffic follows the intended private endpoint path, provides direct evidence. PrivateLink can provide private connectivity to supported service endpoints.
Incorrect Answers
A:The required observable condition is: Network tests confirm traffic follows the intended private endpoint path.
B:This observation would be evidence for another case: A data lake used for model grounding may contain customer identification numbers inside unstructured S3 objects, not the current scenario.
C:Verify the actual outcome: Network tests confirm traffic follows the intended private endpoint path.
Question 442
The initial scenario was: A regulated AI application must access a supported AWS service privately rather than over the public internet. The original control is already in place. Now the service is deployed in an aws region without a required endpoint. Which ADDITIONAL action BEST addresses the new condition?
- Reassess regional endpoint support and a compliant architecture
- Network tests confirm traffic follows the intended private endpoint path
- Use a publicly shared access key in a document to achieve private networking
- Evaluate AWS PrivateLink connectivity where supported
Explanation
Correct Answer
A:The service is deployed in an AWS Region without a required endpoint. The appropriate added action is: Reassess regional endpoint support and a compliant architecture.
Incorrect Answers
B:This describes evidence for the original approach, not a new action: Network tests confirm traffic follows the intended private endpoint path.
C:This original shortcut remains unsuitable: Exposing credentials does not create private network connectivity.
D:This was the original control and is already implemented. PrivateLink can provide private connectivity to supported service endpoints.
Question 443
A data lake used for model grounding may contain customer identification numbers inside unstructured S3 objects. Which TWO statements jointly identify the best initial approach and an observable verification result? (Select TWO.)
- Configure output-side sensitive-information guardrails and tests
- Findings locate example sensitive objects for authorized remediation
- Assume IAM policy names automatically identify PII inside documents
- Treat retrieved content as untrusted and enforce prompt-injection defenses
- Evaluate Amazon Macie for discovering sensitive data in supported S3 content
Explanation
Correct Answers
B:Correct operating evidence. Observing findings locate example sensitive objects for authorized remediation verifies this approach instead of relying on unperformed plans.
E:Correct approach. Macie helps discover sensitive data patterns in Amazon S3. This directly serves the stated situation: A data lake used for model grounding may contain customer identification numbers inside unstructured S3 objects.
Incorrect Answers
A:This would be appropriate if the separate requirement arose: The team next wants to block AI output that repeats sensitive identifiers. That is not the present condition.
C:Access policy labels do not classify sensitive object content. The suggested shortcut would not meet the present requirement.
D:This addresses a different scenario: A knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets. It does not constitute the requested approach or operating result for the current case.
Question 444
A data lake used for model grounding may contain customer identification numbers inside unstructured S3 objects. A colleague suggests: Assume IAM policy names automatically identify PII inside documents. Which finding BEST explains why the suggestion should be rejected?
- The shortcut is safe because its terminology is familiar to the team
- External content does not gain instruction authority from its wording
- Prompt text alone does not guarantee protection against disclosure
- Access policy labels do not classify sensitive object content
Explanation
Correct Answer
D:Access policy labels do not classify sensitive object content. This is the direct defect in the proposed action: Assume IAM policy names automatically identify PII inside documents.
Incorrect Answers
A:Familiar wording is not proof that an action is appropriate. The actual defect is: Access policy labels do not classify sensitive object content.
B:This finding is associated with a different problem: A knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets.
C:This concern belongs to another situation: A public-facing Bedrock application needs to mask sensitive information in generated responses.
Question 445
A data lake used for model grounding may contain customer identification numbers inside unstructured S3 objects. After implementing Evaluate Amazon Macie for discovering sensitive data in supported S3 content, which observation BEST verifies the intended result?
- Findings locate example sensitive objects for authorized remediation
- Red-team documents fail to trigger protected-data disclosure
- The team approves Evaluate Amazon Macie for discovering sensitive data in supported S3 content but does not run an operational test
- The team schedules a future test of whether findings locate example sensitive objects for authorized remediation, but has no result
Explanation
Correct Answer
A:The observed result, Findings locate example sensitive objects for authorized remediation, provides direct evidence. Macie helps discover sensitive data patterns in Amazon S3.
Incorrect Answers
B:This observation would be evidence for another case: A knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets, not the current scenario.
C:The required observable condition is: Findings locate example sensitive objects for authorized remediation.
D:An unperformed future check is not evidence. Verify the actual outcome: Findings locate example sensitive objects for authorized remediation.
Question 446
The initial scenario was: A data lake used for model grounding may contain customer identification numbers inside unstructured S3 objects. The original control is already in place. Now the team next wants to block ai output that repeats sensitive identifiers. Which ADDITIONAL action BEST addresses the new condition?
- Configure output-side sensitive-information guardrails and tests
- Evaluate Amazon Macie for discovering sensitive data in supported S3 content
- Assume IAM policy names automatically identify PII inside documents
- Findings locate example sensitive objects for authorized remediation
Explanation
Correct Answer
A:The team next wants to block AI output that repeats sensitive identifiers. The appropriate added action is: Configure output-side sensitive-information guardrails and tests.
Incorrect Answers
B:This was the original control and is already implemented. Macie helps discover sensitive data patterns in Amazon S3.
C:This original shortcut remains unsuitable: Access policy labels do not classify sensitive object content.
D:This describes evidence for the original approach, not a new action: Findings locate example sensitive objects for authorized remediation.
Question 447
A knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets. Which option BEST meets the stated AI/ML or AWS requirement?
- Treat any retrieved sentence beginning with 'system' as privileged
- Configure supported sensitive-information controls in Amazon Bedrock Guardrails
- Treat retrieved content as untrusted and enforce prompt-injection defenses
- Apply tool-level permission scoping and confirmation gates
Explanation
Correct Answer
C:Retrieval text can carry adversarial instructions that must not override trusted policy. In this situation, choose: Treat retrieved content as untrusted and enforce prompt-injection defenses.
Incorrect Answers
A:External content does not gain instruction authority from its wording. The proposal, Treat any retrieved sentence beginning with 'system' as privileged, does not satisfy this need.
B:This can be valid for a different task: A public-facing Bedrock application needs to mask sensitive information in generated responses. The current request is different and needs the selected approach.
D:This is appropriate only after a distinct change: The assistant later needs an external transaction tool. It is not the best initial response to a knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets.
Question 448
A knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets. A colleague suggests: Treat any retrieved sentence beginning with 'system' as privileged. Which finding BEST explains why the suggestion should be rejected?
- Shared broad credentials erase identity boundaries and increase risk
- The shortcut is safe because its terminology is familiar to the team
- External content does not gain instruction authority from its wording
- Prompt text alone does not guarantee protection against disclosure
Explanation
Correct Answer
C:External content does not gain instruction authority from its wording. This is the direct defect in the proposed action: Treat any retrieved sentence beginning with 'system' as privileged.
Incorrect Answers
A:This concern belongs to another situation: A support agent must access third-party tools on behalf of individual authenticated employees with different permissions.
B:Familiar wording is not proof that an action is appropriate. The actual defect is: External content does not gain instruction authority from its wording.
D:This finding is associated with a different problem: A public-facing Bedrock application needs to mask sensitive information in generated responses.
Question 449
A knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets. After implementing Treat retrieved content as untrusted and enforce prompt-injection defenses, which observation BEST verifies the intended result?
- The team schedules a future test of whether red-team documents fail to trigger protected-data disclosure, but has no result
- Red-team documents fail to trigger protected-data disclosure
- The team approves Treat retrieved content as untrusted and enforce prompt-injection defenses but does not run an operational test
- Test prompts containing synthetic identifiers produce the prescribed safe handling
Explanation
Correct Answer
B:The observed result, Red-team documents fail to trigger protected-data disclosure, provides direct evidence. Retrieval text can carry adversarial instructions that must not override trusted policy.
Incorrect Answers
A:An unperformed future check is not evidence. Verify the actual outcome: Red-team documents fail to trigger protected-data disclosure.
C:Approval is a plan, not a demonstrated result. The required observable condition is: Red-team documents fail to trigger protected-data disclosure.
D:This observation would be evidence for another case: A public-facing Bedrock application needs to mask sensitive information in generated responses, not the current scenario.
Question 450
The initial scenario was: A knowledge-grounded assistant encounters a document containing malicious instructions to leak system secrets. The original control is already in place. Now the assistant later needs an external transaction tool. Which ADDITIONAL action BEST addresses the new condition?
- Apply tool-level permission scoping and confirmation gates
- Red-team documents fail to trigger protected-data disclosure
- Treat any retrieved sentence beginning with 'system' as privileged
- Treat retrieved content as untrusted and enforce prompt-injection defenses
Explanation
Correct Answer
A:The assistant later needs an external transaction tool. The appropriate added action is: Apply tool-level permission scoping and confirmation gates.
Incorrect Answers
B:This describes evidence for the original approach, not a new action: Red-team documents fail to trigger protected-data disclosure.
C:This original shortcut remains unsuitable: External content does not gain instruction authority from its wording.
D:This was the original control and is already implemented. Retrieval text can carry adversarial instructions that must not override trusted policy.