Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Practice Test 36

Question 350

An application needs to identify drift in retrieval relevance. Which governance implementation produces defensible evidence or authorization?

Use vector index uptime alone as proof of retrieval relevance.
Compare different unrelated test questions between versions and attribute score differences to drift.
Refresh source documents without recording which version/index produced prior benchmark results.
Monitor golden-set recall trends and index/model changes with alerts.

Explanation

Correct Answer

Track golden-set retrieval recall/precision across source/index and embedding-model versions, alerting when reproducible quality declines.

Incorrect Answers

A healthy index can return systematically wrong documents.

Confounded populations do not establish regression.

Lack of versioned evidence prevents diagnosis of relevance changes.

Question 351

A user says a prompt response caused an unauthorized account change. Which action resolves this policy-control failure?

Use the model's final sentence as proof that the user's account was modified.
Inspect only the access log of the chat frontend and ignore downstream tool execution.
Assume an error response means no account side effects occurred.
Correlate request, model route, tool execution and signed transaction logs.

Explanation

Correct Answer

Join authenticated request, model route, tool call arguments and the signed account-system receipt before determining whether an unauthorized change occurred.

Incorrect Answers

Model output can disagree with actual database writes.

Front-end activity does not prove which business action committed.

A failure may follow a partial or completed transaction.

Question 352

A team uses external LLM judges to approve every new model automatically. What is the correct compliance or accountability change?

Allow a single related FM judge to approve all model versions without thresholds or review.
Remove disagreement cases from the dataset so judges achieve near-perfect agreement.
Require independent thresholds and escalation for material judge disagreements.
Use semantic similarity as proof of safe business outcomes for irreversible actions.

Explanation

Correct Answer

Treat LLM-judge scores as one signal; calibrate with held-out human-labeled fixtures and escalate material judge disagreement under accountable rules.

Incorrect Answers

Correlated model errors can bias acceptance.

Suppressing hard cases conceals model uncertainty and risk.

Text similarity cannot validate real-world authorization or effects.

Question 353

A bank claims FM compliance simply because output looks professional. Which governance implementation produces defensible evidence or authorization?

Accept an answer as compliant solely because its formatting and tone look professional.
Validate policy-specific rules and retain source/result evidence for each decision.
Keep only high-level model benchmark scores without transaction-specific policy checks.
Allow the model to choose the compliance interpretation without named rule ownership.

Explanation

Correct Answer

Evaluate the bank's actual policy rules and documented evidence for each decision; polished language is not a test of financial or regulatory compliance.

Incorrect Answers

Fluency can coexist with unsupported financial claims.

General benchmarks cannot establish each regulated business rule was followed.

Policy constraints require authoritative definitions and review accountability.

Question 354

A large company has inconsistent tool permission practices. Which action resolves this policy-control failure?

Establish minimum enterprise controls and accountable department owners.
Allow each department to deploy privileged tools with no shared permission baseline.
Use one shared administrator token for all departments to simplify operations.
Treat a generic safe-prompt instruction as the enterprise tool permission policy.

Explanation

Correct Answer

Establish common minimum tool identity, action scope, approvals, logging and exception requirements with accountable business owners.

Incorrect Answers

Inconsistent controls create unreviewed cross-unit exposure.

Central convenience obscures identities and grants excessive rights.

Text instructions do not enforce actual backend actions.

Question 355

A production incident exposes personal data in traces. What is the correct compliance or accountability change?

Delete all evidence immediately without determining legal hold or incident scope.
Contain access, apply retention controls and preserve legally required incident evidence.
Leave the shared log group open while rewriting only customer-facing text.
Publish the affected prompt traces in a broad incident channel so everyone can help.

Explanation

Correct Answer

Contain trace access, investigate exposed recipients and retention obligations, and preserve required forensic evidence securely during incident response.

Incorrect Answers

Indiscriminate deletion can obstruct investigation and violate retention duties.

The underlying sensitive data exposure continues.

Broadcasting the data magnifies the leak.

Question 356

A deployment promotion has no exception process. Which governance implementation produces defensible evidence or authorization?

Require documented exception owner, reason, expiry and reassessment.
Set an indefinite blanket exception for every release from the same team.
Allow the model to decide whether its own failed safety tests deserve an exception.
Erase the failure finding once a release has been promoted under exception.

Explanation

Correct Answer

Document a named exception owner, bounded justification and expiry with re-approval rules; an exception is not a permanent bypass.

Incorrect Answers

Open-ended exclusions eliminate meaningful governance and reassessment.

Risk acceptance requires accountable authorized humans.

The exception must remain traceable to the original risk and decision.

Question 357

A framework changes its control mapping between quarters. Which action resolves this policy-control failure?

Reconcile new assessment coverage instead of treating old score as equivalent.
Compare old and new headline percentages as identical despite changed control definitions.
Delete superseded controls from the audit without recording what was replaced.
Assume a renamed framework label proves all previous tests still apply.

Explanation

Correct Answer

Re-map each changed control requirement and scope to fresh evidence before comparing quarterly assessment scores.

Incorrect Answers

Different denominator and requirements make scores noncomparable.

Historical lineage is needed to explain the assessment change.

Names do not establish substantive control equivalence.

Question 358

A model version is retired but an app silently switches providers. What is the correct compliance or accountability change?

Silently select any currently available model with the same brand name.
Use only latency comparison to approve a new model handling restricted records.
Record lifecycle transition and assess effects on data processing and outputs.
Keep quoting a retired model card as evidence of the replacement model's limitations.

Explanation

Correct Answer

Document the old/new provider and model IDs, assess processing locality, safety/quality differences and approve the lifecycle cutover explicitly.

Incorrect Answers

Different versions/providers can alter outputs and data handling.

Quality, privacy and policy controls must also be checked.

Version-specific documentation may not transfer to the new resource.

Question 359

An agent uses a data source with unclear license. Which governance implementation produces defensible evidence or authorization?

Verify permitted purpose/processing rights before indexing and model use.
Index any dataset found on the Internet because it is publicly downloadable.
Assume paying for the FM grants content rights to all third-party training/retrieval materials.
Record only the original source URL and ignore its applicable license/version.

Explanation

Correct Answer

Verify source license, intended purpose, derivative/indexing rights and retention restrictions before model processing or retrieval exposure.

Incorrect Answers

Public access does not imply unrestricted downstream licensing.

Model-service fees do not license unrelated source documents.

A locator alone does not establish permitted uses.

Question 360

A governance reviewer wants proof of current responsible-AI policy. Which action resolves this policy-control failure?

Submit only a generic responsible-AI promise written by the model.
Reuse last year's approval without checking new features or model/provider versions.
Report a single aggregated score without denominators, failures or known limitations.
Produce versioned tests, risk findings, approvals and operational metrics.

Explanation

Correct Answer

Provide policy version, evaluated risk cases, accountable approvals, incident findings and production metrics demonstrating current adherence.

Incorrect Answers

A promise without tests and operational evidence is not auditable.

Old acceptance may not cover changed behavior and data use.

Averages hide material subgroup and control gaps.

Pass Your Amazon AWS Certified Generative AI Developer - Professional AIP-C01 Exam Easy!

Question 350

Explanation

Correct Answer

Incorrect Answers

Question 351

Explanation

Correct Answer

Incorrect Answers

Question 352

Explanation

Correct Answer

Incorrect Answers

Question 353

Explanation

Correct Answer

Incorrect Answers

Question 354

Explanation

Correct Answer

Incorrect Answers

Question 355

Explanation

Correct Answer

Incorrect Answers

Question 356

Explanation

Correct Answer

Incorrect Answers

Question 357

Explanation

Correct Answer

Incorrect Answers

Question 358

Explanation

Correct Answer

Incorrect Answers

Question 359

Explanation

Correct Answer

Incorrect Answers

Question 360

Explanation

Correct Answer

Incorrect Answers

AWS Certified Generative AI Developer - Professional AIP-C01 Practice Tests