IAPP AIGP Practice Test 2 - Prepaway
Practice Exams:

Pass IAPP AIGP Exam in First Attempt Guaranteed!

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

30 Days Free Updates, Instant Download!

Verified By Experts
Premium Bundle
$29.98
$79.99 $109.97

AIGP Premium Bundle


  • Premium File 222 Questions & Answers. Last update: Nov 15, 2026
  • Training Course 124 Lectures
  • Study Guide 566 Pages

IAPP AIGP Practice Test 2

Question 10

A municipality integrates a vendor high-risk AI system and assumes the supplier's technical file covers local instructions and operator oversight. Which test result would best demonstrate that the actual governance concern has been addressed?

  1. Contracts and local procedures show provider evidence plus deployer oversight in the correct roles.
  2. The interface gives an appropriate clear AI interaction notice and validates user understanding.
  3. The compliance file links model design, evaluation, mitigations and released version.
  4. Applicable market-entry checks and escalation paths are documented before distribution.

Explanation

Correct Answer

A:

Contracts and local procedures show provider evidence plus deployer oversight in the correct roles. Roles in AI-specific laws can impose separate duties throughout a distributed supply chain.

Incorrect Answers

B:

The interface gives an appropriate clear AI interaction notice and validates user understanding. Roles in AI-specific laws can impose separate duties throughout a distributed supply chain. Contracts and local procedures show provider evidence plus deployer oversight in the correct roles.

C:

The compliance file links model design, evaluation, mitigations and released version. Roles in AI-specific laws can impose separate duties throughout a distributed supply chain. Contracts and local procedures show provider evidence plus deployer oversight in the correct roles.

D:

Applicable market-entry checks and escalation paths are documented before distribution. Roles in AI-specific laws can impose separate duties throughout a distributed supply chain. Contracts and local procedures show provider evidence plus deployer oversight in the correct roles.

Question 11

An open-feedback classifier accepts untrusted public labels directly into an automated retraining pipeline. Which additional observation would most strongly confirm the particular risk or control gap?

  1. The proposed pipeline treats anonymous labels as authoritative without provenance checks.
  2. The benchmark covers only the selected large model with no baseline or limitation analysis.
  3. Requirements list fixed eligibility rules with no evidence of complex inference needs.
  4. The risk sheet records only occurrence frequency and ignores injury severity.

Explanation

Correct Answer

A:

The proposed pipeline treats anonymous labels as authoritative without provenance checks.

Incorrect Answers

B:

The benchmark covers only the selected large model with no baseline or limitation analysis. The proposed pipeline treats anonymous labels as authoritative without provenance checks.

C:

Requirements list fixed eligibility rules with no evidence of complex inference needs. The proposed pipeline treats anonymous labels as authoritative without provenance checks.

D:

The risk sheet records only occurrence frequency and ignores injury severity. The proposed pipeline treats anonymous labels as authoritative without provenance checks.

Question 12

A customer-support AI routes uncertain legal questions to itself instead of human specialists because confidence is never checked. Which test result would best demonstrate that the actual governance concern has been addressed?

  1. Test cases outside supported knowledge reliably trigger the documented fallback process.
  2. The approved impact report documents claimant harms, mitigations and monitored thresholds.
  3. Accessibility testing shows equivalent core journey tasks can be completed without hearing speech.
  4. The pilot records performance under realistic difficult cases and meets pre-agreed guardrail criteria.

Explanation

Correct Answer

A:

Test cases outside supported knowledge reliably trigger the documented fallback process. The required intervention is to define uncertainty thresholds and human escalation for unsupported or high-stakes requests. The particular diagnostic issue is that low-confidence advice is sent automatically while specialists never receive the flagged conversation. The governing distinction is that model confidence and scope limits must influence system decisions, not merely appear in an appendix.

Incorrect Answers

B:

The approved impact report documents claimant harms, mitigations and monitored thresholds. Model confidence and scope limits must influence system decisions, not merely appear in an appendix. Test cases outside supported knowledge reliably trigger the documented fallback process.

C:

Accessibility testing shows equivalent core journey tasks can be completed without hearing speech. Model confidence and scope limits must influence system decisions, not merely appear in an appendix. Test cases outside supported knowledge reliably trigger the documented fallback process.

D:

The pilot records performance under realistic difficult cases and meets pre-agreed guardrail criteria. Model confidence and scope limits must influence system decisions, not merely appear in an appendix. Test cases outside supported knowledge reliably trigger the documented fallback process.

Question 13

A chatbot team closes a harmful-output ticket once it edits the prompt, without testing whether the problem is resolved. Which governance intervention most directly addresses this specific situation?

  1. Maintain validated rollback artifacts and a rehearsed controlled reversion procedure.
  2. Require root-cause investigation, test evidence, owner and verified closure for model issues.
  3. Repeat adversarial testing when capabilities or access expand and at planned intervals.
  4. Classify and document AI harms using an incident taxonomy with escalation and reporting review.

Explanation

Correct Answer

B:

Require root-cause investigation, test evidence, owner and verified closure for model issues. The action addresses issue tracking and closure evidence rather than treating every AI control as interchangeable.

Incorrect Answers

A:

Maintain validated rollback artifacts and a rehearsed controlled reversion procedure. The action addresses issue tracking and closure evidence rather than treating every AI control as interchangeable. Require root-cause investigation, test evidence, owner and verified closure for model issues.

C:

Repeat adversarial testing when capabilities or access expand and at planned intervals. The action addresses issue tracking and closure evidence rather than treating every AI control as interchangeable. Require root-cause investigation, test evidence, owner and verified closure for model issues.

D:

Classify and document AI harms using an incident taxonomy with escalation and reporting review. The action addresses issue tracking and closure evidence rather than treating every AI control as interchangeable. Require root-cause investigation, test evidence, owner and verified closure for model issues.

Question 14

A bank builds an internal retrieval agent using documents that employees may edit; it performs no adversarial document testing. Which test result would best demonstrate that the actual governance concern has been addressed?

  1. The organization can document how the request was handled across relevant data and model artifacts.
  2. Independent rare-event tests meet safety-relevant detection criteria with documented limits.
  3. Field-like trials meet the required false-match and false-nonmatch criteria.
  4. Adversarial tests show external content cannot bypass authorization or alter privileged actions.

Explanation

Correct Answer

D:

Adversarial tests show external content cannot bypass authorization or alter privileged actions. The required intervention is to test untrusted retrieved instructions and prevent them from overriding system policy or tool permissions. The particular diagnostic issue is that injected instructions inside a retrieved document cause unauthorized retrieval in a dry run. The governing distinction is that retrieved data can contain hostile instructions and must not become trusted control input.

Incorrect Answers

A:

The organization can document how the request was handled across relevant data and model artifacts. Retrieved data can contain hostile instructions and must not become trusted control input. Adversarial tests show external content cannot bypass authorization or alter privileged actions.

B:

Independent rare-event tests meet safety-relevant detection criteria with documented limits. Retrieved data can contain hostile instructions and must not become trusted control input. Adversarial tests show external content cannot bypass authorization or alter privileged actions.

C:

Field-like trials meet the required false-match and false-nonmatch criteria. Retrieved data can contain hostile instructions and must not become trusted control input. Adversarial tests show external content cannot bypass authorization or alter privileged actions.

Question 15

A benefits agency deploys fraud screening without a simple channel for beneficiaries to dispute incorrect flags. Which additional observation would most strongly confirm the particular risk or control gap?

  1. Affected residents cannot correct their information or learn why assistance is delayed.
  2. Access reports show the tool can read records outside its current authorized business function.
  3. Backtesting shows sustained error growth concentrated in newly volatile product categories.
  4. Policy settings have no reliable geography or use-case restriction despite a regional compliance decision.

Explanation

Correct Answer

A:

Affected residents cannot correct their information or learn why assistance is delayed. This is the diagnostic observation that distinguishes complaint and appeal route from other concerns within govern the deployment and use of the ai system. It supports the targeted response: Establish an accessible appeal and human case review route with recorded outcomes. The underlying governance reason is that contestability helps protect individuals when AI informs consequential decisions. A meaningful follow-up would be to verify that a test appeal reaches an authorized reviewer and resolves documented factual errors.

Incorrect Answers

B:

Access reports show the tool can read records outside its current authorized business function. Affected residents cannot correct their information or learn why assistance is delayed.

C:

Backtesting shows sustained error growth concentrated in newly volatile product categories. Affected residents cannot correct their information or learn why assistance is delayed.

D:

Policy settings have no reliable geography or use-case restriction despite a regional compliance decision. Affected residents cannot correct their information or learn why assistance is delayed.

Question 16

A translation service uses a subcontracted inference provider but does not disclose that downstream data processing arrangement. Which test result would best demonstrate that the actual governance concern has been addressed?

  1. Every customer-facing experiment has a signed risk record and constrained pilot plan before access.
  2. A signed control map identifies applicable policies and closes conflicting requirements.
  3. Subcontractor changes trigger documented risk review and compliant data-handling checks.
  4. A tabletop exercise pauses the model and moves cases to an accountable manual workflow.

Explanation

Correct Answer

C:

Subcontractor changes trigger documented risk review and compliant data-handling checks. The required intervention is to add approved-subprocessor disclosure, flow-down controls and change notification to procurement processes. The particular diagnostic issue is that a support ticket reveals requests are routed to an unreviewed fourth-party hosting service. The governing distinction is that aI supply-chain controls should follow material processing and contractual dependencies.

Incorrect Answers

A:

Every customer-facing experiment has a signed risk record and constrained pilot plan before access. AI supply-chain controls should follow material processing and contractual dependencies. Subcontractor changes trigger documented risk review and compliant data-handling checks.

B:

A signed control map identifies applicable policies and closes conflicting requirements. AI supply-chain controls should follow material processing and contractual dependencies. Subcontractor changes trigger documented risk review and compliant data-handling checks.

D:

A tabletop exercise pauses the model and moves cases to an accountable manual workflow. AI supply-chain controls should follow material processing and contractual dependencies. Subcontractor changes trigger documented risk review and compliant data-handling checks.

Question 17

A regional call center buys an AI triage assistant but agents have no training on when to disregard its recommendations. Which governance intervention most directly addresses this specific situation?

  1. Separate advisory steps from action tools and limit autonomy, spending and approval authority.
  2. Assess staff readiness, operating procedures and override skills before launch.
  3. Select an appropriate predictive time-series model rather than defaulting to generative text output.
  4. Set a defined human decision boundary matching applicable policy and risk tolerance.

Explanation

Correct Answer

B:

Assess staff readiness, operating procedures and override skills before launch. The action addresses workforce readiness for automation rather than treating every AI control as interchangeable. The decisive signal to verify is that pilot agents follow low-confidence suggestions because they believe the system is always authoritative. This matters because operational adoption requires prepared human users as well as working software. Success should be measured by confirming that trained staff correctly override unsupported advice in scenario tests and record exceptions.

Incorrect Answers

A:

Separate advisory steps from action tools and limit autonomy, spending and approval authority. Operational adoption requires prepared human users as well as working software. Assess staff readiness, operating procedures and override skills before launch.

C:

Select an appropriate predictive time-series model rather than defaulting to generative text output. Operational adoption requires prepared human users as well as working software. Assess staff readiness, operating procedures and override skills before launch.

D:

Set a defined human decision boundary matching applicable policy and risk tolerance. Operational adoption requires prepared human users as well as working software. Assess staff readiness, operating procedures and override skills before launch.

Question 18

A public-service translation model is trained mainly on urban speech and performs poorly for rural dialect speakers. Which governance intervention most directly addresses this specific situation?

  1. Assess memorization and re-identification risk before sharing synthetic data.
  2. Version code, dependencies, datasets and evaluation settings for reproducible experiments.
  3. Stratify and validate training and test samples for the actual served language communities.
  4. Add schema, range and distribution validation at ingestion with lineage and rollback.

Explanation

Correct Answer

C:

Stratify and validate training and test samples for the actual served language communities. The action addresses representative sampling for rural users rather than treating every AI control as interchangeable. The decisive signal to verify is that dataset profiling reveals that rural utterances represent a tiny share of the intended population. This matters because training distribution should represent the deployment population rather than the easiest data source. Success should be measured by confirming that evaluation demonstrates acceptable subgroup coverage and performance against defined criteria.

Incorrect Answers

A:

Assess memorization and re-identification risk before sharing synthetic data. Training distribution should represent the deployment population rather than the easiest data source. Stratify and validate training and test samples for the actual served language communities.

B:

Version code, dependencies, datasets and evaluation settings for reproducible experiments. Training distribution should represent the deployment population rather than the easiest data source. Stratify and validate training and test samples for the actual served language communities.

D:

Add schema, range and distribution validation at ingestion with lineage and rollback. Training distribution should represent the deployment population rather than the easiest data source. Stratify and validate training and test samples for the actual served language communities.

Question 19

A legal publisher weighs a community model against a commercial API for confidential document summaries. Which governance intervention most directly addresses this specific situation?

  1. Benchmark acceptable accuracy against cost, latency and energy use rather than optimizing one measure.
  2. Assess licenses, support, security, update controls, data use terms and governance before choosing.
  3. Assess edge inference with a fallback and model-update process suited to disconnected sites.
  4. Compare prompt-based controls and fine-tuning on data rights, evaluations and operational cost.

Explanation

Correct Answer

B:

Assess licenses, support, security, update controls, data use terms and governance before choosing. The action addresses open-source versus proprietary models rather than treating every AI control as interchangeable. The decisive signal to verify is that the open model lacks a clear maintenance owner while the hosted API offers limited training-data controls. This matters because open availability and commercial purchase each create distinct governance obligations. Success should be measured by confirming that a choice memo records tested privacy, performance, licensing and support tradeoffs.

Incorrect Answers

A:

Benchmark acceptable accuracy against cost, latency and energy use rather than optimizing one measure. Open availability and commercial purchase each create distinct governance obligations. Assess licenses, support, security, update controls, data use terms and governance before choosing.

C:

Assess edge inference with a fallback and model-update process suited to disconnected sites. Open availability and commercial purchase each create distinct governance obligations. Assess licenses, support, security, update controls, data use terms and governance before choosing.

D:

Compare prompt-based controls and fine-tuning on data rights, evaluations and operational cost. Open availability and commercial purchase each create distinct governance obligations. Assess licenses, support, security, update controls, data use terms and governance before choosing.

Question 20

An employment service corrects a client's record but its risk-scoring system still uses the stale feature snapshot. Which additional observation would most strongly confirm the particular risk or control gap?

  1. The original notice explains handling complaints but contains no clear explanation of model training reuse.
  2. The feature inventory includes diagnoses with no demonstrated predictive value for the reminder task.
  3. The source file is corrected while the deployed feature store still contains the inaccurate old field.
  4. The draft contract describes infrastructure support but not processing limits or audit cooperation.

Explanation

Correct Answer

C:

The source file is corrected while the deployed feature store still contains the inaccurate old field.

Incorrect Answers

A:

The original notice explains handling complaints but contains no clear explanation of model training reuse. The source file is corrected while the deployed feature store still contains the inaccurate old field.

B:

The feature inventory includes diagnoses with no demonstrated predictive value for the reminder task. The source file is corrected while the deployed feature store still contains the inaccurate old field.

D:

The draft contract describes infrastructure support but not processing limits or audit cooperation. The source file is corrected while the deployed feature store still contains the inaccurate old field.

Artificial Intelligence Governance Professional (AIGP) Practice Tests