A good AIGP practice exam mirrors the real one: 100 questions in 2 hours 45 minutes, spread across the four BoK v2.1 domains, mostly applied scenarios with some multi-select items. Use it twice. Take one mock about two to three weeks before your exam to find weak domains, and one in the final week to confirm your readiness. If you consistently score 75% or more on unseen questions, you have a reasonable buffer. Below are ten original sample questions with explanations.
Practice questions only help if they look like the real exam and if you review them properly. This page gives you ten AIGP-style questions, one or more from every domain, and explains how to turn practice results into a study plan.
Ten AIGP practice questions (BoK v2.1)
These are original questions written by Archuz for practice. They are not IAPP exam items.
An HR software company builds an AI tool that ranks job candidates and sells it to employers in the EU. A retailer licenses it to screen applicants. Who is the provider under the EU AI Act?
- AThe retailer
- BThe HR software company
- CBoth equally
- DNeither, because recruitment tools aren't regulated
Show answer
Correct: B — The HR software company
The developer that places the system on the market under its own name is the provider. The retailer is a deployer. Recruitment AI is an Annex III high-risk use, so D is wrong.
A model performs well in testing but its accuracy declines over the following months as customer behaviour changes. What is this called?
- AOverfitting
- BHallucination
- CData drift
- DModel inversion
Show answer
Correct: C — Data drift
Changes in the input data distribution after deployment (data or concept drift) degrade performance. That's why post-deployment monitoring is a core governance control.
Which of these is a prohibited AI practice under Article 5 of the EU AI Act?
- AA customer chatbot that doesn't disclose that it's AI
- BA credit-scoring system used to assess loan applicants
- CInferring employees' emotions at work, other than for medical or safety reasons
- DA CV-screening tool used to rank job applicants
Show answer
Correct: C — Inferring employees' emotions at work, other than for medical or safety reasons
Emotion recognition in the workplace and in education is prohibited under Article 5(1)(f), except for medical or safety reasons (C). A chatbot that fails to disclose it's AI (A) breaches a transparency duty but isn't a prohibited practice. Credit scoring (B) and CV screening (D) are high-risk uses, regulated but allowed.
An organisation wants a certifiable management system for its AI activities. Which standard fits best?
- AISO/IEC 42001
- BNIST AI RMF
- CISO/IEC 42005
- DOECD AI Principles
Show answer
Correct: A — ISO/IEC 42001
ISO/IEC 42001 is a certifiable AI management system standard. The NIST AI RMF is a voluntary framework, ISO/IEC 42005 gives guidance on AI system impact assessment, and the OECD Principles are high-level policy principles.
Which TWO statements about AI-specific laws outside the EU are accurate?
- ASouth Korea has adopted a framework AI law, the AI Basic Act
- BNo US state has passed AI-specific legislation
- CNew York City requires bias audits for certain automated employment decision tools
- DChina has no rules on generative AI
Show answer
Correct: A and C
South Korea's AI Basic Act took effect in January 2026, and NYC Local Law 144 requires bias audits for automated employment decision tools. Several US states have AI laws, and China regulates generative AI services.
A team wants to use customer support transcripts to fine-tune a model. What should governance require first?
- AFine-tune first, then test the model's outputs for privacy leakage
- BUpdate the privacy notice to mention AI training, then proceed
- CConfirm the vendor's training environment is secure, then proceed
- DConfirm a lawful basis and purpose compatibility, and minimise or de-identify the data
Show answer
Correct: D — Confirm a lawful basis and purpose compatibility, and minimise or de-identify the data
Reusing personal data for training raises purpose-limitation and lawful-basis questions, which come before use, along with minimisation (D). Testing outputs afterwards (A) is useful but too late as a first step. A notice update (B) informs people but doesn't establish a lawful basis, and environment security (C) is necessary but doesn't settle whether the use is permitted.
Before releasing a generative AI feature, which activity best finds ways users might bypass safety controls?
- AUnit testing
- BRed teaming
- CLoad testing
- DA/B testing
Show answer
Correct: B — Red teaming
Red teaming is adversarial testing designed to find failures such as jailbreaks, prompt injection and harmful outputs. See what AI red teaming is.
A company is buying an AI model from a vendor. Which contract term most directly supports ongoing governance?
- AA service-level agreement for uptime and response times
- BAn obligation for the vendor to notify the customer of material model changes
- CAn intellectual property indemnity for model outputs
- DA cap on the vendor's liability under the contract
Show answer
Correct: B — An obligation for the vendor to notify the customer of material model changes
Model updates can change risk, so change notification lets the deployer re-test and reassess, which is what ongoing governance needs (B). An SLA (A) covers availability, an IP indemnity (C) covers a specific legal risk, and a liability cap (D) allocates financial exposure. None of them tells you when the model has changed.
Reviewers approve 99% of an AI system's loan-denial recommendations within seconds. What risk does this most suggest?
- AModel overfitting to the training data
- BData poisoning of the training pipeline
- CModel drift reducing accuracy over time
- DAutomation bias undermining meaningful human oversight
Show answer
Correct: D — Automation bias undermining meaningful human oversight
Near-total approval within seconds suggests reviewers are deferring to the system rather than assessing it: automation bias (D). Oversight only works if reviewers can and do challenge outputs. Overfitting (A), poisoning (B) and drift (C) are model problems; the scenario describes a human review problem.
An AI agent can send emails and issue refunds on behalf of customer service staff. Which control is most important?
- ALimiting its permissions and requiring human approval above a defined refund threshold
- BGiving the agent broad administrative access so it can resolve any issue
- CTurning off logging to protect privacy
- DLetting the agent update its own instructions
Show answer
Correct: A — Limiting its permissions and requiring human approval above a defined refund threshold
Least privilege and human approval gates for high-impact actions are core agentic AI controls. Logging should be kept for accountability.
How to use AIGP practice exams properly
- Start with a short diagnostic (30–40 questions) before you study, so you can see where you stand by domain.
- Practise by domain while you learn, in sets of 20–30. Review every question, including the ones you got right by guessing.
- Keep an error log. For each miss, record the domain, the concept and the reason: knowledge gap, misread stem, or a distractor that fooled you.
- Take a full 100-question mock about 2–3 weeks out, under real timing and with the break.
- Retarget your weakest two competencies, then take a second mock or a large fresh set in the final week.
Interpreting your practice scores
| Score on unseen, mixed questions | What it suggests |
|---|---|
| Below 65% | Significant gaps; consider moving your exam date |
| 65–74% | Borderline; target your weak domains and re-test |
| 75–84% | Reasonable buffer for most candidates |
| 85%+ | Strong; focus on pacing and rest |
These bands are Archuz guidance, not IAPP thresholds. They also depend on how realistic your question source is.
For how to combine a question bank with full mocks, see question bank vs practice exam. For the question types, see the AIGP exam format.
290 more questions where these came from
The AIGP Complete Pack has 300 BoK v2.1-mapped questions, including a 100-question timed mock, with an explanation for every option.
Get the Complete PackA full-mock routine that builds exam fitness
- Book a three-hour block with no interruptions, at the same time of day as your real exam.
- Use exam conditions: no notes, no pausing, a 15-minute break at the halfway point.
- Use checkpoints: about 50 questions by 80 minutes.
- Mark your confidence on each answer (sure, unsure, guess) so you can see whether your guesses are good.
- Review the same day or the next: every wrong answer and every guess, with the reason recorded.
- Re-weight your plan using your domain scores.
Two full mocks are usually enough: one about two to three weeks before the exam, one about a week before.
Frequently asked questions
Is there an official AIGP practice exam?
Yes. IAPP sells an official practice exam through its store. It's useful for checking the format. Many candidates add a larger third-party bank for volume.
How many practice questions should I do before the AIGP?
Around 300 quality questions, including at least one full timed mock, is a solid target, as long as you review every answer.
What practice score means I'm ready for the AIGP?
Consistently scoring about 75% or more on realistic questions you haven't seen before is a reasonable sign of readiness. IAPP doesn't publish an equivalent threshold.
Are free AIGP practice questions reliable?
Quality varies. Check that questions are mapped to BoK v2.1, use scenarios rather than pure recall, and explain why each wrong option is wrong.
How long is a full AIGP practice exam?
A realistic mock has 100 questions in 2 hours 45 minutes, with a short break at the halfway point.
When should I take an AIGP full mock exam?
A common pattern is one full mock about two to three weeks before the exam and another about a week before, both under timed exam conditions with a full review afterwards.
Sources
- IAPP Store: AIGP Exam
- IAPP: AIGP Body of Knowledge
- Regulation (EU) 2024/1689 (EU AI Act), EUR-Lex
- ISO/IEC 42001:2023
Last reviewed: October 2026. Sample questions are original Archuz practice items, not IAPP exam content.