Quick answer

In the final week before the AIGP, stop learning new material and start consolidating. Days 7–5: review your weakest two competencies and your error log. Day 4: sit a final timed mock or a large fresh set. Days 3–2: go through cram sheets on roles, EU AI Act tiers and timelines, framework structures and lifecycle controls. Day 1: light review and logistics only. Sleep matters more than one more hour of reading.

The last week isn't the time to read a new textbook. Your goal is to make what you already know fast and reliable under exam conditions, to fix the few gaps that cost the most points, and to arrive rested. Here's a day-by-day plan, a list of high-yield facts, and how to use a cram guide well.

The 7-day AIGP plan

DayFocusTime
7Review your error log; list your two weakest competencies2–3 h
6Targeted study and 30–40 questions on weak competency 12–3 h
5Targeted study and 30–40 questions on weak competency 22–3 h
4Final timed mock (100 questions, real timing, with the break)3.5 h + review
3Review the mock; cram sheets for Domains I and II2–3 h
2Cram sheets for Domains III and IV; 20 mixed questions2 h
1Light review of one-page summaries; logistics; rest1 h max

High-yield facts to lock in

Exam mechanics

  • 100 questions (85 scored), 2 hours 45 minutes, 15-minute break that locks the first half.
  • Pass at a scaled 300 (100–500). No penalty for guessing, so answer everything.
  • Multi-select questions state how many answers to pick; no partial credit.

EU AI Act essentials

  • Roles: provider, deployer, importer, distributor, authorised representative. A deployer becomes a provider if it puts its name or trademark on a high-risk system, substantially modifies it, or changes its intended purpose so that it becomes high-risk (Article 25).
  • Risk tiers: prohibited (Article 5), high-risk (Article 6, Annexes I and III), transparency obligations (Article 50), minimal risk.
  • High-risk provider duties: risk management, data governance, technical documentation, record-keeping, transparency to deployers, human oversight, accuracy and robustness, quality management system, conformity assessment, registration, post-market monitoring, serious incident reporting.
  • Deployer duties (Article 26): use in line with instructions, assign competent human oversight, monitor, keep logs, inform workers, plus a FRIA for certain deployers (Article 27).
  • Timeline after the 2026 Digital Omnibus: prohibitions 2 Feb 2025; GPAI 2 Aug 2025; Article 50 transparency 2 Aug 2026; Annex III high-risk 2 Dec 2027; Annex I high-risk 2 Aug 2028.
  • Penalties: up to €35m or 7% of turnover (prohibited practices); €15m or 3% (most other obligations); €7.5m or 1% (supplying incorrect information).

Frameworks and standards

  • NIST AI RMF: Govern (cross-cutting), Map, Measure, Manage. Voluntary and not certifiable.
  • ISO/IEC 42001: certifiable AI management system standard.
  • ISO/IEC 42005: guidance on AI system impact assessment (new in BoK v2.1).
  • OECD AI Principles: high-level principles adopted by many governments.

Lifecycle controls

  • Development: purpose definition, data provenance and lawful basis, bias testing, validation, red teaming, documentation, release gates.
  • Deployment: procurement due diligence, contract terms (data use, change notification, audit rights), human oversight design, monitoring for drift, incident response, decommissioning.
  • Agentic AI: least privilege, human approval for high-impact actions, logging, kill switches.
Don't do this in the final week: start a new course, read the full EU AI Act, re-take questions you've memorised, or study late the night before. Each one costs more than it gains.

High-yield content by domain

DomainShare of examReview in your final week
I. Foundations of AI governance16–20 questionsAI types and lifecycle terms; harms (individual, group, societal, organisational); governance roles and accountability; AI principles (OECD, UNESCO)
II. Laws, standards and frameworks19–23 questionsEU AI Act risk tiers, roles and key dates; GDPR overlaps (Articles 22, 35); US state and federal approaches; South Korea's AI Basic Act; NIST AI RMF functions; ISO/IEC 42001 and 42005
III. Governing AI development21–25 questionsData governance and provenance; impact assessments; testing, validation and red teaming; documentation (model cards, datasheets); release decisions
IV. Governing AI deployment and use21–25 questionsProcurement and vendor due diligence; deployer duties; monitoring and drift; incident response; agentic AI controls; decommissioning

Domains III and IV together make up about half the exam. If you only have time for one thing, practise lifecycle scenarios from those two domains.

What to stop doing in your last week

  • Stop reading the full EU AI Act. You need its structure, roles, tiers and key articles, not every recital. See why reading the Act isn't enough.
  • Stop collecting new resources. A new book or course in the final week adds anxiety, not points.
  • Stop doing questions without reviewing them. Twenty questions with careful review beat a hundred done quickly.
  • Stop memorising pre-2026 domain weights. BoK v2.1 shifted weight toward AI-specific laws generally; old materials that put 30% on Domain II are out of date.

Three final-week practice questions

Original Archuz questions in the AIGP style, not real exam items.

Q1

A company plans to use a vendor's AI tool to rank job applicants in the EU. What is the company's role under the EU AI Act?

  • AProvider.
  • BDeployer.
  • CDistributor.
  • DImporter.
Show answer

Correct: B — Deployer

It uses the system under its authority. It would become a provider only if it put its name on the system, substantially modified it or changed its intended purpose.

Q2

A deployed model's accuracy falls steadily as customer behaviour changes. What is this called, and what should governance require?

  • AData poisoning; isolate the model and investigate the training pipeline
  • BOverfitting; retrain with stronger regularisation before redeploying
  • CHallucination; add output filters and a user disclaimer
  • DModel drift; monitoring with thresholds that trigger review or retraining
Show answer

Correct: D — Model drift; monitoring with thresholds that trigger review or retraining

Steadily falling accuracy as real-world behaviour changes is drift, and the governance answer is monitoring with defined thresholds and actions (D). Poisoning (A) is a deliberate attack on training data, overfitting (B) shows up at training time rather than gradually afterwards, and hallucination (C) is a generative model fabricating content.

Q3

An AI agent can send emails and update records on users' behalf. Which control most directly limits harm from a successful prompt injection?

  • AA longer system prompt instructing the agent to ignore injected text
  • BLeast-privilege permissions and human confirmation for high-impact actions
  • CSwitching to a larger, more capable model
  • DAnnual security awareness training for users
Show answer

Correct: B — Least-privilege permissions and human confirmation for high-impact actions

You can't guarantee a model won't be tricked, so limiting what the agent can do bounds the damage (B). Prompt instructions (A) are routinely bypassed by injection, a larger model (C) isn't a reliable defence, and user training (D) doesn't help when the attack comes through content the agent reads.

Pattern check: if you missed any of these, note why. Was it a role confusion, a framework mix-up or a lifecycle-stage error? Your final-week revision should target the pattern, not the individual question.

A one-page memory sheet

Write these from memory on the day before the exam. If you can do it in ten minutes, you're in good shape.

  • EU AI Act roles and the four risk tiers, with one example each
  • Key dates: 2 Feb 2025, 2 Aug 2025, 2 Aug 2026, 2 Dec 2027, 2 Aug 2028
  • NIST AI RMF: Govern, Map, Measure, Manage
  • ISO/IEC 42001 (management system), 42005 (impact assessment), 23894 (AI risk management guidance)
  • GDPR Article 22 conditions and safeguards; when a DPIA is required
  • Lifecycle controls from data sourcing to decommissioning

How to use a cram guide well

  1. Active recall first. Cover a section, write out what you remember, then check. Re-reading alone feels good but sticks poorly.
  2. Annotate it with your error log. Mark each cram-guide item linked to a question you got wrong.
  3. Reduce it to one page per domain by Day 2. That page is your Day 1 review.
  4. Work through scenarios. For each lifecycle control, imagine a situation where it would be the "first" step.

Exam-day reminders

  • Confirm your ID matches your Pearson VUE profile. Online: pass the system test and clear your room.
  • Pacing: aim to finish the first half in about 75–80 minutes, with time to review flags before you submit.
  • Read the qualifier ("first", "best", "most appropriate", "NOT") before the options.

See AIGP exam day: what to expect for the full checklist.

If your final mock goes badly

If you score below about 65% on your Day 4 mock, consider rescheduling, as long as you're still within the reschedule window and your voucher's one-year validity. A short delay with targeted work usually costs less than a retake. Use the readiness checklist to decide.

Built for exactly this week

The AIGP Complete Pack's Cram Guide condenses BoK v2.1 into final-week review sheets, and its timed mock is ready for Day 4.

Get the Complete Pack

Frequently asked questions

Can I cram for the AIGP in one week?

A final week is for consolidation. Learning the whole AIGP from scratch in a week is unrealistic for most candidates, and most need 40–80 hours overall.

Should I take a practice exam the day before the AIGP?

No. Sit your last full mock about four days before, so you have time to review it. Keep the day before light.

What should I memorise for the AIGP?

Roles, EU AI Act risk tiers and key duties, the current timeline, penalty tiers, the NIST AI RMF functions, and the purposes of ISO/IEC 42001 and 42005.

How many hours a day should I study in the final week?

About 2–3 focused hours on most days, the mock day aside, is enough. More hours usually means less retention.

What are the most important AIGP topics to review in the last week?

Lifecycle governance scenarios from Domains III and IV, EU AI Act roles, tiers and dates, the NIST AI RMF functions, ISO/IEC 42001 and 42005, and GDPR overlaps such as Article 22 and DPIAs.

How many practice questions should I do in the final week?

Quality matters more than volume. One timed full-length mock early in the week plus 20–40 reviewed questions a day on weak areas is a sensible pattern for most candidates.

Sources

  1. IAPP Store: AIGP Exam
  2. Regulation (EU) 2024/1689 (EU AI Act), EUR-Lex
  3. Gibson Dunn: EU AI Act Omnibus Agreement
  4. NIST AI Risk Management Framework

Last reviewed: October 2026.