AAISM Advanced in AI Security Management: 450 Questions
450 exam-style questions across AI governance, AI risk management and AI security
You have read the AAISM review manual, you hold CISM or CISSP, and the exam is booked. The question that keeps you awake is not whether you know the material – it is whether you can pick the best of four defensible answers, ninety times, in a hundred seconds each. That is a different skill, and it is the one these tests train.
Five practice tests, 90 questions each, 450 questions in total. Every test uses the official domain proportions from ISACA's published AAISM exam content outline: AI Governance and Program Management 31 percent, AI Risk Management 31 percent, AI Technologies and Controls 38 percent, so your domain report after each attempt is directly comparable to the real score report. The questions are written from primary sources: ISACA's own exam content outline, the NIST AI Risk Management Framework and its adversarial machine learning taxonomy, MITRE ATLAS, the OWASP GenAI guidance, the NCSC and CISA secure AI development guidelines, the EU AI Act and the GDPR.
AAISM is new, launched in 2025, and it is not a renamed version of anything. That matters when you choose practice material. Older AI security question banks were written before ISACA defined this job practice, and they test a different mix – too much data science, too little program management, and nothing on the split between what an AI provider owes and what a deployer owes. AAISM sits on top of CISM or CISSP and assumes you already know general security management, so it spends its questions on what changes when the asset is a model: risk that shifts without a change ticket, incidents with no attacker, suppliers who will not show you their training data, and controls that only work if they are enforced at retrieval time.
The five tests get harder in a planned way. Test 1 checks the vocabulary one concept at a time. Test 2 puts each concept in a short situation. Test 3 is built around trade-offs, where two options are both reasonable and one condition in the stem decides. Test 4 goes into the structures underneath – which framework says what, which artefact documents which layer, what order the incident steps come in. Test 5 is the simulation: 90 questions in 150 minutes, multi-condition scenarios, domains mixed, exactly the shape of the real thing.
Every answer option has its own explanation. Choose wrongly and you are told not just that it was wrong but what that option actually describes, which is how the distinction sticks. On top of that, each question carries a full explanation of roughly a paragraph, ending in a link to the primary source so you can read further at the point where you were unsure. Nothing is left as "the correct answer is C".
The pass mark is 80 percent on tests 1 to 4 and 85 percent on the simulation. ISACA scores the AAISM exam on a scaled score, which is not the same thing as a percentage of questions answered correctly, so there is no honest way to convert one into the other. The bar here is set above the real exam on purpose, because working through questions at your own desk is easier than sitting in a proctored test centre.
What this is not: it is not a course, it is not a set of leaked or recalled exam questions, and it will not teach the body of knowledge from scratch. It is a question bank that tests you, explains every option, and shows you per domain where you still need to read. Use it alongside ISACA's own review material, not instead of it.
Groundwork: Core AI Security Management Concepts
90 questions · 160 minutes · pass mark 80%
Start here. This first test checks whether the vocabulary of AI security management is solid before you put it under pressure. Ninety questions, one concept each, spread across the three AAISM domains in the same 31/31/38 proportion as the real exam: AI Governance and Program Management, AI Risk Management, and AI Technologies and Controls. You will be asked to separate governance from management, risk appetite from risk tolerance, a policy from a standard, model inversion from membership inference, and data drift from concept drift - the distinctions that every later scenario question quietly depends on. Every answer option carries its own feedback, so a wrong choice tells you what that option actually describes, and each question closes with a fuller explanation and a link to the primary source. The clock is generous at 160 minutes and the home under no time pressure is a much easier task than choosing between four plausible options in a test centre, so the bar here is set above the one you will actually have to clear. Take this test cold, before studying, and use your domain report to decide where to read first.
In Practice: Applying Frameworks, Policies and Controls
90 questions · 155 minutes · pass mark 80%
Ninety questions that move from knowing a term to using it. Each one opens with a short situation - an approved chatbot behaving oddly, a supplier that will not release logs, a training set of uncertain origin, an agent with more tool access than it needs - and asks which concept applies or what to do about it. The pairs that get confused under time pressure are here on purpose: KPI against KRI against KCI, a datasheet against a model card, an impact assessment against a risk assessment, input filtering against output filtering, workload identity against a shared API key. The domain split matches the exam at 31/31/38 and the reporting rolls up cleanly with the other four tests. Every option is explained individually and every question ends with a source link to ISACA, NIST, MITRE ATLAS, OWASP or the regulation itself, so you can read further on anything that did not land. 155 minutes, 80% to pass.
Trade-offs: When Two Answers Look Right
90 questions · 155 minutes · pass mark 80%
This is the test where two of the four options are both defensible and one condition in the scenario decides between them. Data residency, a contract without an audit right, a regulator already asking questions, confidential source data, missing logs, limited capacity, an action that cannot be reversed - the constraint is always in the stem, and reading past it is how candidates lose marks on the real exam. Ninety questions phrased the way ISACA phrases them: BEST, MOST appropriate, FIRST, GREATEST. Expect to choose between retrieval-augmented generation and fine-tuning, between accepting and mitigating inside tolerance, between notifying a regulator and informing customers, between a contractual control and a technical one. The explanations spell out why the runner-up is a reasonable answer and what would have to change in the scenario for it to become the right one. Same domain proportions as the exam, 155 minutes, 80% to pass.
Under the Hood: Taxonomies, Artefacts and Sequence
90 questions · 135 minutes · pass mark 80%
Ninety questions on the structure underneath the vocabulary. Which framework says what, which artefact documents which layer, which mitigation belongs to which class of attack, who performs which assessment, and what order the steps come in. You will be asked to keep the NIST adversarial machine learning taxonomy apart from the ATLAS matrix, an AI system impact assessment apart from a data protection impact assessment, competence apart from awareness, a model card apart from a dataset datasheet and a bill of materials, and containment apart from eradication when the thing being eradicated is a model version. Nothing here turns on counting controls in an annex or reciting a list number - the difficulty is in telling neighbouring structures apart, not in recalling obscure trivia. Because these questions are short and precise, the clock is the tightest of the first four tests at 135 minutes for 90 questions. 80% to pass.
Exam Day: Full-Length Simulation
90 questions · 150 minutes · pass mark 85%
Ninety questions in 150 minutes: the exact length and clock of the real AAISM exam. Every question is a full scenario of several sentences with more than one condition working at once, and the domains are deliberately mixed, so a question that opens with a supplier may turn out to be a governance question, a risk question or a controls question depending on what is actually asked at the end. This is the readiness gate, and the pass mark is 85% rather than 80%. Sitting a test at your own desk, able to pause and think, is a gentler exercise than sitting in a proctored test centre with 100 seconds a question and no way back, so clearing the bar here should mean clearing it comfortably there. Save this test until you have worked through the other four and read up on whatever your domain reports flagged. If you pass it at 85% on the clock, without looking anything up, you are ready to book.
What you will practice
- Test your readiness for the ISACA AAISM exam with 450 questions in the official 31/31/38 domain
- Tell neighbouring concepts apart under pressure: appetite and tolerance, KPI and KRI, provider
- Learn from every attempt through feedback on each individual answer option plus a full
- Build up from single-concept questions to full multi-condition scenarios, ending with a
- See where you stand per domain after every attempt, so you know whether to study governance,
- Practise the ISACA question style - BEST, FIRST, MOST appropriate, GREATEST concern - where more
Exam domains covered
Who is this for
Security managers, CISOs, security architects and GRC professionals who already hold CISM or CISSP and are adding ISACA's Advanced in AI Security Management credential. Risk and compliance specialists who now own AI risk alongside information risk, and who need to be fluent in the AI Act's split between providers and deployers, in impact assessments, and in what an AI incident is when nobody attacked anything. Cloud and platform architects being asked to sign off retrieval architectures and agent permissions. Auditors and consultants who advise on AI governance and want to check their knowledge against the security management side of it rather than the audit side - if independent assurance over AI is your actual job, ISACA's AAIA credential is the closer fit and these tests are not written for it. Also useful to anyone preparing for the exam who has read the review manual once and cannot yet tell whether that reading has stuck. Note what this is and is not: a question bank that tests and explains. It will show you your gaps, teach you the distinctions that decide exam questions, and get you used to the pace. It will not teach the body of knowledge from scratch, and it is not a set of real exam questions - every question here is written from the published exam content outline and from primary sources.
Requirements
- An active CISM or CISSP certification, which ISACA requires before you may register for the
- Working knowledge of information security management: risk treatment, policy structure,
- Familiarity with how AI systems are built and run at a conceptual level - training and
- These are practice tests, not a course. They are designed to be used alongside ISACA's own
Free exam guide
Not sure what to expect? Read how the exam is built, what it covers and how to prepare.