Last updated:
AAISM preparation guide
AAISM Domain 3: AI technologies and controls
Direct answer
Domain 3 asks whether you can choose the control that fits each stage of an AI system's life: design, data, training, validation, deployment, monitoring and retirement. It is the largest domain, at 38%, and the one where technical understanding matters most.
Questions describe a stage of the lifecycle and a weakness. The best control is the one that belongs at that stage, not one from earlier or later that would catch the problem too late.
Privacy, ethics, trust and safety are controls here, not slogans: explanations people can use, human review where it is required, and limits on what a model may disclose.
Start AAISM baselineDomain 3 subareas (38% of the exam)
| Subarea | What the questions test |
|---|---|
| 3A AI security architecture and design | Types of AI and secure architecture and design choices |
| 3B AI life cycle (for example model selection, training and validation) | What each lifecycle phase requires before the next begins |
| 3C Data management controls | Data governance and data security for training and use |
| 3D Privacy, ethical, trust and safety controls | Privacy, ethics, trust and safety in practice |
| 3E Security controls and monitoring | Security controls, AI security awareness training and continuous monitoring |
Original teaching example
A fraud model is retrained monthly on new transaction data. A pipeline change could let mislabelled records into training without anyone noticing. Which control fits this stage?
Label-quality checks before training, such as sampled label review and comparison of label distributions with earlier months, with lineage and an owner who reviews exceptions. Monitoring the deployed model is still useful, but it acts later in the life cycle and complements these preventive checks.
Validation evidence must match the intended use
An overall accuracy figure can hide failure in the cases that matter. Look for evidence broken down by the conditions the model will meet in service.
Monitoring is a control, not an afterthought
Drift, misuse and new attack techniques appear after deployment. Expect questions on what to monitor, who reviews it and what triggers action.
Why Domain 3 needs the most study time
It is 38% of the exam and has no direct CISM counterpart.
CISM and CISSP judgment carries into Domains 1 and 2. Domain 3 adds how AI systems are built, tested and run, so plan most new study time here.
Worked examples by subarea
One original scenario per subarea, with the stronger answer.
- 3A: A customer-service agent built on a language model can call internal tools, and a crafted prompt could make it issue refunds. Stronger: let the model call only the tools the task needs, with the user's own permissions, and require confirmation outside the model for high-impact actions.
- 3B: A model in validation was trained without data from regions it will serve. Stronger: validate it on representative data from those regions before sign-off; if it falls short, improve the data, retrain and revalidate, or narrow the intended use.
- 3C: Training data for a recommendation model starts arriving from a new pipeline with no checks, so corrupted or poisoned records could change what the model learns without any visible error. Stronger: data validation and lineage before training, with an owner for exceptions.
- 3D: A lending model gives declines with no reasons. Stronger: explanations reviewers and customers can use, and human review where required.
- 3E: A deployed model has no drift monitoring. Stronger: thresholds, an owner and a response when a threshold is crossed.
Sources and fact check
Source checked: 2026-10-12
- Study methods and examples are independent CertArc teaching guidance. ISACA's pages are the authority for exam rules and fees, which can change.
- Practice results do not convert to an official score or guarantee a pass. CertArc is not affiliated with or endorsed by ISACA.
Frequently asked questions
How technical is Domain 3?
You need to understand how AI systems are built and tested well enough to choose a control; you are not asked to build one.
Which lifecycle model does AAISM use?
Follow ISACA's outline; CertArc's questions use design, data, training, validation, deployment, monitoring and retirement.
How is Domain 3 weighted?
38% of the exam, the largest domain.
How many subareas does Domain 3 have?
Five, 3A to 3E, in ISACA's outline.
Is privacy part of Domain 3?
Yes. Subarea 3D covers privacy, ethical, trust and safety controls.
