AAIA Practice Test

Skills this practice test covers

• Evaluate whether an organization has named owners for AI risk, controls, decisions and standards.
• Assess training data for appropriateness, bias and privacy concerns before a model relies on it.
• Review change control and configuration management for AI models and their settings.
• Judge whether human oversight and output monitoring fit the impact of an AI decision.
• Match AI threats such as data poisoning and model extraction with the controls that counter them.
• Pick a sampling method for an AI audit, gather the evidence and write up the findings.

3 domains · 17 topics · 254 of 262 questions sorted by domain

Requirements

• Required: the CISA certification, or another credential ISACA accepts, such as the CIA or CPA.
• A CIA holder must also work in an IT audit or IT advisory role.
• Recommended: confirm on ISACA's current list that the credential held qualifies before applying.

About the AAIA exam

How to earn the certification

Earns

Advanced in AI Audit (AAIA)Advanced

AAIA stays active through continuing professional education (CPE).

Before

• CISA Certified Information Systems Auditor (CISA)

  The usual qualifying credential. Every CISA holder meets the AAIA entry requirement.

How to prepare for AAIA

1. 1

   Pair each objective with a control

   Almost every objective starts with "Evaluate" or "Assess". For each one, write down the control an auditor expects to find and the evidence that proves it works.

2. 2

   Learn the AI attack names

   Threats and Vulnerabilities Specific to AI uses exact names: data poisoning, model inversion, membership inference, model extraction and backdoor attacks. Drill them in flashcard mode until each one is distinct.

3. 3

   Follow one model end to end

   AI Operations is the largest domain. Trace one model from its training data through MLOps, change control, monitoring and incident response, and note the audit question at each stage.

4. 4

   Argue each sampling choice

   Take one population, such as a month of loan decisions made by a model. Decide when judgmental sampling, dynamic sampling or full population testing fits, and say why.

5. 5

   Close with a timed full run

   Take the AAIA practice test in Timed Mode once all three domains feel familiar. Ask Brainy AI Assistant to explain each missed question and the control it points to.

   Start a timed AAIArun

Who AAIA is for

• IT auditors who hold CISA and now find AI systems in their audit plans.
• Internal auditors with the CIA credential who work in IT audit or IT advisory.
• External auditors who hold a CPA and review how their clients use AI.
• Audit managers and risk consultants who advise on AI governance.

Typical job titles: AI audit manager · AI risk consultant · AI governance specialist · IT auditor · Internal auditor · AI assurance director

AAIA glossary

18 terms from the exam objectives and the vendor's official material, each defined in a sentence.

AI Act (EU)

The European Union law that sets rules for AI systems according to the level of risk they carry.

Backdoor Attacks

Attacks that hide a trigger in a model, so that the model behaves normally until an input contains the trigger.

CMDB for AI Assets

A configuration management database that records AI models, data sets and their settings as managed items.

Continuous Auditing

Running audit tests automatically and often, so that a control failure shows up soon after it happens.

Cross-Validation

Training and testing a model several times on different splits of the data, to check that its results hold.

Data Lineage

The record of where data came from and every change it went through before it reached the model.

Data Poisoning

An attack that slips false or harmful records into training data, so that the model learns the wrong behavior.

Explainability Risk

The risk that no one can show why a model reached a decision, which makes the decision hard to check or defend.

Full Population Testing

Testing every item in a data set, usually with analytics tools, instead of a sample.

Holdout Validation

Testing a model on data that was set aside at the start and never used to train it.

Human-in-the-Loop

An oversight design in which a person must review or approve an AI decision before it takes effect.

Human-on-the-Loop

An oversight design in which the AI acts on its own while a person watches it and can step in.

Judgmental Sampling

Choosing audit samples by the auditor's judgment, for example the riskiest items, instead of at random.

Known Error Database

A store of problems whose cause is known, with the documented workaround for each one.

Membership Inference

An attack that tests whether one specific record was part of a model's training data.

Model Extraction

An attack that sends many queries to a model and uses the answers to build a copy of it.

Model Inversion

An attack that uses a model's outputs to rebuild sensitive details of the data the model was trained on.

Model Registry

A central store of models with their versions, owners, approval status and documentation.