ISACA AAISM Sample Questions & Practice Test

ISACA Advanced in AI Security Management (AAISM) is a focused route for professionals managing AI security risk, controls, monitoring, and incident response.

These original questions preview the decision style a full IT Mastery route should use. They are not official ISACA exam questions.

What this route should test

• identifying AI-specific threats such as prompt injection, data leakage, model abuse, and insecure integrations
• selecting controls for data, identity, monitoring, and model-change governance
• responding to AI incidents without destroying evidence
• aligning AI security controls with enterprise risk and business use

Sample Exam Questions

Question 1

Topic: prompt injection

A customer-facing AI assistant follows user instructions that override system rules and expose internal policy text. What is the primary risk?

• A. Disk fragmentation
• B. Prompt injection or instruction manipulation
• C. Printer misconfiguration
• D. Spreadsheet formatting

Best answer: B

Explanation: Prompt injection attempts to manipulate the model into ignoring intended instructions or revealing restricted information. Mitigations include input controls, grounding, output filtering, and privilege boundaries.

Question 2

Topic: data leakage

Developers want to paste production customer records into a public AI tool for troubleshooting. What should security require first?

• A. Larger prompts
• B. No review because the tool is convenient
• C. Disabling all logs
• D. Approved data-use controls, masking or synthetic data where appropriate, and vendor-risk review

Best answer: D

Explanation: Sensitive data should not be placed into AI tools without approved handling, privacy, contractual, and security controls.

Question 3

Topic: model access

An AI model can call backend tools that update customer records. What control is most important?

• A. Tool access limited by identity, authorization, audit logs, and approved action boundaries
• B. Anonymous write access
• C. No monitoring because AI is automated
• D. Public administrator tokens

Best answer: A

Explanation: When AI can invoke tools, security must control what actions are allowed, under which identity, with what authorization, and with what evidence.

Question 4

Topic: AI incident response

An AI agent sent unauthorized account-change instructions to several users. What should happen first?

• A. Delete all conversations
• B. Rename the model
• C. Contain the agent, preserve logs, assess impacted users, and follow the incident plan
• D. Ignore it because AI output is not security-relevant

Best answer: C

Explanation: AI incidents need containment and evidence preservation. Logs, prompts, outputs, tool calls, and data access records can be critical.

Question 5

Topic: threat modeling

Which scenario should be included in an AI threat model?

• A. A monitor changes brightness
• B. A user tries to make the model reveal restricted system instructions
• C. A printer runs out of paper
• D. A user chooses a dark theme

Best answer: B

Explanation: AI threat models should include prompt manipulation, data exfiltration, model abuse, tool misuse, and insecure integrations.

Question 6

Topic: supply chain

A business team wants to adopt a third-party model API. What should security review?

• A. Only the vendor logo
• B. Only the font on the vendor website
• C. Whether the API name is short
• D. Data handling, contractual terms, model-use limits, logging, availability, and vendor controls

Best answer: D

Explanation: AI supply-chain risk includes data processing, confidentiality, availability, model behavior, contractual protections, and security posture.

Question 7

Topic: monitoring

After deployment, the model begins producing responses outside approved policy. What monitoring signal matters most?

• A. Output-quality and policy-violation indicators tied to alerting and response
• B. Office attendance
• C. Keyboard inventory
• D. The length of the model name

Best answer: A

Explanation: AI security monitoring should detect policy violations, unsafe outputs, tool misuse, abnormal access, and drift from approved behavior.

Question 8

Topic: least privilege

An AI service account has broad read and write access across systems it does not use. What is the best remediation?

• A. Add more permissions
• B. Turn off authentication
• C. Reduce permissions to required data and actions, then monitor use
• D. Share the service account with all users

Best answer: C

Explanation: Least privilege applies to AI services and tool integrations. Excessive permissions increase blast radius if the model, prompt, or integration is abused.

Question 9

Topic: model change control

A vendor silently changes the model version used by a regulated workflow. What should the organization require?

• A. No change records
• B. Version-change notification, testing, approval, and rollback planning
• C. Blind trust in every update
• D. Deleting the workflow

Best answer: B

Explanation: Model version changes can affect behavior and risk. Secure management requires notification, validation, approval, and contingency planning.

Question 10

Topic: human oversight

An AI system recommends high-impact decisions. What control reduces risk?

• A. Fully anonymous decisions
• B. No appeal or review process
• C. Automatic deletion of rationale
• D. Human review for defined high-risk decisions and documented override criteria

Best answer: D

Explanation: Human oversight can provide accountability and risk control when outputs affect people, finances, compliance, or safety.

Question 11

Topic: red teaming

Why run adversarial tests against an AI assistant before launch?

• A. To discover unsafe behavior, policy bypasses, data leakage, and tool misuse scenarios
• B. To guarantee no future incident can occur
• C. To replace all monitoring
• D. To choose a product name

Best answer: A

Explanation: Red-team testing helps identify weaknesses before release. It reduces risk but does not eliminate the need for governance and monitoring.

Question 12

Topic: governance

Which artifact best supports secure AI operations over time?

• A. A list of office plants
• B. A screenshot of one chat response
• C. A register of AI systems, owners, data sources, use cases, risks, controls, and review cadence
• D. A marketing slide only

Best answer: C

Explanation: An AI inventory or register helps security teams understand exposure, assign accountability, and review controls as systems change.