This is not a replacement for the official study guide. Please read the official study guide first — use this exam prep only to reinforce or practice your understanding.
Free AAISM exam practice questions with answers and explanations, organized by domain and part.
Domain 3: AI Technologies and Controls
Q481 Which security control framework MOST specifically catalogs adversary tactics and techniques against AI systems?
- A. ISO 9001
- B. MITRE ATLAS
- C. COSO
- D. PCI DSS
Answer: B
MITRE ATLAS provides a knowledge base of adversary tactics and techniques against AI-enabled systems. PCI DSS covers payment card security, ISO 9001 quality management and COSO internal control.
Domain 3: AI Technologies and Controls
Q482 Which resource lists the most critical security risks for LLM applications and is MOST useful for threat awareness?
- A. The ITIL service catalog
- B. The COBIT 5 process list
- C. The ASQ quality handbook
- D. The OWASP Top 10 for LLM applications
Answer: D
OWASP publishes top risks for LLM applications such as prompt injection and insecure output handling. ITIL, ASQ and COBIT do not focus on LLM security.
Domain 3: AI Technologies and Controls
Q483 An organization applies zero trust to its AI platform. Which principle is MOST central?
- A. Trust all internal traffic because it passes the perimeter firewall
- B. Never trust, always verify every request regardless of network location
- C. Grant permanent privileges to administrators for convenience
- D. Rely on perimeter firewalls as the only line of defense
Answer: B
Zero trust requires continuous verification of users, devices and requests irrespective of location. Trusting internal traffic or relying on a perimeter alone contradicts it.
Domain 3: AI Technologies and Controls
Q484 Which access control practice is MOST appropriate for model repositories and training environments?
- A. Shared team accounts so that activity is easy to hand over between analysts during shift changes
- B. Anonymous read access to simplify onboarding of analysts
- C. Role-based access with least privilege and multifactor authentication
- D. Hard-coded credentials in notebooks for repeatable jobs
Answer: C
Strong identity, role-based access, least privilege and MFA limit unauthorized access to models and data. Shared accounts and embedded credentials remove accountability.
Domain 3: AI Technologies and Controls
Q485 An auditor requests evidence of which data and model versions produced a decision. Which capability is MOST needed?
- A. Reduced token usage for each customer prompt
- B. Higher model accuracy on the held-out validation set, as measured during the final round of model testing
- C. Faster inference speed for the production endpoint
- D. Audit logs and traceability across data, model and decision records
Answer: D
Traceability through logs of data, model and decision lineage provides audit evidence. Accuracy, speed and token use do not prove lineage.
Domain 3: AI Technologies and Controls
Q486 An employee uses an unapproved public chatbot to process client data. Which control is MOST effective to detect this?
- A. Rotating office seats on a regular schedule among teams
- B. Network and application monitoring with an approved AI tool catalog
- C. Installing a suggestion box for staff ideas on AI tools
- D. Printing policy reminders and posting them near the workstations
Answer: B
Monitoring traffic and applications against an approved catalog identifies shadow AI usage. Reminders alone do not detect use, and seating or suggestion boxes are unrelated.
Domain 3: AI Technologies and Controls
Q487 What is the MOST effective approach to reduce shadow AI over time?
- A. Ignore usage until an incident forces a policy response
- B. Ban all AI tools outright without offering any approved alternative, and rely on disciplinary action to enforce the ban
- C. Provide approved AI tools and clear acceptable use guidance alongside monitoring
- D. Rely on employees' voluntary self-reporting as the only control
Answer: C
Giving approved options and guidance, with monitoring, channels demand safely. Blanket bans without alternatives push usage underground, and ignoring it or relying on self-reporting leaves blind spots.
Domain 3: AI Technologies and Controls
Q488 A company applies input validation and output filtering to an LLM application. Which risk is MOST directly reduced?
- A. Power outage affecting the data center
- B. Physical theft of laptops used by developers
- C. Prompt injection and insecure output handling
- D. Hardware failure in the servers hosting the model
Answer: C
Validating inputs and filtering outputs limits manipulation of prompts and unsafe output handling. Hardware, power and theft risks need different controls.
Domain 3: AI Technologies and Controls
Q489 A team hires specialists to attack their own AI system to find weaknesses before release. Which practice does this BEST describe?
- A. A change freeze before the quarterly release
- B. Capacity planning for the inference cluster
- C. AI red teaming
- D. Penetration testing of the payroll application
Answer: C
Red teaming simulates adversaries against AI systems to uncover weaknesses such as jailbreaks and data leakage. Payroll penetration, capacity planning and change freezes are not adversarial testing.
Domain 3: AI Technologies and Controls
Q490 Which control is MOST appropriate to secure the AI supply chain for third-party models?
- A. Disable scanning to avoid false positives in the build pipeline
- B. Verify provenance, scan models and dependencies, and apply SBOM or AIBOM practices
- C. Download any popular model from public sources and deploy it immediately, to avoid delays in the project schedule
- D. Trust models by download counts and community popularity alone
Answer: B
Verifying provenance and scanning models and dependencies, supported by bills of materials, reduces supply chain compromise. Deploying unvetted models or trusting popularity invites risk.