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 2: AI Risk Management
Q381 Which mitigation is MOST effective against bias amplification?
- A. Faster GPUs and larger batch sizes during every training run, so that retraining completes in less time
- B. Tighter firewall rules around the training environment
- C. Bias detection audits, diverse training datasets and fairness audits
- D. Additional storage capacity for archived model versions
Answer: C
Bias is addressed through audits, diverse training data and fairness testing. Hardware speed, firewall rules and storage do not change fairness outcomes.
Domain 2: AI Risk Management
Q382 A public LLM endpoint is hit by high volumes of adversarial queries. Which mitigation is MOST suitable?
- A. Rate limiting
- B. Increasing token limits
- C. Disabling authentication
- D. Adding more training data
Answer: A
Rate limiting helps against adversarial or DoS-style query volume. Raising token limits or removing authentication worsens exposure, and more training data does not stop abuse.
Domain 2: AI Risk Management
Q383 Which set of mitigations is MOST suitable for model inversion?
- A. Strong cryptography, access management and query parameterization
- B. Larger context windows and longer conversation history limits
- C. Public awareness campaigns and strict media forensics procedures
- D. Watermarking of outputs combined with digital signatures only
Answer: A
Model inversion risk is reduced by strong cryptography, identity and access management and query controls. Awareness campaigns and watermarking address other threats.
Domain 2: AI Risk Management
Q384 Which mitigation is MOST appropriate for explainability failures?
- A. Longer password policies and shorter rotation periods for administrators of the AI platform and its hosting accounts
- B. Faster hardware and bigger caches for the inference service
- C. Logging, alerting and monitoring with auditing and regulatory adherence
- D. More marketing communications explaining model improvements
Answer: C
The manual lists logging, monitoring, auditing and assessments and adherence to regulation for compliance and explainability failures. Hardware and password length do not make decisions explainable.
Domain 2: AI Risk Management
Q385 Consumers distrust an AI product because they worry it deceives them. Which mitigation is MOST relevant?
- A. Larger servers and added capacity for the customer-facing service
- B. Cost cutting across the model development and test environments
- C. Consumer transparency, ethical AI principles and opt-in policies
- D. Reduced logging so that less customer activity is recorded
Answer: C
For consumer distrust and deception concerns, the manual suggests transparency, ethical AI principles and opt-in policies. Capacity and cost steps do not rebuild trust.
Domain 2: AI Risk Management
Q386 An AI-related scandal damages an enterprise's brand. Which mitigation is MOST suitable in advance?
- A. Longer log retention for all production AI systems
- B. Additional perimeter firewalls around the AI hosting network
- C. Crisis communication plans and proactive reputation management
- D. Increased storage capacity for archived model outputs
Answer: C
Brand and reputational risk is mitigated through crisis communication plans and proactive reputation management. Logs, firewalls and storage do not manage public perception.
Domain 2: AI Risk Management
Q387 Which mitigation set is MOST suitable for AI usability challenges that affect customers?
- A. Explainable AI models, interface testing and user feedback loops
- B. Additional firewalls between the interface and the model host
- C. Faster backups of model artifacts to offsite storage
- D. Stronger encryption on stored training data and model files
Answer: A
Usability challenges are addressed through explainable models, UI/UX testing and feedback loops. Encryption, backups and firewalls do not improve usability.
Domain 2: AI Risk Management
Q388 Which mitigation is MOST suitable for a model backdoor attack?
- A. Continuous monitoring and AI fairness and bias testing
- B. Larger monitors and ergonomic workstations for the model developers
- C. Quarterly social events to improve team culture and morale
- D. Reduced network bandwidth to slow down any hidden activity
Answer: A
Backdoor attacks are mitigated by continuous monitoring and fairness and bias testing to detect unexpected behavior. The other options have no effect on hidden triggers.
Domain 2: AI Risk Management
Q389 Under the EU AI Act framing in the manual, which responsibility BEST belongs to the AI provider rather than the deployer?
- A. Controlling which data is supplied to the system as input
- B. Using the system strictly according to the supplier's written instructions, and nothing beyond those instructions
- C. Assigning qualified staff to oversee the system during everyday operation
- D. Performing conformity assessments for high-risk systems and maintaining technical documentation
Answer: D
Providers ensure high-risk systems comply with law, perform conformity assessments and keep documentation. Deployers use systems as directed, assign human oversight and control the data provided. Mixing these roles creates gaps in accountability.
Domain 2: AI Risk Management
Q390 A company subscribes to a third-party AI tool and feeds its own customer records into it. Which role does the company hold and what is its MOST direct duty?
- A. Provider, completing the conformity assessment before the tool is sold and keeping the technical documentation up to date
- B. Deployer, controlling what data is provided to the system and ensuring human oversight
- C. Regulator, certifying the tool for the market on the company's behalf
- D. Auditor, issuing independent assurance over the vendor's model design
Answer: B
The company uses the solution, so it is the deployer and it controls the data entered and assigns oversight. Conformity assessment belongs to the provider. Regulators and auditors play separate roles.