Isaca AAISM Online Practice Questions and Exam Preparation

Isaca AAISM Online Questions & Answers

• Question 261:

  An organization plans to leverage AI in the software development process to speed up coding. Which of the following should the information security manager do FIRST?

  A. Conduct an impact assessment
  B. Train developers to verify AI output
  C. Update the security policy to include AI controls
  D. Perform a cost-benefit analysis
  A. Conduct an impact assessment

  ---

  AAISM guidance specifies that before introducing AI into any business or technical workflow, an AI Impact Assessment must be conducted early to determine potential risks, privacy implications, misuse scenarios, governance gaps, and required security controls. This aligns with the principle that AI adoption must begin with governance and risk identification, not training or policy modification. Training developers (B) is important but occurs after identifying risks. Updating policies (C) is also downstream of the assessment. Cost-benefit analysis (D) supports business justification but does not address security.

  References: AAISM Study Guide - AI Governance; Impact Assessment Required Before Deployment.

• Question 262:

  A global organization experienced multiple incidents of staff pasting confidential data into public chatbots. Which action is MOST important to reduce short-term risk?

  A. Deliver role-based, scenario-driven AI security training mapped to job functions
  B. Require employees to complete an annual generic phishing and deepfake module
  C. Publish an AI acceptable use policy and collect signatures
  D. Block access to public LLMs at the network perimeter
  A. Deliver role-based, scenario-driven AI security training mapped to job functions

  ---

  AAISM states that the most effective short-term mitigation for unintentional data leakage into public AI tools is targeted, role-based AI security awareness, focused on: what data cannot be entered consequences of leakage real-world scenarios employees face An acceptable-use policy (C) is necessary but insufficient alone. Blocking LLMs (D) may reduce access but does not change user behavior and may cause shadow AI usage. Generic phishing training (B) does not address AI misuse risks.

  References: AAISM Study Guide - AI Security Awareness; Human-Centric Data Leakage Prevention.

• Question 263:

  Which of the following is the MOST effective action an organization can take to address data security risk when using generative AI features in an application?

  A. Rely on the AI provider's independent third-party audit reports for assurance
  B. Establish policies and awareness training for acceptable use of AI
  C. Require opt-out provisions for data usage in service agreements
  D. Establish guidelines and best practices with third parties for intellectual property ownership
  C. Require opt-out provisions for data usage in service agreements

  ---

  AAISM directs organizations to manage third-party AI risks through contractual and technical controls that explicitly govern data use, retention, training/fine-tuning, isolation, and deletion. The most effective data-security action when consuming generative AI features is to require enforceable opt-out provisions that prohibit the provider from using the organization's data for training or secondary purposes and that mandate retention limits and secure deletion. Third-party audit reports (A) provide assurance but do not guarantee provider behavior for your specific data; awareness policies (B) are necessary but insufficient to control external processing; IP ownership guidelines (D) address legal rights, not data-security risk.

  References: AI Security Management. (AAISM) Body of Knowledge - Third-Party/Procurement Controls; Data Use and Retention Clauses; Training/Fine-tuning Opt-Out; Secure Deletion and Purpose Limitation.

• Question 264:

  Personal data used to train AI systems can BEST be protected by:

  A. Erasing personal data after training
  B. Ensuring the quality of personal data
  C. Anonymizing personal data
  D. Hashing personal data
  C. Anonymizing personal data

  ---

  AAISM guidance on privacy-preserving AI highlights anonymization as the most effective means of protecting personal data used in training. By irreversibly removing or masking identifiable attributes, anonymization ensures that training data cannot be linked back to individuals, thereby meeting key privacy obligations under laws such as GDPR. Erasing data after training may limit exposure but does not protect it during the training process. Ensuring data quality improves accuracy but does not mitigate

  privacy risk. Hashing protects data integrity but does not guarantee anonymity, as hashes can sometimes be reversed or correlated. Therefore, anonymization is the recommended control for protecting personal data in AI training.

  References:

  AAISM Study Guide - AI Technologies and Controls (Privacy-Preserving Methods)

  ISACA AI Security Management - Data Anonymization Practices

• Question 265:

  Which of the following MOST effectively secures ongoing stakeholder support for AI initiatives?

  A. Quantifying and communicating the value of AI solutions
  B. Conducting periodic staff training
  C. Addressing and optimizing AI-related risk
  D. Developing and monitoring an AI strategic roadmap
  A. Quantifying and communicating the value of AI solutions

  ---

  AAISM governance guidance emphasizes that stakeholder buy-in is sustained when the measurable value of AI initiatives is clearly communicated. Value demonstrations include: . improved efficiency . reduced cost . reduced risk . business growth Training (B) and risk optimization (C) are important but do not guarantee stakeholder support. A roadmap (D) guides planning but does not secure buy-in.

  References: AAISM Study Guide - AI Governance; Stakeholder Engagement and Value Communication.

• Question 266:

  Which of the following datasets is used to tune hyperparameters?

  A. Validation
  B. Test
  C. Configuration
  D. Training
  A. Validation

  ---

  Per AAISM's ML lifecycle controls, hyperparameter tuning is performed on the validation set, reserving the test set strictly for final, unbiased performance estimation. The training set is used to fit parameters; the validation set guides model selection and hyperparameter optimization; the test set is untouched until the end to prevent leakage and optimistic bias. "Configuration" is not a dataset type in the lifecycle split. References:. AI Security Management (AAISM) Body of Knowledge: Model Development Controls-Data Splitting and Evaluation Integrity. AAISM Study Guide: Overfitting Avoidance; Validation vs. Test Separation; Leakage Prevention. AAISM Mapping to Standards: Evaluation Integrity-Hold-out Protocols and Tuning Practices

• Question 267:

  A large corporation has received an influx of sophisticated credential-phishing emails and wants to leverage an AI solution to detect and quarantine these messages before they reach employees. Which of the following blue-team AI features is BEST suited to this task?

  A. Large language model (LLM)
  B. Natural language processing (NLP)
  C. Natural language generation (NLG)
  D. Retrieval-augmented generation (RAG)
  B. Natural language processing (NLP)

  ---

  For pre-delivery phishing detection and classification, the most appropriate capability is NLP-tokenization, feature extraction, semantic similarity, and supervised classifiers (e.g., transformer-based classifiers) tuned on phishing corpora and indicators. NLP models score messages and drive automated quarantine policies. An LLM (Option A) is a model type, not a specific blue-team feature; NLG (Option C) is for generation, not detection; RAG (Option D) augments responses with retrieved knowledge but does

  not by itself optimize classification and quarantine of phishing emails.

  References:

  AAISM Body of Knowledge: Defensive AI Use Cases; Text Classification Pipelines for Security Operations; Email Security and AI-Driven Triage.

  AAISM Study Guide: NLP for Threat Detection; Model Evaluation for Precision/Recall in Security Classifiers; SOC Integration and Automated Containment.

• Question 268:

  When robust input controls cannot prevent prompt injections in an LLM, what is the BEST compensating control?

  A. Fine-tune the system to validate inputs
  B. Implement identity and access management (IAM)
  C. Conduct human reviews of AI system inputs
  D. Review and annotate the AI system's outputs
  D. Review and annotate the AI system's outputs

  ---

  AAISM identifies output review and annotation as the most practical compensating control when robust input validation cannot be applied. Output moderation detects: . maliciously influenced responses . unsafe outputs . security-policy violations IAM (B) does not mitigate prompt injection itself. Human review of inputs (C) is unrealistic at scale. Fine-tuning (A) cannot guarantee full prevention.

  References: AAISM Study Guide - Generative AI Safeguards; Output Moderation Controls.

• Question 269:

  A SaaS-based LLM system has risks including prompt injection, data poisoning, and model exfiltration. What is the BEST way to ensure consistent risk treatment?

  A. Apply control baselines from a recognized industry standard
  B. Implement an AI threat control matrix mapping threats to controls and assurance
  C. Focus on post-deployment red teaming
  D. Rely on vendor audit reports and SLAs
  B. Implement an AI threat control matrix mapping threats to controls and assurance

  ---

  AAISM highlights the importance of an AI threat-control matrix that:

  maps threats (e.g., poisoning, exfiltration, injection)

  identifies required controls

  defines assurance/testing activities

  ensures consistent coverage across all AI components

  Control baselines (A) are useful but not AI-specific. Red teaming (C) is reactive and not comprehensive.

  Vendor reports (D) are supplemental, not sufficient.

  References: AAISM Study Guide - AI Risk Treatment; Threat-Control Mapping.

• Question 270:

  Which strategy BEST ensures generative AI tools do not expose company data?

  A. Conducting an independent AI data audit
  B. Implementing a solution prohibiting input of sensitive data
  C. Testing AI tools before implementation
  D. Ensuring AI tools comply with local regulations
  B. Implementing a solution prohibiting input of sensitive data

  ---

  AAISM identifies the strongest immediate control for preventing data leakage into generative AI systems as technically restricting or blocking user entry of sensitive data. Audits (A) are retrospective. Testing tools (C) does not prevent user error. Regulatory compliance (D) does not stop operational leakage.

  References: AAISM Study Guide - Data Leakage Prevention and Input Restriction Controls.