Free CompTIA SecAI+ Practice Questions

Q1.A security team deploys AI-enhanced honeypots across their network that mimic database servers containing synthetic data. An attacker interacts with a honeypot and attempts SQL injection. The AI system enriches the alert with context about the attacker's techniques and cross-references the source IP against threat intelligence feeds. Which of the following describes the MOST valuable intelligence this AI-enhanced deception system provides?

• A.Detailed attacker TTP mapping correlated with threat intelligence to identify the likely threat actor and predict their next actions
• B.The attacker's IP address which can be permanently blocked at the firewall to prevent future attacks
• C.The specific SQL injection payload used which proves the attacker has advanced technical skills
• D.The success rate of honeypot deployments as a percentage of total network assets

Domain: 3.0 AI-Assisted Security

Q2.An organization is configuring AWS SageMaker for a sensitive model training job. The security team requires that training data and model artifacts are encrypted with keys the organization fully controls and that all training occurs within the organization's VPC. Which combination of AWS services BEST meets these requirements? (Select TWO.)

• A.AWS KMS Customer Managed Keys for encrypting all training data and model artifacts
• B.VPC configuration with private subnets and security groups for SageMaker training jobs
• C.S3 Transfer Acceleration for faster data upload to the training bucket
• D.CloudWatch alarms set on SageMaker training job completion metrics
• E.Deploying SageMaker endpoints in public subnets for easier access

Domain: 2.0 Securing AI Systems

Q3.In a federated learning system, multiple hospitals contribute to training a shared diagnostic model. A security researcher discovers that by analyzing the gradient updates sent by one hospital, they can reconstruct individual patient records from that hospital's data. Which of the following attacks does this describe?

• A.Gradient leakage attack
• B.Model inversion attack
• C.Byzantine attack on the aggregation server
• D.Membership inference attack

Domain: 1.0 Basic AI Concepts

Q4.An organization wants to map their AI governance controls to the EU AI Act classification requirements. Their AI system is used for employee performance evaluation and is classified as high-risk under the Act. Which of the following obligations does this high-risk classification impose? (Select TWO.)

• A.Implementing mandatory human oversight mechanisms that allow humans to monitor and intervene in system decisions
• B.Maintaining comprehensive technical documentation including design specifications and risk management measures
• C.Open-sourcing the model weights and training data for public auditing
• D.The system is prohibited from deployment under any circumstances
• E.Registering for a voluntary self-regulation program without government oversight

Domain: 4.0 AI GRC

Q5.An organization uses AI to automatically correlate alerts from multiple security tools and group them into unified incidents. The AI system identifies that five seemingly unrelated alerts from the firewall, EDR, SIEM, email gateway, and identity provider are all part of a single coordinated attack. Which AI capability is MOST responsible for this correlation?

• A.Cross-source pattern recognition that identifies temporal and contextual relationships between alerts from disparate security tools
• B.Signature-based matching that compares each alert against a database of known attack signatures
• C.Log compression algorithms that reduce the volume of data analysts need to review
• D.Automated vulnerability scanning that identifies weaknesses exploited by the attacker

Domain: 3.0 AI-Assisted Security

Q6.A security architect is evaluating the risk of input transformation defenses against adversarial attacks on an image classification model. The defense applies JPEG compression and spatial smoothing to all inputs before they reach the classifier. Which limitation of this defense approach should the architect be MOST aware of?

• A.Adaptive adversaries can craft perturbations that survive the specific transformation by incorporating it into their attack optimization process
• B.Input transformation defenses always introduce unacceptable latency for real-time classification systems
• C.The defense requires complete retraining of the classification model after each transformation change
• D.Input transformation techniques can only be applied to grayscale images and not to color photographs

Domain: 2.0 Securing AI Systems

Q7.A data scientist is evaluating a malware classifier and wants to ensure robust performance estimates that minimize the impact of how the data happens to be split. She divides the dataset into five equal parts, trains the model on four parts, tests on the remaining part, and repeats this process five times using a different part for testing each time. Which evaluation technique is she using?

• A.K-fold cross-validation
• B.Holdout validation
• C.Bootstrapping
• D.Stratified random sampling

Domain: 1.0 Basic AI Concepts

Q8.A financial services organization uses an AI model for automated loan decisions. Regulators require that every decision must be explainable to the affected individual in plain language. The organization currently uses a deep neural network that achieves the highest accuracy. Which approach BEST balances regulatory compliance with model performance?

• A.Replacing the neural network with a simple linear regression model that is inherently interpretable but less accurate
• B.Implementing post-hoc explainability techniques such as SHAP or LIME that generate human-understandable explanations for each decision while retaining the neural network's accuracy
• C.Deploying the neural network without explanations and accepting the regulatory risk
• D.Providing the raw neural network output probabilities to applicants without any interpretation

Domain: 4.0 AI GRC

Q9.A security team uses AI to automate the initial phase of incident response. When a potential security incident is detected, the AI system automatically collects relevant log data from affected systems, captures memory snapshots, preserves network packet captures, and creates a timeline of events. All evidence is stored with cryptographic hashes for integrity verification. Which incident response principle does this AI-automated process PRIMARILY support?

• A.Evidence preservation and chain of custody
• B.Proactive threat hunting
• C.Vulnerability management
• D.Risk assessment

Domain: 3.0 AI-Assisted Security

Q10.An organization deploys an LLM with a plugin that can query a customer database. The system prompt instructs the model to only respond to customer inquiries about their own accounts. However, through a carefully crafted conversation, a user convinces the model to query another customer's account data. According to OWASP Top 10 for LLM Applications, which vulnerability does this exploit?

• A.Prompt Injection
• B.Excessive Agency
• C.Insecure Output Handling
• D.Model Denial of Service

Domain: 2.0 Securing AI Systems