42 free CompTIA SecAI+ CY0-001 practice questions from the CertStash pack of 169, each with the correct answer and a full explanation. The rule is the same for every exam: ten free on the product page, 42 free here, all 169 in the pack, and each step includes the one before it.
Get all 169 questions (US$39) · Download the first 10 as a PDF · Read the study guide
Question 1
Which of the following job roles in an organizational governance structure develops a model from business use cases?
Show answer and explanation
Correct answer: D. Data scientist
Data scientists develop predictive models from business use cases and requirements. They translate business problems into machine learning models, working directly with use case specifications to create the model architecture and approach. While platform architects design systems and MLOps engineers operationalize models, the data scientist role is specifically focused on model development from business requirements.
Why the other options are wrong
- A. Platform architects design overall system infrastructure, not develop individual models from use cases.
- B. AI risk analysts focus on identifying and mitigating risks, not developing models from business requirements.
- C. MLOps engineers manage model deployment and operations, not the initial model development from use cases.
Question 2
An administrator, who works for a financial institution, is required to implement data security controls for data at rest within AI systems that involve data disclosure. Which of the following is the most suitable control?
Show answer and explanation
Correct answer: C. Encryption
Encryption is the most suitable control for protecting data at rest, especially in financial institutions where data disclosure is a concern. Encryption renders data unreadable without the proper decryption keys, providing a fundamental security layer that prevents unauthorized access to sensitive information regardless of how the data is stored or breached.
Why the other options are wrong
- A. Data lineage tracks data flow and origins but does not protect data from disclosure.
- B. Rate limits control access frequency but do not protect the data itself from being read if accessed.
- D. Masking obscures data display but does not prevent access to underlying plaintext data at rest.
Question 3
A security engineer needs to monitor an AI-based system for runtime operations. The engineer is mostly concerned about the visibility of internal activity. Which of the following is the most appropriate monitoring solution?
Show answer and explanation
Correct answer: D. Enabling stack call and debugging level traces at the function level
Enabling stack call and debugging level traces at the function level provides the deepest visibility into internal activity and execution flow of an AI system. This approach captures detailed function-level execution information, variable states, and control flow, which is essential for understanding internal behavior during runtime operations.
Why the other options are wrong
- A. SIEM tools are designed for security event management across infrastructure but may not provide sufficient granularity into AI model internal operations.
- B. WAF with header logging focuses on web layer attacks and request/response headers, not internal AI system activity.
- C. Relying solely on vendor controls and prompt input monitoring does not provide visibility into the internal computational processes and function-level execution of the AI system.
Question 4
Which of the following should an auditor reference when reviewing a company’s human resources AI systems for legal non-compliance?
Show answer and explanation
Correct answer: C. European Union (EU) AI Act
The European Union AI Act is the most directly applicable legal framework for reviewing HR AI systems for compliance. This regulation specifically addresses AI systems used in employment and hiring decisions, establishing legal requirements for high-risk AI applications in human resources. It provides the binding legal standards that auditors must reference for legal compliance in this context.
Why the other options are wrong
- A. OECD standards provide guidance and principles but are not binding legal requirements like the EU AI Act.
- B. NIST RMF is a risk management framework for the US and provides best practices but is not the legal standard for EU HR AI systems.
- D. ISO standards provide general quality and management frameworks but do not specifically address the legal requirements for AI systems in HR.
Question 5
An airline corporation wants to implement a chatbot application using a large language model (LLM) so its customers: Can ask question and receive answers about flight details. Have the option to upload files. Which of the following security controls should the airline use to protect against malicious input and unauthorized use beyond the service-level agreement? (Choose two.)
Show answer and explanation
Correct answer: A, D. Prompt guardrails; Model token quotas
Prompt guardrails prevent malicious input by filtering and validating user prompts before they reach the model, protecting against prompt injection and harmful queries. Model token quotas limit resource usage and prevent unauthorized use beyond the service agreement by capping the number of tokens each user or session can consume. Together, these controls address both the malicious input threat and the unauthorized usage concern specific to the airline's requirements.
Why the other options are wrong
- B. Role-based access controls manage who can access the system but do not specifically protect against malicious input content or token-based overuse.
- C. Firewall rules control network traffic but do not address malicious input within valid API requests or token consumption limits.
Question 6
A security operations center (SOC) has a very high volume of logs and alerts. The manager proposes the implementation of machine learning (ML) system to help with triage. Which of the following tasks is most suitable?
Show answer and explanation
Correct answer: C. Identifying and classifying alerts
Identifying and classifying alerts is the most suitable ML task for handling high-volume alert triage. Machine learning excels at pattern recognition and categorization, allowing the system to automatically sort and classify alerts by severity, type, and relevance. This reduces noise and helps SOC teams focus on the most critical incidents, directly addressing the high-volume problem.
Why the other options are wrong
- A. Applying filters is a static, rule-based task that does not leverage ML capabilities and requires manual filter creation.
- B. Automatically patching systems is a remediation action, not a triage task, and carries operational risk that requires human oversight.
- D. Summarizing alert content is a generative task but does not address the core triage challenge of identifying and prioritizing which alerts matter most.
Question 7
An organization recently created a custom model that integrates with a language model (LLM). The developer notices that the application programming interface (API) costs have increased. Which of the following is the best control to reduce cost?
Show answer and explanation
Correct answer: D. Adjusting token limits
Adjusting token limits directly controls API costs for LLM integrations, as most LLM providers charge per token consumed. By setting appropriate token limits on requests, responses, or user sessions, the organization can prevent excessive token usage that drives up costs while maintaining service functionality.
Why the other options are wrong
- A. Prompt templates improve consistency but do not directly reduce token consumption or API costs.
- B. Increasing CPU and memory increases infrastructure costs rather than reducing LLM API costs.
- C. Reducing model size may impact quality and capabilities; token limits are a more direct cost control mechanism.
Question 8
A security administrator needs to improve an AI model. During an initial investigation, the administrator notices that two successive login features are recorded every day, and then a successful login occurs after a specific time interval. All the successful login attempts have been during office hours. Which of the following techniques should the administrator use to improve the AI model’s security?
Show answer and explanation
Correct answer: B. Pattern recognition
Pattern recognition is the technique that identifies the recurring pattern of two failed login attempts followed by a successful login during office hours. This behavioral pattern detection is essential for improving the model's ability to distinguish normal from anomalous login behavior. Pattern recognition can identify this repeating sequence and potentially flag deviations from it as security risks.
Why the other options are wrong
- A. Access management controls who can access systems but does not analyze the pattern of login attempts and their timing.
- C. Signature matching looks for known malicious indicators or hashes, not behavioral patterns in login sequences.
- D. Vulnerability analysis identifies weaknesses in systems or code, not behavioral patterns in user activity and login attempts.
Question 9
Which of the following is the most concerning risk for a company that allows corporate end users to use public-facing large language models (LLMs)?
Show answer and explanation
Correct answer: C. Data security regulatory violations
When corporate end users input data into public-facing LLMs, that data may be retained, used for model training, or accessed by third parties, creating exposure of confidential business information and personal data. This violates data protection regulations like GDPR, CCPA, and HIPAA. While hallucinations and malicious code are real concerns, they affect output quality rather than regulatory compliance. Outdated acceptable use policies are a governance issue but less immediately threatening than direct data exposure.
Why the other options are wrong
- A. Hallucinations reduce output quality but don't create regulatory or compliance violations.
- B. Outdated policies are a governance problem, not the most acute risk compared to data exposure.
- D. Malicious code generation is a capability concern, not the primary risk from data submission to public services.
Question 10
Which of the following requires developers to harden infrastructure to protect AI systems?
Show answer and explanation
Correct answer: D. Configuration standards
Configuration standards establish baseline security requirements, secure defaults, and hardening procedures that developers must implement to protect AI systems from attack, unauthorized access, and data breaches. Intake processes handle onboarding, acceptable use policies define user behavior rules, and development guidelines provide coding best practices, but only configuration standards directly mandate the technical hardening of infrastructure.
Why the other options are wrong
- A. Intake processes manage initial requirements gathering, not infrastructure hardening.
- B. Acceptable use policies define user conduct, not infrastructure security controls.
- C. Development guidelines cover coding practices but configuration standards specifically mandate hardening deployment and infrastructure setup.
What people say about the full CY0-001 pack 5.0 from 4
Sat the exam last week and passed. Thank you.
The CompTIA SecAI+ CY0-001 exam questions were exactly what I needed at the time. They made the content easier to review and helped me feel ready for the exam.
Your CompTIA SecAI+ dumps are really very helpful.
Question 11
Which of the following is required first in order to send a prompt query and response in a language model (LLM) system when authentication is enabled?
Show answer and explanation
Correct answer: C. Application programming interface gateway
An API gateway is the primary access control point that must be established first when authentication is enabled. The API gateway handles authentication, authorization, and routing of requests to the LLM system, serving as the security checkpoint through which all authenticated prompt queries and responses must pass before reaching the backend model.
Why the other options are wrong
- A. A front-end web proxy gateway operates at the user interface layer but is not the first required authentication control for LLM queries.
- B. Endpoint access control manages permissions at the endpoint level but does not serve as the initial authentication gateway for LLM systems.
- D. Back-end access gateway operates after the API layer and is not the first authentication control required.
Question 12
A developer is selecting authentication controls for an AI system. Which of the following is the best way to prevent threat actor replay attacks?
Show answer and explanation
Correct answer: C. Expiring session tokens
Expiring session tokens directly prevent replay attacks by ensuring that captured tokens become invalid after a set time period, rendering them useless to an attacker even if they intercept them. This is the most direct mitigation against replay attacks in authentication. IdP federation addresses identity verification but not replay prevention; SSH certificates authenticate users but don't prevent token replay; access keys without expiration don't solve replay vulnerability.
Why the other options are wrong
- A. IdP federation handles identity verification but does not inherently prevent replay attacks.
- B. SSH-based certificate authentication provides strong authentication but does not specifically prevent session replay attacks.
- D. Access keys alone without expiration do not prevent an attacker from replaying captured credentials.
Question 13
An attacker successfully completes a denial-of-service (DoS) attack through the context window of an AI system. Thousands of characters are obfuscated and hidden behind an emoji. Which of the following techniques best mitigates this type of attack?
Show answer and explanation
Correct answer: D. Prompt filter
A prompt filter examines incoming text for obfuscation techniques (hidden characters, emojis masking content, Unicode tricks) and can detect and block attempts to bypass context window limits with encoded payloads. This directly addresses the attack vector described. Fraud detection targets financial anomalies, LLM-as-a-judge evaluates output quality, and pattern recognition is too generic; only prompt filters specifically validate and sanitize input before it reaches the model's context window.
Why the other options are wrong
- A. Fraud detection identifies financial anomalies, not obfuscation or context window exhaustion attacks.
- B. LLM-as-a-judge evaluates generated output correctness, not input validation or DoS prevention.
- C. Pattern recognition is too broad and doesn't specifically prevent obfuscated prompt injection at the input layer.
Question 14
A team of data scientists is ready to release a model for enterprise use. The team wants to protect the model from unintentional changes or tampering. Which of the following is the most appropriate action?
Show answer and explanation
Correct answer: D. Integrate an application programming interface (API) with identity and access management (IAM) roles to interact with the model.
Integrating an API with IAM roles provides the most comprehensive protection by combining secure access control with authentication and authorization mechanisms. This approach ensures only authorized personnel can interact with or modify the model through authenticated API calls with role-based permissions, protecting against both unintentional changes and deliberate tampering while maintaining usability for legitimate enterprise users.
Why the other options are wrong
- A. Converting to an LLM with guardrails changes the model's architecture rather than protecting an existing model from tampering.
- B. Providing local copies increases tampering risk by distributing copies across multiple systems without centralized access control.
- C. Restricting to only IT professionals is too narrow and doesn't provide technical protection mechanisms against tampering.
Question 15
Which of the following responsible AI standards refers to a principle that clearly states the reasons behind the decisions for a particular conclusion?
Show answer and explanation
Correct answer: D. Explainability
Explainability is the responsible AI principle that ensures AI systems provide clear, understandable reasons and justifications for their decisions and conclusions. It addresses the need for interpretability and transparency in the decision-making process. Accountability refers to responsibility for outcomes, auditability refers to the ability to track and review systems, and transparency refers to openness about system operations, but explainability specifically focuses on clarifying the reasoning behind conclusions.
Why the other options are wrong
- A. Accountability refers to responsibility for AI system outcomes and actions, not the reasoning behind decisions.
- B. Auditability refers to the ability to review and verify AI system behavior and decisions after the fact, not explaining reasoning.
- C. Transparency is about openness regarding AI system operations and data usage, broader than explaining decision reasoning specifically.
Question 16
A cybersecurity analyst must use pattern recognition on a data set containing unstructured data. Which of the following models is the best for this task?
Show answer and explanation
Correct answer: B. Convolutional neural network
Convolutional neural networks excel at pattern recognition in unstructured data, particularly images, text, and complex signal data. CNNs use convolutional layers to automatically detect spatial patterns and features, making them ideal for extracting meaningful patterns from unstructured datasets.
Why the other options are wrong
- A. Long short-term memory networks are specialized for sequential data with temporal dependencies, not general unstructured pattern recognition.
- C. Decision trees work best with structured, tabular data with clear feature boundaries, not unstructured data.
- D. Logistic regression is a linear classifier requiring structured numerical features and is not designed for unstructured data analysis.
Question 17
A security analyst is preparing a presentation for the sales team that describes the most common vulnerabilities that are specific to AI applications. Which of the following is the best source for the analyst to consult?
Show answer and explanation
Correct answer: C. Open Worldwide Application Security Project (OWASP)
OWASP is the primary organization that publishes guidance specific to application security vulnerabilities, including AI-specific risks such as prompt injection, data poisoning, and model extraction. OWASP's Top 10 for LLMs is the industry standard resource for AI security vulnerabilities. While ISO 27001 covers information security and NIST-RMF covers risk management frameworks, neither specializes in application-specific vulnerabilities like OWASP does.
Why the other options are wrong
- A. ISO 27001 is a general information security management standard, not specific to AI application vulnerabilities.
- B. CWE covers software weaknesses broadly but is not specialized for AI-specific threats.
- D. NIST-RMF is a risk management framework applicable across domains but does not focus on AI application vulnerabilities specifically.
Question 18
A manufacturing company wants to use AI within its operations to improve the efficiency and accuracy of its processes. Which of the following should the organization do first to enable adoption and achieve the business objectives?
Show answer and explanation
Correct answer: B. Hire a data and AI architect.
Hiring a data and AI architect should be the first step because this role establishes strategic direction, assesses organizational readiness, identifies business objectives, evaluates technical requirements, and creates an implementation roadmap. This foundational planning enables all subsequent decisions. ISO 42001 certification is beneficial but comes after strategy. LLM selection and GAN implementation are specific technical choices that should follow architectural planning.
Why the other options are wrong
- A. ISO 42001 certification is a governance framework that should be implemented after establishing an AI strategy and architecture.
- C. Selecting an LLM is a technical tool decision that should follow strategic planning and architectural assessment.
- D. Implementing a GAN is a specific technical application that should follow strategic planning and architectural design.
Question 19
A cybersecurity analyst wants to choose a machine learning (ML) model to classify log entries while providing the best explainability. Which of the following models should the analyst use?
Show answer and explanation
Correct answer: C. Decision trees
Decision trees provide superior explainability by making classification rules transparent and interpretable: analysts can trace the exact decision path for any classification decision. This is critical for security log analysis where understanding the reasoning is essential. LLMs lack explainability and are black-box models. Neural networks are also difficult to interpret. GANs are generative models unsuitable for classification tasks.
Why the other options are wrong
- A. Large language models are black-box systems that lack explainability for individual predictions.
- B. Neural networks, particularly deep networks, are difficult to interpret and explain their classification decisions.
- D. Generative adversarial networks are designed for data generation, not classification of log entries.
Question 20
A security architect performs threat modeling of an AI system. The architect needs to determine which attacks can be performed against the system. Which of the following actions should the architect take next?
Show answer and explanation
Correct answer: D. Analyze MITRE Adversarial Threat Landscape for AI Systems (ATLAS) for tactics, techniques, and procedures (TTPs).
MITRE ATLAS (Adversarial Threat Landscape for AI Systems) is specifically designed for threat modeling of AI systems and provides a comprehensive knowledge base of tactics, techniques, and procedures used by adversaries against AI/ML systems. It is the most appropriate framework for an architect performing threat modeling to systematically identify which attacks can be performed against the AI system. While OWASP Top 10 is useful for application security, ATLAS is the specialized framework for AI threat modeling.
Why the other options are wrong
- A. Using an LLM to map attack paths is not a systematic threat modeling approach and lacks the rigor of established frameworks.
- B. Quantifying risk of known vulnerabilities comes after identifying potential threats and attack paths.
- C. While trust boundaries and OWASP Top 10 are useful, they are application-focused, not specifically designed for AI system threats like ATLAS.
Question 21
Which of the following controls is the best way to mitigate a denial-of-service (DoS) attack?
Show answer and explanation
Correct answer: B. Rate limiting
Rate limiting directly mitigates DoS attacks by restricting the number of requests from individual sources within a time window, preventing resource exhaustion. This is the standard and most effective control for DoS protection. Model guardrails relate to AI safety, not network attacks. Encryption protects data in transit but doesn't prevent resource consumption. Access controls manage permissions but don't limit request frequency.
Why the other options are wrong
- A. Model guardrails are AI safety measures unrelated to network denial-of-service attacks.
- C. End-to-end encryption protects data confidentiality but does not prevent DoS attacks based on request volume.
- D. Access controls determine who can access resources but do not limit the rate of requests from authenticated users.
Question 22
An organization is concerned with the exposure of sensitive data. Which of the following is the most relevant security concern?
Show answer and explanation
Correct answer: B. Model inversion
Model inversion is a security attack where an adversary reconstructs sensitive training data (including personal medical records) from a trained model, making it the most relevant security concern for sensitive data exposure. Overfitting is a performance issue, data normalization is a preprocessing technique, and hyperparameter tuning is a modeling optimization task; none specifically address sensitive data exposure risks.
Why the other options are wrong
- A. Overfitting is a model performance issue, not a mechanism for sensitive data exposure.
- C. Data normalization is a preprocessing technique unrelated to data security or exposure.
- D. Hyperparameter tuning is a model optimization process, not a security vulnerability.
Question 23
Which of the following helps in managing potential security issues related to model training?
Show answer and explanation
Correct answer: A. National Institute of Standards and Technology (NIST) AI Risk Management Framework (RMF)
The NIST AI Risk Management Framework (RMF) is specifically designed to address risks associated with AI systems throughout their lifecycle, including model training. It provides structured guidance on identifying, measuring, and managing AI-specific security and safety risks during development and deployment.
Why the other options are wrong
- B. ISO 27001 addresses general information security management but is not specifically tailored for AI model training risks.
- C. OECD provides principles for AI governance and responsible development but is not a dedicated framework for managing training-specific security issues.
- D. GDPR focuses on data privacy and personal data protection rights rather than the security of model training processes themselves.
Question 24
During the selection of a machine learning (ML)-based threat classification model, a cybersecurity administrator verifies that label distribution is highly unbalanced. Which of the following processing techniques should the engineer use to balance the model?
Show answer and explanation
Correct answer: B. Data augmentation
Data augmentation is the technique used to address imbalanced label distribution in machine learning models. It involves creating synthetic examples of underrepresented classes through transformations, replication, or generative methods to balance the training dataset and improve model performance across all classes.
Why the other options are wrong
- A. Data lineage tracks the origin and flow of data but doesn't address class imbalance.
- C. Data provenance documents data source and history but doesn't resolve imbalance issues.
- D. Data verification checks data quality and correctness but doesn't balance class distributions.
Question 25
Which of the following attacks would be the best to automate with AI during dynamic application software testing (DAST)?
Show answer and explanation
Correct answer: C. Payload creation
Payload creation is best automated with AI during DAST because AI can generate diverse, context-aware attack payloads at scale, test them against applications, and iterate based on responses. This directly supports dynamic testing by automatically crafting and delivering varied malicious inputs without manual effort.
Why the other options are wrong
- A. DDoS attacks involve coordinated network traffic and are not a primary focus of application-level DAST.
- B. Data poisoning involves corrupting training data and is not applicable to dynamic application software testing.
- D. Threat modeling is a planning and analysis activity, not an automated attack technique executed during DAST execution.
Question 26
Which of the following International Organization for Standardization (ISO) standards should be selected for certification to use for third-party assurance for responsible AI practices?
Show answer and explanation
Correct answer: D. 42001
ISO 42001 is the international standard specifically designed for artificial intelligence management systems and responsible AI practices. It provides a framework for organizations to implement, manage, and maintain AI governance. ISO 20000 covers IT service management; ISO 27001 addresses information security management; ISO 27701 extends 27001 for privacy; none are specifically designed for responsible AI practices certification.
Why the other options are wrong
- A. ISO 20000 is for IT service management, not AI governance.
- B. ISO 27001 is a general information security management standard, not AI-specific.
- C. ISO 27701 extends privacy controls to ISO 27001 but is not the primary AI governance standard.
Question 27
A security operations center (SOC) analyst needs to automate multiple security tasks by breaking them down into smaller parts. Which of the following AI tools is the best for this task?
Show answer and explanation
Correct answer: A. Agentic AI
Agentic AI refers to autonomous AI agents that can break down complex tasks into smaller subtasks, plan sequences of actions, execute them independently, and adapt based on results. This is ideal for automating multiple security tasks by decomposing them into manageable steps and executing them with minimal human intervention. RAG AI is for information retrieval, generative AI is for content creation, and chatbots are for user interaction; none are designed for autonomous multi-step task automation.
Why the other options are wrong
- B. RAG AI is designed for retrieving and augmenting information, not for breaking down and automating multiple security tasks.
- C. Generative AI creates content but is not specifically designed for autonomous task decomposition and multi-step automation.
- D. Chatbots are conversational interfaces and are not designed to autonomously break down and execute complex multi-step security tasks.
Question 28
Which of the following is used to train an AI model with unstructured data?
Show answer and explanation
Correct answer: B. Fine-tuning
Fine-tuning is the process of taking a pre-trained model and adapting it with additional training on new, unstructured data (text, images, audio, etc.). This approach is specifically designed to leverage unstructured data by building on existing model knowledge. Statistical learning requires structured numerical data, supervised learning requires labeled data, and reinforcement training focuses on reward-based learning rather than learning from unstructured data directly.
Why the other options are wrong
- A. Statistical learning typically requires structured, numerical data rather than unstructured data.
- C. Supervised learning requires labeled training data with explicit input-output pairs, not raw unstructured data.
- D. Reinforcement training uses reward signals and agent interactions, not unstructured raw data as its primary training mechanism.
Question 29
An organization recently developed an AI-powered product and discovers that it is vulnerable to attacks in which malicious actors can alter the input, causing the system to recommend inappropriate information. Which of the following techniques is the most effective way to secure the system against manipulation attacks?
Show answer and explanation
Correct answer: D. Guardrails
Guardrails are the most effective technique to prevent the AI system from recommending inappropriate information when inputs are altered. Guardrails establish strict boundaries on what the model can output, filtering responses for safety violations and ensuring outputs remain appropriate regardless of malicious input manipulation. This directly addresses the vulnerability where altered inputs cause inappropriate recommendations.
Why the other options are wrong
- A. Cross-validation improves model accuracy but doesn't prevent responding to adversarial or malicious inputs.
- B. Feature regularization prevents overfitting but doesn't defend against adversarial input manipulation attacks.
- C. Feature scaling normalizes input ranges but doesn't filter or prevent inappropriate outputs from malicious prompts.
Question 30
A security analyst finds that the AI system is under a denial-of-wallet attack. Which of the following should the analyst enforce to protect the company? (Choose two.)
Show answer and explanation
Correct answer: E, F. Application programming interface (API) rate controls; Output token controls
A denial-of-wallet attack exploits AI systems by making them perform expensive computational operations, draining financial resources through excessive API usage. API rate controls (E) limit the number of requests to prevent resource exhaustion, while output token controls (F) limit the computational cost per response. Both directly address the resource consumption mechanism of such attacks.
Why the other options are wrong
- A. Endpoint access controls restrict who can access systems but don't prevent expensive computational operations once authenticated.
- B. CDNs optimize content delivery but don't limit the computational cost of AI inference operations.
- C. Model fine-tuning improves model performance but doesn't protect against resource exhaustion attacks.
- D. Modality controls restrict input/output types but don't directly limit computational resource consumption.
Question 31
Which of the following is most resistant to AI manipulation?
Show answer and explanation
Correct answer: D. Attack surface reduction
Attack surface reduction is most resistant to AI manipulation because it fundamentally limits the number of potential entry points and exposed components available for an attacker to exploit. By reducing what is exposed and accessible, the system has fewer vectors for manipulation regardless of AI sophistication. This architectural approach is inherently defensive against AI-specific attack techniques.
Why the other options are wrong
- A. Payloads are the vehicle for attacks and are directly subject to manipulation by malicious actors.
- B. AI-generated content is produced by AI systems and is vulnerable to prompt injection and manipulation attacks.
- C. API gateways, while protective, can still be bypassed or manipulated through sophisticated attacks on their logic.
- E. Antivirus protects against known malware but doesn't address AI manipulation or prompt injection attacks.
Question 32
An organization develops a chatbot with the following requirements: Does not provide harmful or explicit responses Must use clean and professional language Ensures that responses are accurate Which of the following should the organization conduct after the chatbot is fully developed but before a customer-facing deployment?
Show answer and explanation
Correct answer: C. Guardrail testing and validation
Guardrail testing and validation is the specific practice of verifying that an AI system adheres to defined safety constraints and behavioral boundaries. This directly tests whether the chatbot meets all three stated requirements: avoiding harmful/explicit responses, using clean professional language, and ensuring accuracy. Guardrails are the safeguards that enforce these constraints before customer-facing deployment.
Why the other options are wrong
- A. Data labeling and classification are upstream processes for preparing training data, not deployment validation.
- B. Model auditing is a broader review process; guardrail testing is the specific validation needed for these behavioral requirements.
- D. Regression modeling is a statistical technique unrelated to chatbot safety validation or guardrail enforcement.
Question 33
A company uses human review for software development validation and wants to add another validation layer. Which of the following should a security administrator use to accomplish task?
Show answer and explanation
Correct answer: A. AI-assisted approval
AI-assisted approval adds an additional automated validation layer that complements human review by flagging issues, analyzing code patterns, and providing recommendations, thereby strengthening the validation process. Low-code plug-ins facilitate development but don't add validation. Automated rollback addresses deployment failures, not validation layers. Regression testing validates functionality but is not an additional approval layer.
Why the other options are wrong
- B. Low-code plug-ins facilitate development capabilities, not validation or approval workflows.
- C. Automated rollback is a deployment safety mechanism, not a validation layer for software review.
- D. Regression testing validates that changes don't break existing functionality but is not an approval or validation layer.
Question 34
An AI security team must assess the probability of an attack on its new system and the impact associated with such an attack. Which of the following threat-modeling resources best addresses the threat landscape for machine learning (ML)?
Show answer and explanation
Correct answer: B. MITRE Adversarial Threat Landscape for AI Systems (ATLAS)
MITRE Adversarial Threat Landscape for AI Systems (ATLAS) is specifically designed as a threat-modeling framework that addresses the unique attack surface and threat landscape of machine learning systems. It provides attack patterns, techniques, and case studies tailored to AI/ML security concerns, making it the best resource for assessing attack probability and impact on ML systems.
Why the other options are wrong
- A. CVE AI working group focuses on vulnerability enumeration but is not a comprehensive threat-modeling framework specific to ML.
- C. MIT risk repository is a general-purpose resource and lacks the ML-specific threat modeling framework needed.
- D. OWASP provides application security guidance but is not ML-specific; ATLAS is the dedicated ML threat-modeling resource.
Question 35
Which of the following is an example of how a security analyst uses generative AI in the triage process?
Show answer and explanation
Correct answer: C. To summarize security findings by category
Summarizing security findings by category is a direct application of generative AI capabilities in triage, where the AI rapidly processes and organizes large volumes of alerts and findings for analyst review. Predicting attack targets requires predictive ML not generative AI, statistical analysis for malware uses classical ML not generative AI, and tagging malware uses classification algorithms not generative AI; none represent how generative AI specifically enhances triage.
Why the other options are wrong
- A. Predicting attack targets requires predictive modeling, not generative AI capabilities.
- B. Statistical malware analysis uses classical ML algorithms, not generative AI.
- D. Malware tagging uses classification algorithms, not generative AI summarization.
Question 36
A data set containing medical information is put into a machine learning (ML) model that is designed to predict specific illnesses for a population. In the process of verifying the reliability of the system, the compliance officer realizes that the system cannot reliably predict illnesses for certain segments of the population. Which of the following types of risk is most applicable to this case?
Show answer and explanation
Correct answer: A. Bias
Bias is the most applicable risk when an ML model performs reliably for some population segments but not others, which indicates systematic disparities in prediction accuracy across demographics. This is the definition of algorithmic bias. Consistency refers to reproducibility of results, transparency concerns model explainability, and inclusiveness relates to representation in design; none directly address differential prediction performance across populations.
Why the other options are wrong
- B. Consistency refers to reproducible performance, not differential accuracy across demographic segments.
- C. Transparency is about model explainability, not unequal predictive reliability across populations.
- D. Inclusiveness in design differs from bias in model predictions causing differential reliability.
Question 37
Which of the following describe the practice of providing examples in a prompt? (Choose two.)
Show answer and explanation
Correct answer: E, F. One-shot; Multi-shot
One-shot and multi-shot learning are practices where examples are provided in a prompt to guide model behavior. One-shot provides a single example, while multi-shot provides multiple examples. Both are example-based prompting techniques that improve model performance by showing the desired input-output pattern rather than only describing it.
Why the other options are wrong
- A. User prompt refers to the input itself, not the practice of providing examples.
- B. System prompt is an instruction to guide overall behavior, distinct from providing examples in the user prompt.
- C. Prompt template is a reusable structure for prompts but doesn't specifically describe the practice of including examples.
- D. Quantization is a model optimization technique unrelated to prompting practices.
Question 38
A line of business wants to onboard an application that uses a custom AI model for employee assessments. The Chief Information Officer (CIO) agrees to allow the engagement to proceed but first wants a threat model. Which of the following is the most appropriate to use for an AI threat model?
Show answer and explanation
Correct answer: B. Adversarial Threat Landscape for AI Systems (ATLAS)
The Adversarial Threat Landscape for AI Systems (ATLAS) is specifically designed as a threat model framework for AI systems. It provides a structured approach to identifying, categorizing, and understanding threats specific to AI/ML systems, making it the most appropriate choice for threat modeling an AI-based application.
Why the other options are wrong
- A. Responsible AI focuses on ethical principles and governance rather than threat modeling and security vulnerabilities.
- C. OECD provides AI principles and governance guidelines but is not a formal threat modeling framework.
- D. ISO standards like ISO 27001 address information security generally but ATLAS is specifically tailored for AI threat modeling.
Question 39
A machine learning (ML) engineer is working with a security engineer to identify the best practices for securing a system with various AI models. Which of the following actions should the engineers suggest?
Show answer and explanation
Correct answer: B. Following a secure model development life cycle (MDLC)
Following a secure model development life cycle (MDLC) is the foundational best practice that encompasses all security activities for AI systems, including threat modeling, secure training, validation, and deployment phases specific to machine learning. A secure MDLC is purpose-built for AI security unlike a generic SDLC, and it encompasses guardrail testing and security architecture as components within a comprehensive framework.
Why the other options are wrong
- A. Guardrail testing is important but represents only one component of comprehensive AI security.
- C. Security architecture is necessary but incomplete without the systematic MDLC framework.
- D. Generic SDLC lacks AI-specific security practices needed for ML model security.
Question 40
A security administrator must provide access controls for AI systems to list tables. Which of the following should the administrator implement?
Show answer and explanation
Correct answer: D. Data access
Data access controls regulate which AI systems and users can access specific tables and datasets, directly addressing the security requirement to control AI system permissions. Network access control lists operate at the network layer for traffic filtering, agentic AI access relates to autonomous agent capabilities, and model access is less specific than data access controls which are needed to restrict table visibility.
Why the other options are wrong
- A. Agentic AI access controls autonomous agent capabilities, not database table access.
- B. Network ACLs filter network traffic, not database or table-level access permissions.
- C. Model access is too broad; data access controls specifically govern table and dataset permissions.
Question 41
A human resources officer is using AI to evaluate resumes and help select candidates that meet minimum criteria. To improve the results, the human resources officer adjusts the query parameters and includes an example resume that matches a successful candidate. Which if the following best describes this query?
Show answer and explanation
Correct answer: C. One-shot prompting
One-shot prompting involves providing a single example of the desired behavior to guide the AI model's output. The HR officer is including an example resume of a successful candidate to demonstrate what kind of results they want, which is the defining characteristic of one-shot prompting. This technique helps the model understand the pattern by showing one concrete example.
Why the other options are wrong
- A. Distillation is a model compression technique where a larger model's knowledge is transferred to a smaller model, not a prompting strategy.
- B. A prompt template is a reusable structure with placeholders, not specifically about providing example outputs to demonstrate desired behavior.
- D. System role defines the AI's persona or function but doesn't involve providing examples of successful outputs.
Question 42
Which of the following improves the observability and auditing of an AI system?
Show answer and explanation
Correct answer: C. Implementing machine learning operations (MLOps)
MLOps (Machine Learning Operations) encompasses practices and tools for monitoring, logging, versioning, and auditing AI systems throughout their lifecycle. It provides comprehensive observability through model tracking, performance monitoring, data validation, and audit trails that enable systematic oversight and accountability. Redeploying doesn't improve observability, manual detection doesn't scale, and anomaly detection is a component within MLOps rather than a comprehensive solution.
Why the other options are wrong
- A. Redeploying a model alone does not improve observability or auditing capabilities without supporting infrastructure.
- B. Manual detection is not scalable or systematic for comprehensive observability of AI systems.
- D. Anomaly detection is a technique used within observability frameworks but is not comprehensive enough to address all auditing requirements.
That was 42 of 169.
The full CompTIA SecAI+ CY0-001 pack has all 169 questions, each with the answer, the explanation and why the other options are wrong, plus a questions-only copy for timed runs. US$39, paid once, with free monthly updates and a pass-or-your-money-back guarantee.
Not ready yet? Get the first 84 free as one PDF by email. Get the Half Pack →