CompTIA CY0-001 CompTIA SecAI+ v1 Exam Exam Practice Test

Basic Concept: A replay attack occurs when an attacker captures a valid authentication token or credential and reuses it to impersonate a legitimate user. Preventing replay attacks requires ensuring that captured credentials cannot be successfully reused after a defined period or after their intended single use. CompTIA SecAI+ Study Guide covers replay attack prevention under AI system authentication.

Why C is Correct: Expiring session tokens have a limited validity window, typically a few minutes to hours. If an attacker captures a token, they can only use it until it expires. Short expiration times dramatically reduce the window of opportunity for replay attacks. This is the most direct and effective control specifically targeting replay attack prevention, as expired tokens are rejected even if intercepted.

Why A is Wrong: IdP federation enables single sign-on across multiple systems using federated identity providers. While it standardizes authentication, it does not inherently prevent replay attacks on captured tokens unless combined with short token expiration and proper validation.

Why B is Wrong: SSH certificate authentication uses cryptographic certificates for strong authentication. While more secure than password-based SSH, certificates alone do not prevent replay attacks unless they include timestamps, nonces, or other anti-replay mechanisms that invalidate captured credentials.

Why D is Wrong: IAM access keys are long-lived credentials that provide programmatic access to services. They are typically static and do not expire automatically, making them vulnerable to replay attacks if intercepted. They are less suitable for replay attack prevention than expiring session tokens.

An audit trail is the governance mechanism that preserves evidence of actions, decisions, approvals, validation activities, and responsible parties throughout the lifecycle of an AI system. If the organization cannot establish who authorized deployment or how the system was validated, the fundamental problem is missing traceability and historical evidence. This is precisely what an audit trail should provide. CompTIA SecAI+ emphasizes governance roles, model validation, monitoring, auditing for quality and compliance, and accountability throughout the AI lifecycle. A completed threat model documents threats and attack scenarios, but it would not necessarily record deployment approval or validation evidence. Blueprint controls are not the primary mechanism for preserving historical authorization records. Organizational policies may define that approval and validation must occur, but merely having policies does not demonstrate that the required actions actually occurred. Proper audit records would identify the approver, validation method, relevant results, dates, model version, and other governance evidence. Therefore, the inability to reconstruct those decisions indicates the absence of an adequate audit trail.

================

Basic Concept: Input manipulation attacks — including adversarial examples and prompt injection — alter inputs to cause AI systems to produce unintended, harmful, or inappropriate outputs. Defending against these attacks requires mechanisms that validate and constrain both inputs and outputs at runtime. CompTIA SecAI+ Study Guide identifies guardrails as the primary defense against input manipulation in AI systems.

Why D is Correct: Guardrails implement real-time validation and filtering of both incoming inputs and outgoing recommendations. They detect manipulated inputs that deviate from expected patterns, enforce content policies on outputs, and prevent the system from producing inappropriate recommendations regardless of how cleverly the input was crafted. Guardrails provide the most comprehensive and directly applicable defense against the described manipulation attack scenario.

Why A is Wrong: Cross-validation is a model evaluation technique that assesses how well a model generalizes to independent datasets during training. It measures predictive performance but does not provide runtime protection against input manipulation attacks on a deployed system.

Why B is Wrong: Feature regularization is a training technique that adds penalties to model weights to prevent overfitting. It improves generalization during training but does not inspect or validate inputs at inference time to detect manipulation.

Why C is Wrong: Feature scaling normalizes input feature values to a standard range for training efficiency. Like regularization, it is a preprocessing and training step that has no effect on defending against runtime input manipulation attacks on deployed systems.

Basic Concept: Preventing vulnerable code from reaching production requires an automated, mandatory gate in the software delivery pipeline. The CI/CD pipeline is the enforcement point where all code must pass before deployment. CompTIA SecAI+ Study Guide covers AI integration in secure development pipelines under AI-assisted security.

Why C is Correct: Implementing an LLM in the CI/CD runner creates a mandatory automated security gate that every code change must pass. The LLM can analyze code for vulnerabilities, insecure patterns, and policy violations, then automatically fail the build if issues are found. This prevents vulnerable code from progressing toward production without human bypass capability, making it the most effective enforcement mechanism.

Why A is Wrong: IDE plug-ins provide warnings to developers during coding, but developers can choose to ignore them and proceed with compilation and commits. Warnings are advisory, not preventive, and cannot guarantee vulnerable code is blocked from the pipeline.

Why B is Wrong: SOAR platforms with ML models are excellent for incident response and security operations automation. However, they are not positioned in the code delivery pipeline and do not gate code from progressing to production.

Why D is Wrong: An agentic penetration testing tool validates vulnerabilities reactively after code is written or deployed. This approach does not intercept code before production deployment and is typically used for post-deployment assessment rather than prevention.

Basic Concept: Generative AI produces natural language content based on input data. In a security operations context, triage involves rapidly understanding and prioritizing security events. Generative AI ' s strength lies in synthesizing information and producing readable summaries from complex data. CompTIA SecAI+ Study Guide covers generative AI applications in security operations.

Why C is Correct: Summarizing security findings by category is a natural application of generative AI in triage. The AI can process large volumes of alerts and security events, group them by type or severity, and generate concise natural language summaries that enable analysts to quickly understand the current threat landscape without reading individual alerts. This directly reduces triage time and cognitive load.

Why A is Wrong: Predicting the next attack target requires predictive analytics and threat intelligence correlation. While AI can assist with this, it is a forecasting task better suited to analytical ML models rather than generative AI, and it is a strategic intelligence function rather than a triage task.

Why B is Wrong: Statistical analysis for malicious code assessment uses mathematical and ML techniques to analyze code characteristics. This is a traditional ML classification task, not a generative AI application, and is performed during malware analysis rather than alert triage.

Why D is Wrong: Tagging malware using ML algorithms is a classification task that uses supervised ML models trained on malware features. It is a detection and classification function, not a generative AI triage application.

Basic Concept: AI systems can inadvertently reveal sensitive information such as PII, credentials, or internal data in their outputs when not properly controlled. Sensitive information disclosure is a critical OWASP LLM Top 10 risk. CompTIA SecAI+ Study Guide covers both vulnerability identification and appropriate data protection controls for AI outputs.

Why D is Correct: The scenario describes the AI system outputting sensitive information in its responses, which is a sensitive information disclosure vulnerability. The appropriate control is masking, which replaces sensitive data values such as credit card numbers, SSNs, or API keys with redacted or tokenized equivalents in the model ' s outputs before they are returned to users. This prevents the AI from disclosing sensitive data while still providing useful responses.

Why A is Wrong: Prompt injection involves crafting inputs to override model instructions. If the penetration test revealed sensitive information, the primary vulnerability is the disclosure of that sensitive data, not the injection mechanism itself. EDR monitors endpoint behavior, not AI output content.

Why B is Wrong: Model hallucinations produce fabricated information rather than disclosing real sensitive data. The described scenario involves actual sensitive information being revealed, not fictitious content generation.

Why C is Wrong: Jailbreaking circumvents safety restrictions but the primary harm demonstrated is sensitive data exposure. RBAC manages access permissions but does not prevent the model from including sensitive data in responses once access is granted.

Why E is Wrong: Role impersonation involves the AI pretending to be a different entity. This may be a secondary technique used by the penetration tester but the primary vulnerability described is the disclosure of actual sensitive information in the output.

Basic Concept: AI-generated content including personalized text, synthetic voice, and deepfake video has dramatically enhanced the effectiveness and scalability of social engineering attacks. Understanding how AI amplifies specific attack types is key to CompTIA SecAI+ basic AI concepts in the cybersecurity context.

Why B is Correct: Phishing attacks are most dramatically enabled by AI-generated content. AI can generate highly personalized, grammatically perfect phishing emails tailored to individual targets using publicly available information. It can create convincing deepfake audio and video for voice phishing (vishing) and video phishing, replicate executive communication styles for business email compromise, and generate phishing campaigns at massive scale. The quality and personalization that previously required skilled human social engineers can now be automated with AI.

Why A is Wrong: Model poisoning is a specific attack against AI systems that corrupts training data to manipulate model behavior. While sophisticated, it is a targeted AI security attack rather than a broad cybercrime enabled by AI-generated content at scale.

Why C is Wrong: Ransomware is malware that encrypts victim data and demands payment for decryption keys. While AI can assist in ransomware development, ransomware deployment relies on code execution and network propagation techniques more than AI-generated content.

Why D is Wrong: Remote code execution involves exploiting vulnerabilities to run arbitrary code on a target system. It relies on technical vulnerability exploitation rather than AI-generated content. AI might assist in finding vulnerabilities, but RCE is not primarily enabled by content generation.

Basic Concept: AI models trained on unrepresentative data can produce systematically inaccurate results for certain population groups. This is a form of algorithmic bias where the model ' s performance varies significantly across demographic segments, creating disparate outcomes. CompTIA SecAI+ Exam Objectives cover bias as a core AI governance and risk concept.

Why A is Correct: Bias in AI occurs when a model produces systematically skewed results for certain groups due to biased training data, flawed data collection, or model design choices. In this healthcare scenario, the inability to reliably predict illnesses for specific population segments indicates the training data likely underrepresented those segments, causing the model to learn inadequate patterns for them. This is a critical bias risk with serious health equity implications.

Why B is Wrong: Consistency refers to the model producing the same output given the same input across different runs or time periods. The problem described is not about inconsistent outputs for the same input but about systematically poor performance for specific population groups.

Why C is Wrong: Transparency refers to openness about how the AI model operates, what data it uses, and how it makes decisions. The compliance officer has already assessed the system, suggesting sufficient transparency exists to identify the performance gap.

Why D is Wrong: Inclusiveness is a design principle ensuring AI systems are designed to serve all users regardless of background. While related to the outcome, the specific risk type described — differential predictive accuracy across population segments — is most precisely categorized as bias.

Overreliance occurs when users place excessive trust in AI-generated output without performing sufficient independent verification. The scenario indicates that employees used a chatbot for research and reporting and subsequently published documents containing incorrect references. That pattern strongly indicates that AI-generated citations or factual claims were accepted without adequate human validation. CompTIA SecAI+ explicitly lists overreliance among attacks and weaknesses that security professionals must recognize when analyzing AI systems. Generative AI systems can produce hallucinated or inaccurate citations that appear credible, which is why human oversight and human validation remain important lifecycle controls. Insecure output handling instead concerns unsafe downstream processing of model output, such as executing generated code or passing untrusted output into another system without validation. Excessive agency occurs when an AI system has more autonomous permissions or capabilities than necessary. Sensitive information disclosure involves exposing protected or confidential information. None of those conditions explains publication of believable but inaccurate references as directly as overreliance. Appropriate mitigation includes independent verification, authoritative-source validation, human review, and monitoring AI output for accuracy and hallucinations before publication.

================

Basic Concept: Data integrity ensures that data has not been tampered with, corrupted, or modified during storage or transmission. For AI training data that is procured from external sources, cryptographic integrity verification is essential to confirm the data arrived unmodified. CompTIA SecAI+ Study Guide covers data integrity controls for AI data pipelines.

Why A is Correct: Implementing checksums provides cryptographic verification of data integrity. A checksum or hash value such as SHA-256 is computed from the dataset at the source. The receiver computes the same hash and compares it to the provided value. Any modification to the data during transit or storage will produce a different hash, immediately detecting tampering or corruption. This is the most reliable, automated, and scalable strategy for ensuring the integrity of procured training data.

Why B is Wrong: Human evaluation can verify data quality and relevance but is impractical for verifying integrity across large datasets of medical records. Human reviewers cannot detect subtle bit-level corruption or intentional small modifications, and the process is not scalable.

Why C is Wrong: Querying the model tests model performance rather than verifying the integrity of the underlying training data. The model cannot tell you whether its training data was modified after collection or during procurement.

Why D is Wrong: Log monitoring tracks system activities and events over time. While useful for auditing access to data, it cannot retroactively confirm that data content has not been modified and does not provide cryptographic integrity guarantees.

Basic Concept: Human-in-the-loop is a design pattern where AI systems generate recommendations or decisions but require human review and approval before those recommendations are acted upon. This approach maintains human oversight and accountability in AI-assisted workflows. CompTIA SecAI+ Study Guide covers human-in-the-loop as a key responsible AI principle and operational pattern.

Why C is Correct: The scenario precisely describes the human-in-the-loop pattern: the AI system identifies potential issues and provides recommendations, but an administrator must review those recommendations before any action is taken. This deliberate inclusion of human judgment in the AI ' s decision or recommendation workflow ensures human oversight is maintained, which is the defining characteristic of the human-in-the-loop process.

Why A is Wrong: Data validation verifies that data meets expected quality standards and formats before being used in AI processing. It is a data quality control activity, not a workflow pattern describing human review of AI recommendations.

Why B is Wrong: Data preparation involves transforming raw data into a format suitable for AI model training or inference. It encompasses cleaning, normalizing, and formatting data, not the process of human review of AI-generated recommendations during system updates.

Why D is Wrong: Model evaluation assesses a model ' s performance against metrics such as accuracy, precision, and recall on test datasets. It is a technical assessment of model quality, not a workflow process where humans review AI-generated recommendations before acting on them.

Domain-specific specialization is a major advantage of using a small language model when cost efficiency matters. CompTIA SecAI+ requires candidates to distinguish both LLMs and SLMs within natural language processing concepts. An SLM can be trained or fine-tuned for a narrowly defined task such as interpreting SIEM events, endpoint terminology, security alerts, and incident patterns. Because it does not require the enormous parameter count and broad general-purpose knowledge of a large model, an SLM can provide efficient task-specific inference while requiring less memory and computational capacity. This can reduce infrastructure and operational costs substantially. Broad dataset training is characteristic of general-purpose foundation models and tends to increase rather than reduce training and compute requirements. Data aggregation is a preprocessing or analytics function and is not an inherent SLM advantage. Data privacy can improve when smaller models are deployed locally, but local deployment is an architectural choice rather than the defining cost-related benefit represented by the available choices. A model concentrated on domain-specific cybersecurity information can therefore deliver the required SIEM-analysis capability more economically than a general-purpose LLM.

Basic Concept: When unauthorized changes to an ML training pipeline cause model degradation, the root cause is insufficient access control and change management around the pipeline. Preventing future occurrences requires implementing governance controls that ensure all pipeline changes are reviewed and approved before execution. CompTIA SecAI+ Study Guide covers MDLC change management controls.

Why B is Correct: Enabling human review and approval workflows in the repository creates a mandatory gate requiring authorized reviewers to examine and approve any changes to training pipeline code before they can be merged and executed. This prevents unauthorized modifications from reaching the pipeline by enforcing a review process where any suspicious or unauthorized changes will be caught and rejected before they affect model training and performance.

Why A is Wrong: Static code scanning analyzes code for vulnerabilities and coding standard violations. While it improves code quality and security, it does not prevent unauthorized individuals from submitting and merging malicious changes to the pipeline without proper review.

Why C is Wrong: Retraining with more data and epochs addresses model performance restoration after the fact but does not prevent future unauthorized pipeline modifications. If the pipeline remains unprotected, the same attack could occur again on the new model.

Why D is Wrong: Keeping multiple model copies enables rapid restoration of a previous version when a deployed model is found to be compromised. While useful for recovery, it is a reactive measure that does not prevent unauthorized pipeline changes from occurring and affecting future model training.

Basic Concept: Advances in AI generative technology have enabled the creation of highly realistic synthetic video and audio content that convincingly impersonates real individuals. These AI-generated impersonations are called deepfakes and represent a significant threat for corporate fraud, misinformation, and social engineering. CompTIA SecAI+ covers deepfakes under basic AI concepts and cybersecurity threats.

Why C is Correct: The scenario describes a video in which the CEO ' s likeness was convincingly replicated to deliver a false message about bankruptcy. This is a textbook deepfake attack — AI-generated synthetic media using the target ' s face and potentially voice to make it appear they said something they never said. Deepfake technology uses GANs and CNNs to create this type of convincing impersonation, and the attack specifically exploited AI generation for deception.

Why A is Wrong: An on-path attack intercepts network communications between two parties to eavesdrop or modify traffic. It requires positioning in the network path, not creating synthetic media of a person ' s likeness.

Why B is Wrong: Phishing is a social engineering attack that uses deceptive messages typically via email to trick recipients into taking harmful actions such as clicking malicious links or providing credentials. While this attack has social engineering elements, the use of AI-generated video to impersonate the CEO specifically categorizes it as a deepfake attack.

Why D is Wrong: Social engineering broadly describes psychological manipulation tactics to deceive individuals. While deepfakes can be used as part of social engineering campaigns, the specific technical technique used — AI-generated synthetic video — makes deepfake the most precise and accurate categorization.

Basic Concept: When API sessions or connections remain active and consuming resources after their intended operations have completed, they create resource leaks that progressively degrade system availability. Session lifecycle management is critical for maintaining AI system health. CompTIA SecAI+ Study Guide covers session management as an availability control for AI systems.

Why B is Correct: Enforcing session expiration ensures that external API sessions and connections are automatically terminated after a defined idle period or maximum duration. This prevents resource-consuming zombie sessions from accumulating and exhausting system memory, thread pools, or connection limits. The observed pattern — persistent unavailability that resolves temporarily with restart — is classic resource leak behavior from sessions that never close, making session expiration the direct fix.

Why A is Wrong: Token limits cap the number of tokens processed per request. While useful for controlling per-request resource consumption, they do not address the root cause of sessions persisting and consuming resources long after their operations complete.

Why C is Wrong: Increasing system memory defers the problem rather than solving it. The leak will eventually consume the additional memory too, requiring another restart. Addressing the root cause through session management is superior to scaling resources to accommodate the leak.

Why D is Wrong: MFA adds an additional authentication factor for users accessing the system. It is a security control for identity verification, not a mechanism for managing session lifecycle or preventing resource exhaustion from lingering sessions.

Basic Concept: Before implementing any security controls for an AI system, especially in a highly regulated sector such as healthcare, a risk assessment must first be conducted to understand the specific threats, vulnerabilities, regulatory obligations, and compliance requirements. CompTIA SecAI+ Study Guide emphasizes risk assessment as the foundational first step in any AI security program.

Why C is Correct: A risk assessment identifies what assets need protection, what threats exist, what regulations apply such as HIPAA for healthcare AI, and what the potential impact of various failure modes would be. In healthcare, this is especially critical given the sensitivity of patient records and strict regulatory requirements. The risk assessment results then inform and prioritize all subsequent security control implementations.

Why A is Wrong: Implementing prompt firewalls is a technical security control appropriate after risks have been identified and prioritized. Deploying controls before conducting a risk assessment may address the wrong threats or miss critical vulnerabilities.

Why B is Wrong: Role-based access management is a security control that should be designed based on identified roles and access requirements discovered during risk assessment. It is an implementation step, not the first step.

Why D is Wrong: Using a secure communication channel is a specific technical control for data in transit. While important, it addresses only one specific risk and should be implemented as part of a comprehensive security strategy informed by a prior risk assessment.

Basic Concept: Security posture management for AI systems involves assessing and improving the overall security state of AI deployments, including identifying risks, implementing controls, and maintaining ongoing compliance. Appropriate frameworks provide structure for this assessment. CompTIA SecAI+ Study Guide identifies NIST AI RMF as the primary framework for AI risk and posture management.

Why B is Correct: The NIST AI Risk Management Framework provides comprehensive, actionable guidance for managing and improving AI security and risk posture across the entire AI lifecycle. It includes the GOVERN, MAP, MEASURE, and MANAGE functions that directly address posture management activities including risk identification, assessment, and control implementation for ML use cases. Its technical depth and ML-specific guidance make it ideal for this summarization task.

Why A is Wrong: OECD standards provide high-level policy principles for AI governance at an international level. They lack the technical specificity and operational guidance needed to summarize posture management impact on a specific ML use case.

Why C is Wrong: The EU AI Act is a regulatory compliance framework establishing legal requirements for AI systems. While it addresses risk management, its focus is on legal compliance rather than technical posture management guidance for ML systems.

Why D is Wrong: A Generative Adversarial Network is an AI architecture for generating synthetic data, not a framework or standard. It has no relevance as a reference for AI security posture management.

Basic Concept: API-based AI systems are susceptible to DoS attacks where excessive requests overwhelm the system ' s ability to respond to legitimate users. Rate limiting is the standard control for preventing both intentional and unintentional API abuse. CompTIA SecAI+ Study Guide covers rate limiting as a key availability control for AI APIs.

Why C is Correct: Rate limiting restricts the number of requests a client can make to an API within a defined time window. In this scenario, a brute-force DoS attack works by sending a massive volume of requests to exhaust the model ' s resources. Rate limiting would have automatically throttled or blocked the excessive request volume, preventing the attack from succeeding and preserving service availability for legitimate hospital users.

Why A is Wrong: Tokenization replaces sensitive data values with non-sensitive placeholders. It is a data security control for protecting sensitive information such as patient identifiers, not a control for managing API request volumes or preventing DoS attacks.

Why B is Wrong: Model guardrails filter and constrain model inputs and outputs for safety and policy compliance. They inspect content quality, not request volume, and cannot prevent a volume-based DoS attack.

Why D is Wrong: A prompt firewall inspects the content of prompts for malicious patterns or policy violations. Like guardrails, it analyzes content rather than controlling request frequency and cannot prevent resource exhaustion from a high-volume brute-force attack.

Basic Concept: Pattern recognition is an AI technique that enables a system to identify recurring sequences or structures within data. In cybersecurity, detecting behavioral patterns such as consistent pre-login failure sequences followed by successful access is critical for threat detection. CompTIA SecAI+ Exam Objectives cover this under AI-assisted security.

Why B is Correct: The scenario describes a highly regular, repeating behavioral pattern — two failures followed by success at a specific time interval, consistently during office hours. Pattern recognition enables the AI model to learn this sequence and flag it as indicative of credential stuffing or an automated brute-force attack with timing controls. ML-driven pattern recognition is specifically designed for such behavioral anomaly detection.

Why A is Wrong: Access management controls who can log in and under what conditions. It enforces authorization policies but does not analyze or detect suspicious behavioral sequences in authentication logs.

Why C is Wrong: Signature matching compares known attack signatures against observed data. The described pattern is behavioral and time-based rather than a known malware or exploit signature, making this technique unsuitable.

Why D is Wrong: Vulnerability analysis identifies weaknesses in systems and code. It does not analyze authentication log sequences or detect behavioral patterns in user activity data.

Basic Concept: An on-path (formerly man-in-the-middle) attack intercepts communication between two parties, allowing the attacker to read, modify, or inject content. In the context of a generative AI application, an on-path attack on the session between user and AI service can manipulate prompts being sent to the model or responses being returned to users. CompTIA SecAI+ covers AI-specific attack vectors under securing AI systems.

Why B is Correct: Session hijacking involves an attacker taking control of an active user session by capturing or forging session tokens. In this attack, the attacker intercepts the communication channel between users and the AI application, allowing them to modify prompts sent to the model or replace legitimate model responses with offensive content. This explains why outputs seem unrelated to prompts and contain offensive material.

Why A is Wrong: Application server hijacking involves gaining unauthorized control of the server hosting the application. While severe, this would typically manifest as complete service disruption or data exfiltration rather than targeted modification of individual user session content.

Why C is Wrong: Domain hijacking involves unauthorized transfer of a domain name registration, redirecting all users to a different IP address. This would affect all users simultaneously and typically redirect to a completely different site rather than manipulating individual AI responses.

Why D is Wrong: Model hijacking refers to attacks that steal or replicate an AI model, not to intercepting and modifying the communication between users and an existing model during active sessions.

Basic Concept: Before ingesting data from multiple sources — both structured and unstructured — into an AI-powered SIEM system, the data must be prepared to ensure quality, consistency, and usability. Data from diverse sources often contains noise, errors, duplicates, and formatting inconsistencies that will degrade AI performance if not addressed. CompTIA SecAI+ covers data preparation as a prerequisite for AI system effectiveness.

Why C is Correct: Data cleansing is the process of detecting and correcting or removing corrupt, inaccurate, incomplete, and duplicate data. Before collecting data from multiple structured and unstructured SIEM sources, engineers must cleanse the data to standardize formats, remove duplicates, fill missing values, and eliminate noise. Clean input data is fundamental to producing accurate AI-generated insights and reliable LLM interactions in the SIEM context.

Why A is Wrong: Balancing addresses class distribution imbalance in labeled training data for classification models. While relevant when training ML detection models, it is not the primary consideration before initial data collection from diverse SIEM sources.

Why B is Wrong: Verification confirms that data meets expected quality standards and validates its accuracy against trusted sources. It is a post-collection quality check performed after cleansing, not the first step before data collection.

Why D is Wrong: Vector storage refers to databases that store embeddings for semantic search, relevant for RAG systems. It is a storage architecture decision made after data is collected, processed, and prepared, not a pre-collection technique.

Basic Concept: Responsible AI principles each address different aspects of trustworthy AI behavior. When stakeholders are concerned about both bias and inconsistency — specifically that the same or equivalent work might receive different treatment from the AI system — the principle of consistency is most directly relevant. CompTIA SecAI+ covers responsible AI principles under governance.

Why C is Correct: Consistency in AI systems means the model applies the same rules, standards, and decision criteria uniformly across all inputs and user groups without variation based on characteristics unrelated to the task. An AI plagiarism detection system that produces inconsistent results across different student submissions or demographic groups fails the consistency principle, which directly addresses both the bias concern (differential treatment) and inconsistency concern the faculty have raised.

Why A is Wrong: Transparency relates to openness about how the AI system works and what data it uses. While valuable for understanding the system, transparency alone does not ensure that the system applies its rules uniformly or consistently.

Why B is Wrong: Explainability means the system can articulate why it made a particular decision. While useful for understanding individual cases, it does not guarantee that decisions are made with equal consistency across different submissions or groups.

Why D is Wrong: Accountability identifies who is responsible for AI system decisions and outcomes. It is a governance principle about ownership and responsibility rather than about ensuring uniform application of evaluation criteria.

Basic Concept: Suspicious queries in AI system logs indicate that potentially malicious or policy-violating prompts are reaching the AI model. Proactively intercepting and filtering suspicious prompts before they are processed requires a prompt-level security control. CompTIA SecAI+ Study Guide identifies prompt firewalls as the appropriate control for blocking suspicious AI queries.

Why A is Correct: A prompt firewall analyzes incoming queries using a combination of pattern matching, semantic analysis, and policy rules to identify and block suspicious prompts before they reach the AI model. It can detect prompt injection attempts, jailbreaking patterns, sensitive data extraction queries, and other suspicious prompt characteristics. By intercepting malicious prompts at the perimeter, it prevents them from influencing model behavior or extracting sensitive information.

Why B is Wrong: Data size controls limit the volume or size of data in requests. While controlling input size can prevent some attacks, it does not analyze the content or semantics of queries to detect suspicious patterns. A small suspicious prompt can be just as harmful as a large one.

Why C is Wrong: Rate limiting controls the frequency of requests from a source. While it can slow down automated attack campaigns, it does not inspect query content for suspicious patterns and allows suspicious queries through as long as they are submitted below the rate threshold.

Why D is Wrong: Agentic AI is an AI architecture for autonomous multi-step task execution. It is a type of AI system, not a security control for filtering suspicious queries from an existing AI system ' s logs.

Basic Concept: As AI-generated images become increasingly indistinguishable from authentic photographs, technical mechanisms are needed to definitively identify synthetic content. Embedding verifiable markers in AI-generated content at creation time provides a reliable provenance trail. CompTIA SecAI+ Study Guide covers digital content authentication and AI provenance controls.

Why D is Correct: Watermarking embeds invisible or visible identifying markers into AI-generated images at the point of creation. These watermarks can be cryptographic, steganographic, or perceptual in nature and survive compression, cropping, and other common image manipulations. When an image ' s provenance is questioned, watermark detection tools can analyze it to confirm AI generation. This is the primary technical mechanism for proving AI image origin as supported by the Content Authenticity Initiative and C2PA standards.

Why A is Wrong: Human validation relies on subjective visual inspection by experts to determine if an image appears AI-generated. Modern AI image generation has advanced to the point where human reviewers frequently cannot distinguish synthetic from authentic images, making this approach unreliable for definitive proof.

Why B is Wrong: Guardrails are input and output filtering controls for AI systems. They restrict what content an AI can generate but do not create verifiable proof of an image ' s AI origin after generation.

Why C is Wrong: Diffusion refers to the diffusion model architecture used to generate images. It describes the creation process, not a verification mechanism. Knowing that diffusion models were used does not help prove a specific image was AI-generated after the fact.

Basic Concept: Preventing prompt injection and ensuring output sanitization requires a control that can inspect both the semantic content of incoming prompts and the safety of outgoing responses. This requires an intelligent, context-aware filtering layer specifically designed for LLM traffic. CompTIA SecAI+ Study Guide identifies LLM firewalls as a primary control for prompt security and output safety.

Why C is Correct: An LLM firewall is specifically designed to inspect, filter, and sanitize both incoming prompts and outgoing AI responses. It can detect and block prompt injection attempts using pattern matching, semantic analysis, and behavioral heuristics, while also sanitizing output to remove sensitive data, harmful content, or policy violations before responses reach users. This dual capability makes it the primary control addressing both requirements simultaneously.

Why A is Wrong: Least privilege restricts what resources and actions users and systems can access. It reduces the potential impact of successful attacks but does not inspect prompt content for injection attempts or sanitize model outputs.

Why B is Wrong: Encryption protects data confidentiality in transit and at rest. It does not analyze prompt content for malicious patterns or filter AI-generated responses for unsafe content. Encrypted traffic can still carry prompt injection attacks.

Why D is Wrong: Rate limiting controls request frequency. While it can slow down automated injection attack campaigns, it does not inspect the content of individual prompts to detect injections, nor does it sanitize output responses. Malicious prompts can still succeed within rate limits.

Basic Concept: LLM API billing is primarily based on token consumption. High costs resulting from staff using powerful, verbose models can be controlled by limiting the maximum tokens processed per interaction and restricting which models staff can access. CompTIA SecAI+ Study Guide covers token management as the primary cost control mechanism for AI deployments.

Why D is Correct: Implementing token limits caps the maximum tokens consumed per API call for both input and output. This directly controls the per-interaction cost by preventing excessively long prompts or overly verbose model responses from generating large token bills. Combined with model tier restrictions, token limits ensure that interactions with powerful models remain within budget constraints regardless of how extensively staff use the service.

Why A is Wrong: Storage monitoring tracks the utilization and performance of data storage systems. Storage costs are separate from LLM API token-based billing and monitoring storage does not address the high API charges resulting from excessive model usage by staff.

Why B is Wrong: Modality types refer to the input formats an AI model accepts such as text, images, audio, or video. While different modalities have different pricing, managing modality types is not the direct cost control lever. Token consumption is the primary cost driver for text-based interactions.

Why C is Wrong: Prompt firewalls inspect and filter prompt content for security and policy compliance. While they can block certain types of queries, they are designed for security purposes, not as financial controls to limit token consumption or enforce cost budgets across staff usage.

Basic Concept: LLM-based chatbots accepting user-uploaded files face two critical risk categories: malicious input injection and resource or cost abuse. CompTIA SecAI+ Study Guide highlights prompt security controls and resource management as key defensive layers for public-facing LLM applications.

Why A is Correct: Prompt guardrails intercept and filter user inputs and model outputs, blocking malicious prompts, prompt injection attempts, and harmful file content before affecting model behavior. Since users can upload files, guardrails are essential for sanitizing and validating that content before processing.

Why D is Correct: Model token quotas directly limit how much of the LLM ' s processing capacity a user can consume. This prevents abuse beyond the SLA, including denial-of-wallet attacks or resource exhaustion through excessively large inputs or repeated requests.

Why B is Wrong: Role-based access controls manage who can access what resources. While useful for internal systems, they do not address malicious input content or enforce LLM resource consumption limits for a public-facing chatbot.

Why C is Wrong: Firewall rules operate at the network layer and can block unauthorized IPs or ports but cannot inspect or filter the semantic content of prompts or control token-level LLM usage.

Basic Concept: Different ML model architectures are optimized for different data types and tasks. Unstructured data such as images, raw network packet captures, and visual content requires models capable of automatically extracting hierarchical spatial features. CompTIA SecAI+ covers ML model selection for security tasks under basic AI concepts.

Why B is Correct: Convolutional Neural Networks (CNNs) are specifically designed for pattern recognition in unstructured data, particularly image and grid-structured data. CNNs use convolutional layers to automatically extract local and hierarchical features without requiring manual feature engineering. They excel at recognizing patterns in raw, unstructured inputs, making them the optimal choice for pattern recognition on unstructured datasets in cybersecurity contexts such as image-based malware analysis or visual traffic pattern recognition.

Why A is Wrong: Long Short-Term Memory (LSTM) networks are recurrent neural networks optimized for sequential and time-series data such as network traffic flows over time or log sequences. While they handle unstructured sequential data, they are not specifically designed for spatial pattern recognition in general unstructured data.

Why C is Wrong: Decision trees work on structured, tabular data with defined features. They require feature extraction and engineering before processing unstructured data and are not designed for raw pattern recognition in unstructured inputs.

Why D is Wrong: Logistic regression is a linear classification algorithm that requires structured, numerical input features. It cannot directly process unstructured data and requires extensive preprocessing and feature extraction, making it unsuitable for pattern recognition on raw unstructured datasets.

Basic Concept: Prompt injection occurs when malicious content embedded in user input manipulates an LLM ' s behavior, causing it to leak sensitive data, bypass restrictions, or execute unintended actions. Preventing such attacks requires mechanisms that inspect and filter content at the prompt level. CompTIA SecAI+ covers LLM-specific security controls extensively.

Why A is Correct: Guardrails are purpose-built controls that inspect, filter, and constrain both input prompts and output responses in LLM systems. They can detect sensitive data patterns such as credit card numbers, block prompt injection payloads, enforce content policies, and prevent the model from processing or outputting restricted information. Guardrails are the primary LLM-native defense against prompt injection as cited in the CompTIA SecAI+ Study Guide.

Why B is Wrong: Antivirus software detects known malware signatures in files and executables. It does not inspect or understand the semantic content of LLM prompts and cannot detect or block prompt injection attacks.

Why C is Wrong: A WAF operates at the HTTP layer inspecting web requests and responses against rule sets. While it can block some patterns, it lacks the contextual intelligence to understand LLM prompt semantics and cannot prevent sophisticated injection attacks.

Why D is Wrong: Role-based access control manages who can access which resources. It controls authorization but does not inspect the content of prompts to prevent injection attacks once a user has legitimate access.

Overfitting is the principal risk when a model is trained for too many epochs on the same training data. An epoch represents one complete pass through the training dataset. CompTIA SecAI+ includes epochs under fine-tuning and model-training techniques, making the relationship between repeated training and model performance an important foundational concept. During appropriate training, repeated epochs allow the model to learn useful patterns. However, excessive training can cause the model to fit not only meaningful relationships but also noise, anomalies, and idiosyncrasies in the training set. The model can then perform extremely well against familiar training samples while generalizing poorly to previously unseen data. This condition is overfitting. Output unpredictability is not the defining consequence of excessive epochs. Insufficient system-usage monitoring is an operational observability problem and has no direct relationship to the number of training epochs. Unit testing concerns software and pipeline validation rather than statistical generalization. Proper controls include validation datasets, evaluation throughout training, early stopping, regularization, and monitoring the divergence between training and validation performance. Therefore, overfitting is the most technically accurate answer.

================

Basic Concept: Prompting techniques for LLMs include various approaches to guide model behavior. Providing examples within prompts is a powerful technique that leverages the model ' s in-context learning capability to guide response format and quality. CompTIA SecAI+ Study Guide covers prompting techniques under basic AI concepts.

Why E is Correct: One-shot prompting involves providing exactly one example within a prompt to demonstrate to the model the desired input-output format or response style. This single example guides the model ' s understanding of the task without requiring extensive fine-tuning. It is a well-established prompting technique that uses examples to inform model behavior.

Why F is Correct: Multi-shot prompting (also called few-shot prompting) involves providing multiple examples within a prompt to further clarify the desired output pattern. Multiple examples help the model identify consistent patterns and produce more accurate, consistent responses. Both one-shot and multi-shot are specifically defined by their use of examples in prompts.

Why A is Wrong: A user prompt is the input message submitted by a user to the AI system. It is the general term for any user input, not a specific technique that describes the practice of providing examples.

Why B is Wrong: A system prompt sets the model ' s behavior, persona, and constraints at the session level. While a system prompt could contain examples, the term specifically refers to the system-level instruction context, not the technique of example provision.

Why C is Wrong: A prompt template is a reusable structured format with placeholders for variable inputs. It standardizes prompt structure but is not defined by the practice of including examples.

Why D is Wrong: Quantization is a model compression technique that reduces model size by representing weights with lower precision numbers. It is a model optimization technique completely unrelated to prompting practices.

Basic Concept: When AI models are deployed in production, they interact with real data including sensitive business information, personal data, and confidential records. The intersection of AI capabilities and sensitive data creates significant security risks. CompTIA SecAI+ Exam Objectives identify data exposure as the primary production security risk for AI deployments.

Why D is Correct: Data exposure is the primary security risk in production AI deployments. AI models in production process sensitive data through queries and responses, and vulnerabilities such as prompt injection, model inversion attacks, insecure output handling, and misconfigured access controls can expose confidential training data, user PII, proprietary information, or system credentials. The consequences include regulatory violations, legal liability, and reputational damage, making data exposure the most critical ongoing security concern.

Why A is Wrong: GPU acceleration is a performance optimization technique that uses graphics processors for faster AI computation. While hardware security is important, GPU acceleration itself is not a security risk — it is a performance feature that does not inherently expose data.

Why B is Wrong: Model overfitting is a model quality issue where a model performs poorly on new data after memorizing training data too specifically. While it can indirectly contribute to data memorization, it is primarily a performance and generalization concern during development rather than a primary production security risk.

Why C is Wrong: Model encryption is a security control used to protect AI model weights from unauthorized access, not a risk itself. Framing a protection mechanism as a primary risk conflates controls with threats.

Option B is correct because a human resources AI application processes highly sensitive employee and applicant information and may influence consequential employment decisions. Privacy and security therefore determine whether the system can be deployed responsibly. The design must address data minimization, lawful collection, access control, encryption, retention, auditability, secure model integration, and protection against unauthorized disclosure. It should also evaluate whether training data or outputs expose protected or confidential information and whether human review is required before decisions affect candidates or employees. Option A concerns friendliness and presentation, not the protection of people or organizational data. Option C may improve usability, but visual design cannot compensate for insecure handling of resumes, performance records, or identity data. Option D can make outputs easier to consume, yet brevity and summarization do not control privacy risk or system compromise. The NIST AI Risk Management Framework identifies privacy enhancement, security, resilience, accountability, and transparency as characteristics of trustworthy AI, making privacy and security the controlling considerations in this scenario.

================

Basic Concept: LLM API pricing is primarily based on token consumption — the number of tokens processed in both input prompts and output responses. Controlling token usage is the most direct lever for managing and reducing LLM API costs. CompTIA SecAI+ Study Guide covers AI cost management and resource controls under securing AI systems.

Why D is Correct: Adjusting token limits directly caps the maximum number of tokens used per request for both input and output. By setting appropriate token limits, the organization prevents excessively long prompts or verbose responses from consuming unnecessary tokens, directly translating to lower API costs and providing hard budget control.

Why A is Wrong: Prompt templates standardize how queries are structured, which can indirectly improve efficiency. However, they do not enforce a hard cap on token usage and cannot prevent costs from escalating with large volumes or verbose responses.

Why B is Wrong: Increasing CPU and memory addresses computational infrastructure performance on the client side. LLM API costs are billed by the API provider based on token usage, not on the client ' s hardware resources.

Why C is Wrong: Reducing model size means using a smaller, less powerful model version. While this may lower cost per token, it is a model selection decision, not an ongoing operational control that can be adjusted to manage cost in real time.

Basic Concept: Security assessment of AI-generated code requires a systematic review of the code itself to understand what has been generated and identify potential vulnerabilities before remediation steps are taken. Security assessments follow a structured methodology beginning with understanding the current state. CompTIA SecAI+ Study Guide covers AI-generated code security assessment under AI-assisted security.

Why D is Correct: Performing a source code review is the first and most fundamental step in assessing AI-generated code security. Before removing secrets, enforcing access controls, or scanning for sensitive data, the analyst must understand what the AI tool has generated by reviewing the code for security vulnerabilities, insecure patterns, logic flaws, and policy violations. The review provides the baseline knowledge needed to prioritize and direct all subsequent remediation actions.

Why A is Wrong: Removing hard-coded secrets is a specific remediation action for a specific finding. This step should come after the source code review has identified the presence and location of hard-coded secrets, not before the initial assessment reveals whether they exist.

Why B is Wrong: Enforcing access controls for code repositories is a security hardening measure for the repository infrastructure. It protects access to existing code but does not constitute an assessment of what the AI tool has generated from a security standpoint.

Why C is Wrong: Enabling sensitive data discovery scans repositories for PII and sensitive information patterns. While valuable as part of the assessment, it is a specific automated scanning tool best used after or alongside a manual code review that provides contextual understanding of the codebase.

Basic Concept: Threat modeling for AI systems requires a framework specifically designed to address AI-specific attack techniques, tactics, and procedures. General cybersecurity or governance frameworks do not capture the unique adversarial attack surface of AI and ML systems. CompTIA SecAI+ Exam Objectives identify MITRE ATLAS as the primary AI threat modeling resource.

Why B is Correct: MITRE ATLAS (Adversarial Threat Landscape for AI Systems) is specifically designed as an AI and ML threat modeling framework. It catalogs real-world adversarial tactics, techniques, and procedures targeting AI systems, enabling security architects to identify and assess threats unique to ML models such as data poisoning, model extraction, and evasion attacks. It is the industry standard for AI-specific threat modeling.

Why A is Wrong: Responsible AI is a set of ethical principles and governance guidelines for developing and deploying AI systems fairly and safely. It addresses ethics and fairness, not technical adversarial threat modeling.

Why C is Wrong: The OECD provides non-binding policy recommendations and principles for AI governance at an international level. It does not provide technical threat modeling taxonomies or AI-specific attack catalogs.

Why D is Wrong: ISO standards such as ISO 42001 establish management system requirements for AI governance. They are compliance and management frameworks, not threat modeling tools for identifying adversarial AI attack vectors.

Basic Concept: Generative AI systems in healthcare settings incur costs from multiple operational activities. Understanding the cost drivers specific to generative AI helps administrators implementaccurate cost monitoring and controls. CompTIA SecAI+ Study Guide covers AI cost management under securing AI systems.

Why C is Correct: Storage retrieval and prompt processing are the two primary cost drivers for generative AI systems in healthcare. Storage retrieval refers to the cost of querying vector databases or document stores in RAG-based AI systems to fetch relevant patient records, clinical guidelines, or historical data for context. Prompt processing encompasses the token-based cost of the LLM processing the combined retrieved content and user query to generate a response. Together these two activities represent the billable units that drive generative AI costs in healthcare RAG deployments, making them the most accurate basis for cost calculation and monitoring.

Why A is Wrong: Connection access and exchange gateway costs relate to network infrastructure and API gateway usage fees. While there may be minor costs associated with API calls, these are not the primary cost drivers for generative AI systems where the dominant expenses are computational token processing and data retrieval operations.

Why B is Wrong: Encryption and decryption processing costs relate to cryptographic operations for data security. While encryption is important for healthcare data protection under HIPAA, cryptographic processing overhead is minimal compared to the substantial token-based LLM processing and storage retrieval costs that dominate generative AI operational expenses.

Why D is Wrong: Catalog servicing and exchange processing are terms associated with data catalog management and data exchange infrastructure. These are not recognized primary cost components of generative AI systems in healthcare, where storage retrieval and token-based prompt processing are the established cost measurement criteria.

Basic Concept: Creating realistic deepfake videos that convincingly replicate a real person ' s facial expressions, movements, and voice requires deep learning models capable of learning and synthesizing complex spatial and temporal features from existing video data. CompTIA SecAI+ covers deepfake technologies under basic AI concepts.

Why B is Correct: Convolutional Neural Networks are foundational to deepfake video generation. CNNs excel at learning spatial features from visual data and are used within deepfake architectures to analyze source and target faces, extract facial features, and synthesize realistic face swaps or face animations. Modern deepfake systems typically combine CNNs with autoencoders and GANs to generate convincing video content showing a person saying or doing things they never did.

Why A is Wrong: Statistical analysis involves mathematical methods for analyzing data distributions and relationships. It does not have the capability to generate synthetic video content or replicate a person ' s visual likeness in motion.

Why C is Wrong: An ML classifier assigns input data to predefined categories. Classification models detect and label content rather than generating new synthetic video content of a person ' s likeness. They are detection tools, not generation tools.

Why D is Wrong: Random forest is an ensemble ML method using multiple decision trees for classification and regression tasks. It works on structured, tabular data and cannot process or generate visual, spatial data needed for realistic deepfake video synthesis.

Basic Concept: Different ML learning paradigms handle different data situations. The availability of labeled versus unlabeled data determines which learning approach is appropriate. Building clustering models specifically requires learning from data without predefined category labels. CompTIA SecAI+ Study Guide covers ML learning paradigms under basic AI concepts.

Why C is Correct: Unsupervised learning works with unlabeled data by discovering inherent patterns, structures, and groupings within the data without predefined categories. Clustering is the canonical unsupervised learning task, where algorithms like k-means, hierarchical clustering, or DBSCAN group similar data points together based on feature similarity. Since the data scientist has unlabeled data and wants to find natural groupings, unsupervised learning is the appropriate and correct technique.

Why A is Wrong: Supervised learning requires labeled training data where each example has a corresponding correct output label. The data scientist explicitly has unlabeled data, making supervised learning inapplicable without first completing the labor-intensive task of manually labeling all examples.

Why B is Wrong: Reinforcement learning trains agents to take actions in an environment to maximize cumulative rewards through trial and error. It is designed for sequential decision-making problems, not for finding groupings in static, unlabeled datasets.

Why D is Wrong: Semi-supervised learning combines a small amount of labeled data with a large amount of unlabeled data. It requires at least some labels to guide learning. The scenario specifies working with unlabeled data only, making unsupervised learning the pure fit.

Basic Concept: In a layered AI system security architecture, access control must be established at each layer, beginning from the outermost point of entry. Authentication must be established at the endpoint level first, as this is the first point of interaction between users and the AI system. CompTIA SecAI+ Study Guide establishes endpoint authentication as the initial access control layer for AI systems.

Why D is Correct: Endpoint access control is the first authentication control to implement because it governs the initial connection from user devices or client applications to the AI system. All subsequent access layers including server access, model access, and data access depend on the endpoint being authenticated first. Establishing endpoint authentication ensures that only authorized endpoints can initiate sessions and proceed through subsequent authentication layers.

Why A is Wrong: Model access controls govern who can query, update, or access the AI model ' s parameters and functions. This control layer is implemented after endpoint authentication has been established, as it applies to requests that have already been authenticated at the endpoint level.

Why B is Wrong: Server access controls manage access to the computing infrastructure hosting the AI system. While critical for infrastructure security, server-level controls are configured by administrators and are not the first authentication control for end-user access flows.

Why C is Wrong: Data access controls define what data the AI system and its users can read, write, or query. These are implemented at a deeper layer after endpoint and potentially model authentication have verified that the requester is authorized to interact with the system at all.

Basic Concept: Cybersecurity threats evolve continuously, with new attack variants emerging regularly. The choice between traditional statistical models and ML models for attack classification depends on which better handles the complexity and dynamism of the threat landscape. CompTIA SecAI+ covers ML model advantages for cybersecurity under basic AI concepts.

Why A is Correct: ML models can learn arbitrarily complex, non-linear relationships from training data and adapt to new patterns when retrained with updated data. For attack classification, this means ML can recognize sophisticated, multi-feature attack patterns that exceed the capabilities of simple statistical models and can be updated to detect new attack variants as the threat landscape evolves. This adaptability to complex and changing problems is the primary reason analysts prefer ML over static statistical approaches.

Why B is Wrong: ML model development pipelines are generally more complex than statistical models, requiring data preparation, feature engineering, model selection, training, validation, and deployment steps. Simplicity of development is not a characteristic advantage of ML over statistical models.

Why C is Wrong: ML models typically require large amounts of training data to perform well. Statistical models often perform better than ML with small datasets. Performance with small datasets is actually an advantage of statistical models over ML, not ML over statistical.

Why D is Wrong: Community support and expert availability are ecosystem considerations rather than technical reasons to prefer ML for cybersecurity classification tasks. These factors might influence tool selection but do not explain the fundamental technical preference for ML ' s superior handling of complex attack patterns.

OWASP is the best answer because it publishes the Top 10 for Large Language Model Applications, a resource specifically structured around significant security vulnerabilities and attack patterns affecting LLM-based applications. CompTIA SecAI+ Objective 2.1 directly lists the OWASP Top 10 and specifically the LLM Top 10 as an AI threat-modeling resource. MITRE ATLAS is also highly relevant to AI security, but its purpose is broader: it catalogs adversarial tactics, techniques, and procedures associated with attacks against AI and machine learning systems. The MIT AI Risk Repository aggregates a broad range of AI risks rather than functioning as the recognized LLM application vulnerability Top 10. NIST AI RMF provides a risk-management framework for governing and managing AI risks but is not primarily a vulnerability catalog for LLM applications. Since the question asks for the resource that most directly captures widely recognized threat vectors and common vulnerabilities associated specifically with LLM applications, OWASP ' s LLM-focused Top 10 is the most precise match.

Basic Concept: Threat modeling for any system, and especially for agentic AI systems with multiple interacting components, begins with understanding the system ' s architecture and where trust boundaries exist. Trust boundaries define where data and control flows cross between components with different trust levels, representing potential attack surfaces. CompTIA SecAI+ Study Guide aligns with STRIDE and MITRE ATLAS threat modeling methodologies.

Why B is Correct: Identifying trust boundaries between components is the foundational first step in threat modeling. Agentic systems often involve multiple components such as the orchestrator, tools, APIs, data sources, and external services with different trust levels. Understanding where these boundaries exist reveals where untrusted inputs cross into trusted components, enabling the architect to systematically identify threats at each boundary before proceeding to risk quantification and control application.

Why A is Wrong: Applying compensating controls based on exposure findings is the final step in threat modeling, occurring after threats have been identified and risks quantified. Controls cannot be appropriately designed without first understanding the system ' s trust boundaries and threat landscape.

Why C is Wrong: Calculating risk to resources based on data sensitivity is a risk assessment step that occurs after trust boundaries are mapped and potential threats are identified. Risk quantification requires knowing what threats exist at each boundary first.

Why D is Wrong: Scanning for OWASP Top 10 vulnerabilities is a technical vulnerability assessment activity. While valuable, it comes after the architectural analysis of trust boundaries and threat identification phases of threat modeling.

Model validation is intended to determine systematically whether an AI or machine learning model performs according to its defined objectives and produces acceptable results. This requires a formal methodology, appropriate validation data, measurable criteria, and comparison of actual model behavior against expected outcomes. CompTIA SecAI+ explicitly includes model validation among model-training techniques and also places validation, model evaluation, monitoring, and feedback within the AI lifecycle. Option C therefore represents the actual purpose of validation: objectively establishing whether the model satisfies predetermined performance and quality requirements. Providing feedback between interactions is closer to feedback and iteration or model adaptation rather than validation itself. Checking an exported file is merely an output or serialization check and does not evaluate model quality comprehensively. CI/CD integration checks establish whether software components build, integrate, or deploy correctly; a model could successfully pass those technical checks while still producing inaccurate, biased, unsafe, or otherwise unacceptable predictions. Formal measurement against expected outcomes is therefore the distinguishing characteristic of proper model validation.

================

Basic Concept: LLM firewalls inspect prompts and responses to identify malicious content, policy violations, and attack attempts. To detect known attack patterns, these systems apply inspection techniques that compare content against established threat indicators. CompTIA SecAI+ Study Guide covers LLM security monitoring and detection techniques.

Why A is Correct: Signature matching compares incoming prompts and outgoing responses against a library of known attack signatures, including common prompt injection patterns, jailbreaking attempts, data exfiltration queries, and known malicious payload strings. When content matches a known attack signature, the LLM firewall can block or flag it. Signature matching is an efficient, proven detection technique for identifying known attack patterns traversing an LLM firewall.

Why B is Wrong: Distributed denial-of-service is itself a type of attack, not a detection technique. DDoS floods systems with traffic to cause service unavailability and has no role in detecting attacks through an LLM firewall.

Why C is Wrong: Translation analysis involves converting content between languages or formats. While it might be used to detect obfuscated attacks in different encodings, it is not a standard detection technique for identifying attacks crossing an LLM firewall.

Why D is Wrong: Vulnerability enumeration systematically identifies and catalogs vulnerabilities in systems or applications during security assessments. It is an assessment activity used to discover weaknesses, not a real-time detection technique for attacks traversing an LLM firewall.

Option D is correct because the AI application is taking high-impact remediation actions without adequate human authorization. Human-in-the-loop control requires a qualified person to review and approve consequential actions, such as stopping servers or changing firewall exposure, before execution. The system may still detect misconfigurations, rank risks, and recommend fixes, but it should not autonomously disrupt development or alter network access during a demonstration. Option A restores service temporarily but leaves the unsafe decision path unchanged. Option B removes useful monitoring and does not correct the excessive automation. Option C might close improperly opened ports, yet it addresses only one symptom and does not prevent the AI from repeating other damaging actions. The stronger design is to roll back the unauthorized changes, restrict the agent’s permissions, require approval for disruptive operations, and maintain an auditable record of recommendations and approvals. The NIST AI Risk Management Framework calls for appropriate human-oversight processes to be defined, assessed, and documented, supporting this governance pattern.

================

Basic Concept: Dynamic Application Security Testing (DAST) tests running applications by sending various inputs to discover vulnerabilities. AI can significantly enhance DAST by intelligently generating diverse, targeted test payloads that traditional tools might miss. CompTIA SecAI+ covers AI augmentation of security testing methodologies.

Why C is Correct: Payload creation is highly suitable for AI automation during DAST. AI can generate diverse, contextually appropriate attack payloads such as SQL injection strings, XSS vectors, command injection attempts, and format string exploits tailored to the specific application ' s behavior observed during testing. AI can learn from the application ' s responses to previous payloads and generate increasingly targeted inputs, discovering vulnerabilities more efficiently than static payload databases.

Why A is Wrong: DDoS attacks are volume-based attacks designed to overwhelm network or application infrastructure. Automating DDoS during DAST is inappropriate as it would disrupt service availability rather than discover application security vulnerabilities, and it is harmful to legitimate operations.

Why B is Wrong: Data poisoning is an attack targeting AI/ML model training data integrity. It is relevant to securing AI systems but is not a DAST technique for testing web or software application security vulnerabilities during dynamic testing.

Why D is Wrong: Threat modeling is a structured analysis process performed before development or testing to identify potential threats and design appropriate countermeasures. It is a planning activity, not an attack technique that can be automated during dynamic application security testing.