Visit here for our full IAPP AIGP exam dumps and practice test questions.
Question 16
What is the primary purpose of an AI impact assessment?
A) To measure the financial return on AI investments
B) To evaluate potential risks, benefits, and ethical implications of AI systems before deployment
C) To calculate processing speed of AI algorithms
D) To determine the number of users for an AI system
Answer: B
Explanation:
An AI impact assessment is a systematic evaluation process designed to identify, analyze, and document the potential risks, benefits, and ethical implications of an AI system before it is deployed in production environments. This assessment serves as a critical governance tool that helps organizations make informed decisions about whether and how to deploy AI systems while considering potential impacts on individuals, communities, and society.
The assessment process typically examines multiple dimensions including fairness and bias potential, privacy implications, security risks, transparency and explainability requirements, accountability mechanisms, safety considerations, and broader societal impacts. It involves analyzing how the AI system makes decisions, what data it uses, who might be affected by its outputs, and what harms could result from errors or misuse.
AI impact assessments align with responsible AI principles by requiring organizations to proactively consider consequences before deployment rather than reactively addressing problems after harm occurs. The assessment process often involves stakeholder consultation, including input from affected communities, domain experts, ethicists, and other relevant parties. Assessment findings inform decisions about system design modifications, deployment conditions, monitoring requirements, and whether deployment should proceed at all.
Option A is incorrect because while financial considerations may be part of broader project evaluation, AI impact assessments focus specifically on risks, ethics, and societal impacts rather than return on investment. Option C is wrong as technical performance metrics like processing speed are separate from impact assessment which examines consequences and implications. Option D is incorrect because user volume is an operational consideration, not the focus of impact assessment which examines potential harms and benefits.
Understanding AI impact assessments is fundamental to implementing responsible AI governance that identifies and mitigates potential harms before deployment.
Question 17
Which principle requires that AI systems provide explanations for their decisions that stakeholders can understand?
A) Accuracy
B) Efficiency
C) Explainability
D) Scalability
Answer: C
Explanation:
Explainability is the principle that requires AI systems to provide understandable explanations for their decisions, predictions, or recommendations to relevant stakeholders including users, affected individuals, auditors, and regulators. This principle recognizes that trust, accountability, and effective oversight require understanding how AI systems reach their conclusions rather than treating them as inscrutable black boxes.
Explainability encompasses both technical interpretability (the degree to which humans can understand the cause of a decision) and the ability to communicate that understanding effectively to diverse audiences with varying levels of technical expertise. Different stakeholders may require different types of explanations, from high-level summaries for end users to detailed technical analyses for auditors or data scientists investigating potential issues.
The importance of explainability varies by context and risk level. High-stakes applications such as healthcare diagnosis, credit decisions, criminal justice, and employment determinations typically require strong explainability to support accountability, enable appeals or challenges, detect bias, and build justified trust. Lower-risk applications may require less detailed explanations, though transparency about AI involvement remains important.
Option A is incorrect because accuracy refers to the correctness of AI predictions or outputs, not the ability to explain how those outputs were generated. Option B is wrong as efficiency relates to resource utilization and speed, not explainability of decisions. Option D is incorrect because scalability concerns the ability to handle increased workload or data volume, not the provision of understandable explanations.
Understanding explainability is essential for implementing AI systems that support accountability, enable effective oversight, and maintain stakeholder trust through transparency.
Question 18
What is the primary concern regarding algorithmic bias in AI systems?
A) That algorithms process data too slowly
B) That algorithms may systematically produce unfair outcomes for certain groups
C) That algorithms require too much computational power
D) That algorithms cannot handle large datasets
Answer: B
Explanation:
Algorithmic bias refers to systematic and repeatable errors in AI systems that create unfair outcomes, particularly disadvantaging certain demographic groups defined by characteristics such as race, gender, age, disability status, or other protected attributes. This concern is central to responsible AI governance because biased systems can perpetuate or amplify existing societal inequalities and cause significant harm to affected individuals and communities.
Bias can enter AI systems through multiple pathways including biased training data that does not represent all groups fairly, problematic data labeling that reflects human prejudices, algorithm design choices that optimize for overall accuracy at the expense of fairness across groups, and deployment contexts that differ from training conditions. Historical bias in training data can cause AI systems to learn and perpetuate discriminatory patterns from the past.
The impacts of algorithmic bias can be severe and wide-reaching, affecting access to credit, employment opportunities, healthcare quality, criminal justice outcomes, educational resources, and many other important life domains. Addressing algorithmic bias requires comprehensive approaches including diverse training data, fairness-aware algorithm design, bias testing across demographic groups, ongoing monitoring for disparate impacts, and mechanisms for redress when bias is identified.
Option A is incorrect because processing speed is a performance consideration unrelated to the fairness concerns that define algorithmic bias. Option C is wrong as computational requirements are resource constraints, not bias issues. Option D is incorrect because data volume handling is a scalability concern, whereas algorithmic bias specifically addresses unfair treatment of different groups.
Understanding algorithmic bias is critical for AI governance professionals responsible for ensuring AI systems treat all individuals and groups fairly.
Question 19
Which framework provides guidance specifically for trustworthy AI development?
A) ISO 27001
B) NIST AI Risk Management Framework
C) GDPR
D) SOC 2
Answer: B
Explanation:
The NIST AI Risk Management Framework (AI RMF) provides comprehensive guidance specifically designed for developing and deploying trustworthy AI systems by helping organizations identify, assess, and manage AI-related risks throughout the AI lifecycle. Released by the National Institute of Standards and Technology, this framework addresses the unique challenges of AI systems including opacity, complexity, and potential for unintended consequences.
The NIST AI RMF is organized around four core functions: Govern, Map, Measure, and Manage. The Govern function establishes organizational structures and processes for responsible AI development. The Map function identifies context and categorizes AI risks. The Measure function assesses and analyzes identified risks. The Manage function implements responses to address mapped and measured risks. These functions work together to create a comprehensive approach to AI risk management.
The framework emphasizes characteristics of trustworthy AI including validity and reliability, safety, security and resilience, accountability and transparency, explainability and interpretability, privacy enhancement, and fairness with harmful bias managed. It provides a flexible, risk-based approach that can be adapted to different organizational contexts, AI applications, and risk tolerance levels while promoting consistency with existing risk management practices.
Option A is incorrect because ISO 27001 addresses information security management broadly, not AI-specific trustworthiness concerns. Option C is wrong as GDPR is data protection regulation, not an AI development framework, though it has implications for AI systems processing personal data. Option D is incorrect because SOC 2 addresses service organization controls for security and privacy, not specifically AI trustworthiness.
Understanding AI-specific frameworks like the NIST AI RMF is essential for implementing comprehensive governance that addresses unique AI risks and challenges.
Question 20
What is the purpose of model cards in AI governance?
A) To store AI models efficiently
B) To provide standardized documentation about AI models including intended use, performance, and limitations
C) To encrypt AI model parameters
D) To increase model processing speed
Answer: B
Explanation:
Model cards are standardized documentation artifacts that provide transparent, structured information about AI models including their intended use cases, performance characteristics across different demographic groups, limitations, ethical considerations, and other critical details that stakeholders need to make informed decisions about model deployment and use. Model cards promote transparency and accountability by making model characteristics explicit and accessible.
A comprehensive model card typically includes information about model development including training data sources and characteristics, model architecture and design decisions, intended use cases and users, out-of-scope uses that should be avoided, performance metrics overall and disaggregated by demographic groups, fairness and bias assessments, limitations and known weaknesses, ethical considerations, and recommendations for responsible use. This documentation supports informed decision-making by downstream users and deployers.
Model cards serve multiple governance purposes including enabling appropriate model selection for specific use cases, facilitating bias and fairness assessments, supporting regulatory compliance and audit requirements, promoting responsible model reuse by clarifying intended contexts, and documenting model provenance and accountability. They complement other governance tools like datasheets for datasets and system cards for deployed AI systems.
Option A is incorrect because model cards are documentation tools, not storage mechanisms; model storage is a separate technical infrastructure concern. Option C is wrong as encryption is a security technique unrelated to model cards which focus on transparency through documentation. Option D is incorrect because model cards do not affect computational performance; they document model characteristics for governance purposes.
Understanding model cards and similar documentation practices is important for implementing transparent AI governance that enables informed decision-making by stakeholders.
Question 21
Which approach helps address the challenge of AI systems making decisions that humans cannot understand?
A) Increasing model complexity
B) Using only proprietary algorithms
C) Implementing explainable AI (XAI) techniques
D) Removing all documentation
Answer: C
Explanation:
Explainable AI (XAI) encompasses techniques and methods designed to make AI system decisions, predictions, and behaviors understandable to humans, addressing the black box problem where complex models produce accurate results through opaque processes that cannot be easily interpreted. XAI techniques help stakeholders understand why systems made particular decisions, how different factors influenced outcomes, and when systems might be unreliable.
XAI approaches vary based on model type and explanation needs. Techniques include feature importance methods that identify which input features most influenced a decision, attention mechanisms that highlight which parts of input data the model focused on, counterfactual explanations showing how changing inputs would alter outcomes, rule extraction that approximates complex models with interpretable rules, and visualization techniques that make model behavior more transparent. Some approaches provide local explanations for individual predictions while others offer global understanding of overall model behavior.
The benefits of explainability extend across multiple governance objectives including building justified trust in AI systems, enabling detection and debugging of bias and errors, supporting accountability by clarifying responsibility for decisions, facilitating regulatory compliance with explanation requirements, empowering users to make informed decisions about AI recommendations, and enabling effective human oversight of AI systems. Different stakeholders may require different types and levels of explanation.
Option A is incorrect because increasing complexity typically reduces rather than improves explainability; simpler models are generally more interpretable. Option B is wrong as proprietary algorithms often reduce transparency; open approaches generally support better explainability. Option D is incorrect because removing documentation would eliminate rather than improve understanding; comprehensive documentation supports explainability.
Understanding XAI techniques is essential for AI governance professionals working to ensure AI systems are transparent, accountable, and trustworthy.
Question 22
What is the primary purpose of an AI ethics committee in an organization?
A) To increase AI processing speed
B) To provide governance oversight and guidance on ethical AI development and deployment
C) To reduce cloud computing costs
D) To manage IT infrastructure
Answer: B
Explanation:
An AI ethics committee serves as a governance body that provides organizational oversight, guidance, and decision-making support for ethical considerations in AI development and deployment. This multidisciplinary committee typically includes diverse perspectives from technical experts, ethicists, legal professionals, business leaders, and potentially external stakeholders, ensuring comprehensive evaluation of AI initiatives from multiple viewpoints.
The committee’s responsibilities often include reviewing proposed AI projects for ethical implications, providing guidance on ethical AI design and development practices, evaluating AI impact assessments and recommending mitigation strategies, resolving ethical dilemmas that arise during AI development or deployment, establishing and maintaining organizational AI ethics policies and principles, promoting ethical AI culture through education and awareness, and monitoring deployed AI systems for emerging ethical concerns. The committee provides a structured forum for deliberation on complex ethical questions.
Effective AI ethics committees operate with clear authority, defined processes for escalation and decision-making, adequate resources and organizational support, and regular communication with leadership and development teams. They balance innovation with responsibility, helping organizations navigate ethical challenges while enabling beneficial AI deployment. The committee should have sufficient authority to influence decisions and the ability to escalate concerns when necessary.
Option A is incorrect because processing speed is a technical performance consideration unrelated to the ethical oversight function of an ethics committee. Option C is wrong as cost management is a financial concern, not an ethics committee responsibility. Option D is incorrect because IT infrastructure management is an operational function separate from ethical governance oversight.
Understanding the role of AI ethics committees is important for establishing effective governance structures that embed ethical considerations into AI development processes.
Question 23
Which data governance practice is most important for ensuring AI model quality?
A) Using only unstructured data
B) Ensuring high-quality, representative, and appropriately labeled training data
C) Maximizing data storage capacity
D) Minimizing data security measures
Answer: B
Explanation:
Ensuring high-quality, representative, and appropriately labeled training data is fundamental to AI model quality because models learn patterns from training data and will perpetuate any biases, errors, or limitations present in that data. The principle of “garbage in, garbage out” is particularly acute in AI systems where data quality issues can result in models that perform poorly, exhibit bias, or fail in unexpected ways when deployed.
High-quality training data requires attention to multiple dimensions including accuracy of data values and labels, completeness with minimal missing or corrupted data, consistency in format and representation across the dataset, relevance to the intended use case and deployment context, representativeness ensuring all important groups and scenarios are adequately included, and timeliness reflecting current rather than outdated patterns. Poor quality in any dimension can significantly degrade model performance and fairness.
Data governance practices that support model quality include establishing data quality standards and validation processes, implementing systematic data collection and labeling procedures with appropriate quality control, documenting data provenance and lineage to understand data sources and transformations, regularly assessing training data for representation of relevant populations and scenarios, maintaining appropriate data versioning for reproducibility, and continuously monitoring for data drift that might degrade deployed model performance over time.
Option A is incorrect because data structure type (structured vs. unstructured) is less important than quality and representativeness; both types can be used effectively depending on the use case. Option C is wrong as storage capacity is an infrastructure concern that does not directly ensure data quality. Option D is incorrect because minimizing security would be counterproductive; proper security protects data integrity which supports quality.
Understanding the critical role of data quality in AI governance helps organizations implement practices that ensure models are built on sound foundations.
Question 24
What is the purpose of algorithmic accountability?
A) To maximize algorithm complexity
B) To ensure clear assignment of responsibility for AI system outcomes and effective mechanisms for oversight
C) To reduce algorithm transparency
D) To eliminate human involvement in AI systems
Answer: B
Explanation:
Algorithmic accountability ensures that clear responsibility is assigned for AI system outcomes and that effective mechanisms exist for oversight, redress, and correction when problems arise. This principle recognizes that AI systems can cause significant impacts on individuals and society, requiring identifiable parties who are responsible for ensuring systems operate appropriately and addressing harms when they occur.
Accountability encompasses multiple elements including clear assignment of roles and responsibilities throughout the AI lifecycle from development through deployment and monitoring, documentation of decisions and rationale for design choices, explainability enabling understanding of how systems reach decisions, auditability through logging and records that enable investigation of system behavior, mechanisms for affected individuals to challenge decisions and seek redress, processes for detecting and addressing errors or bias, and organizational structures with authority to enforce accountability requirements.
Implementing algorithmic accountability requires both technical and organizational measures. Technical measures include logging system decisions and confidence levels, maintaining audit trails, implementing human review processes for high-stakes decisions, and building in mechanisms for explanation and appeal. Organizational measures include governance frameworks defining roles and responsibilities, policies establishing accountability requirements, training for personnel involved in AI development and deployment, and oversight mechanisms including audits and impact assessments.
Option A is incorrect because accountability is not about complexity but about responsibility; complexity often reduces rather than enhances accountability. Option C is wrong as transparency typically supports accountability by making system behavior visible for oversight. Option D is incorrect because human oversight and responsibility are central to accountability; eliminating human involvement would undermine rather than support accountability.
Understanding algorithmic accountability is essential for governance frameworks that ensure responsible parties can be identified and held answerable for AI system impacts.
Question 25
Which technique helps detect bias in AI systems?
A) Increasing model size
B) Disaggregated performance evaluation across demographic groups
C) Using only aggregate performance metrics
D) Reducing model transparency
Answer: B
Explanation:
Disaggregated performance evaluation, which analyzes AI system performance separately across different demographic groups or population subgroups, is a critical technique for detecting bias because overall aggregate metrics can mask significant performance disparities that disproportionately affect certain groups. This evaluation approach reveals whether models work equally well for all groups or whether some groups experience higher error rates or other quality issues.
Disaggregated evaluation requires identifying relevant demographic dimensions for analysis such as race, gender, age, disability status, or other characteristics depending on the application context and potential for disparate impact. Performance metrics including accuracy, false positive rates, false negative rates, precision, recall, and other relevant measures are calculated separately for each subgroup. Significant differences across groups indicate potential bias requiring investigation and mitigation.
Beyond detection, disaggregated evaluation informs bias mitigation by revealing which groups are disadvantaged, what types of errors are occurring, and how severe the disparities are. This information guides decisions about whether additional training data is needed for underrepresented groups, whether model recalibration is required, whether fairness constraints should be applied during training, or whether the system should not be deployed for certain populations. Ongoing monitoring with disaggregated metrics helps detect bias that emerges over time.
Option A is incorrect because model size relates to capacity and does not directly detect or address bias; larger models can exhibit bias just as smaller models can. Option C is wrong because using only aggregate metrics is precisely what fails to detect bias affecting specific groups; disaggregation is necessary. Option D is incorrect because reducing transparency makes bias harder to detect; transparency and detailed analysis support bias detection.
Understanding disaggregated evaluation is essential for detecting and addressing bias in AI systems to ensure fair treatment across all demographic groups.
Question 26
What is the primary concern regarding data privacy in AI systems?
A) That data storage is expensive
B) That AI systems may process personal data in ways that violate privacy rights or regulations
C) That data transfers are slow
D) That data formats are incompatible
Answer: B
Explanation:
Data privacy concerns in AI systems center on the potential for these systems to process personal data in ways that violate individual privacy rights, exceed reasonable expectations, or contravene data protection regulations like GDPR or CCPA. AI systems often require large amounts of data for training and operation, creating significant privacy risks if personal information is collected, used, or shared inappropriately.
Privacy risks in AI include unauthorized data collection or use beyond stated purposes, inadequate consent or notice about how personal data will be used in AI systems, potential for re-identification of anonymized individuals through AI analysis, inference of sensitive attributes not directly provided by individuals, data breaches exposing personal information used for AI training or operation, and retention of personal data longer than necessary. The opacity of some AI systems can make it difficult for individuals to understand or control how their data is being used.
Addressing privacy in AI requires comprehensive approaches including privacy by design principles that embed privacy protections into system architecture, data minimization using only necessary data for specified purposes, purpose limitation ensuring data is used only for declared purposes, obtaining appropriate consent with clear explanations of AI use, implementing strong security controls to protect personal data, providing transparency about data practices, enabling individual rights like access and deletion, and conducting privacy impact assessments to identify and mitigate risks before deployment.
Option A is incorrect because while storage costs are a business consideration, they are not the primary privacy concern which focuses on inappropriate use of personal information. Option C is wrong as transfer speed is a technical performance issue unrelated to privacy rights. Option D is incorrect because format compatibility is a technical integration concern, not a privacy issue about appropriate handling of personal data.
Understanding privacy concerns in AI is critical for governance professionals ensuring AI systems respect individual rights and comply with data protection requirements.
Question 27
Which principle requires that AI systems are designed to minimize potential harm?
A) Efficiency
B) Safety
C) Profitability
D) Complexity
Answer: B
Explanation:
The safety principle requires that AI systems are designed, developed, and deployed to minimize potential harm to individuals, communities, and society, recognizing that AI systems operating in the real world can cause various types of harm if they malfunction, are misused, or produce unintended consequences. Safety encompasses both preventing AI systems from causing harm and ensuring they operate reliably within their intended parameters.
AI safety considerations vary by application context and potential impact. Physical safety is paramount for AI systems controlling vehicles, robots, medical devices, or industrial equipment where failures could cause injury or death. Informational safety addresses harms from incorrect, misleading, or manipulated information in applications like content recommendation, search, or decision support. Social safety concerns harms to dignity, opportunity, or well-being from discriminatory, invasive, or manipulative AI systems.
Implementing AI safety requires multiple approaches including conducting risk assessments to identify potential harms and their likelihood, designing fail-safe mechanisms and fallback procedures for when systems encounter unexpected situations, implementing robust testing including edge cases and adversarial conditions, establishing human oversight for high-stakes decisions, maintaining monitoring to detect safety issues in deployed systems, creating incident response procedures for when problems occur, and regular safety audits to identify emerging risks. Safety considerations must be balanced with system functionality and practicality.
Option A is incorrect because efficiency relates to resource utilization, not harm prevention which is the focus of safety principles. Option C is wrong as profitability is a business objective that may need to be balanced against safety but does not require minimizing harm. Option D is incorrect because complexity is a system characteristic that does not inherently minimize harm; simpler systems may actually be safer in some contexts.
Understanding AI safety principles is fundamental to responsible AI governance that protects against various types of harm from AI systems.
Question 28
What is the purpose of human-in-the-loop (HITL) systems in AI governance?
A) To completely automate all decisions
B) To maintain human oversight and judgment in AI decision-making processes
C) To eliminate the need for AI systems
D) To reduce system accuracy
Answer: B
Explanation:
Human-in-the-loop (HITL) systems maintain meaningful human oversight and judgment in AI decision-making processes, ensuring that humans remain actively involved at critical points rather than simply accepting automated outputs without review. HITL approaches recognize that certain decisions have sufficient stakes, complexity, or potential for error that human judgment should complement AI capabilities rather than being replaced entirely by automation.
HITL implementations vary in the degree and timing of human involvement. Some systems require human approval before AI recommendations are implemented, particularly for high-stakes decisions like medical diagnoses, lending decisions, or hiring recommendations. Others use humans to review a sample of AI decisions for quality assurance and bias detection. Still others involve humans in handling exceptions or edge cases where AI confidence is low or the situation falls outside normal parameters.
The benefits of HITL include maintaining accountability by ensuring humans make final decisions, enabling detection of AI errors or bias through human review, building trust by demonstrating responsible use of AI with appropriate oversight, capturing human expertise and judgment that AI may lack, and satisfying regulatory or ethical requirements for human involvement in certain decisions. However, HITL effectiveness depends on humans receiving adequate information to make informed judgments, having sufficient time and expertise to review AI outputs critically, and being empowered to override AI recommendations when appropriate.
Option A is incorrect because HITL specifically involves humans rather than complete automation; full automation would be the opposite of HITL. Option C is wrong as HITL uses AI systems with human oversight, not eliminating AI. Option D is incorrect because HITL aims to improve decision quality through combined human and AI capabilities, not reduce accuracy.
Understanding HITL approaches is important for designing AI systems that appropriately balance automation benefits with human judgment and oversight.
Question 29
Which factor is most important when determining the appropriate level of explainability for an AI system?
A) The size of the development team
B) The stakes and potential impact of the system’s decisions
C) The age of the technology used
D) The marketing budget for the system
Answer: B
Explanation:
The stakes and potential impact of an AI system’s decisions are the most important factors in determining appropriate explainability requirements because higher-stakes applications with greater potential for significant impact on individuals or society require stronger explainability to support accountability, enable effective oversight, and protect against harm. Risk-based approaches to explainability align governance requirements with the severity of potential consequences.
High-stakes applications such as criminal justice decisions affecting liberty, medical diagnoses affecting health outcomes, credit or lending decisions affecting financial opportunity, employment decisions affecting livelihoods, or autonomous vehicle controls affecting physical safety require robust explainability. Stakeholders need to understand how decisions were reached to assess their appropriateness, identify potential bias or errors, exercise rights to challenge adverse decisions, and hold responsible parties accountable. Regulatory frameworks often mandate explanation requirements for these contexts.
Lower-stakes applications with minimal potential for harm may require less detailed explainability, though basic transparency about AI involvement remains important. Content recommendation systems or game-playing AI may not require the same level of explanation as systems making consequential decisions about individuals. However, even lower-stakes systems should provide some explainability to build appropriate trust and enable users to understand and calibrate their reliance on AI recommendations.
Option A is incorrect because team size is an organizational factor that does not determine the need for explainability which depends on system impact. Option C is wrong as technology age is irrelevant to explainability needs which depend on application stakes. Option D is incorrect because marketing budget is a business consideration unrelated to the ethical and governance requirements for explainability based on potential impact.
Understanding how to match explainability requirements to application stakes is essential for implementing proportionate and effective AI governance.
Question 30
What is the primary purpose of conducting regular AI system audits?
A) To increase system complexity
B) To verify ongoing compliance with governance requirements and detect emerging issues
C) To reduce system transparency
D) To eliminate documentation requirements
Answer: B
Explanation:
Regular AI system audits serve to verify ongoing compliance with governance requirements, ethical principles, and regulatory obligations while detecting emerging issues such as performance degradation, bias drift, security vulnerabilities, or changing risk profiles that require attention. Audits provide systematic, independent assessment of AI systems after deployment, recognizing that initial development and testing cannot guarantee ongoing appropriate operation as conditions change.
AI audits examine multiple dimensions including technical performance such as accuracy and reliability metrics, fairness and bias through disaggregated performance analysis across demographic groups, data quality and appropriateness of training and operational data, security controls and vulnerability management, privacy practices and data protection compliance, documentation completeness and accuracy, human oversight effectiveness, and incident response capabilities. Audits compare actual practices against established policies, standards, and regulations.
Different types of audits serve different purposes. Internal audits conducted by the organization provide regular monitoring and continuous improvement. External audits by independent parties provide objective assessment and can satisfy regulatory or contractual requirements. Technical audits focus on algorithmic performance and fairness while process audits examine governance procedures and controls. The audit frequency and scope should be risk-based, with higher-risk systems receiving more frequent and comprehensive audits.
Option A is incorrect because audits assess systems rather than increase their complexity; complexity is a design characteristic unrelated to audit purposes. Option C is wrong as audits typically increase rather than reduce transparency by examining and reporting on system operation. Option D is incorrect because audits rely on and often strengthen documentation requirements rather than eliminating them; proper documentation supports effective auditing.
Understanding the role of audits in AI governance is essential for maintaining ongoing accountability and ensuring systems continue operating appropriately throughout their lifecycle.