Uniqueness<\/span><\/li>\n<\/ol>\nCorrect Answer: 2<\/b><\/p>\n
Explanation<\/b><\/p>\n
Accuracy refers to whether data correctly represents the real-world facts or values it is intended to describe. Accurate data is important for AI systems because models can learn incorrect patterns when training information contains errors. For example, incorrect customer information, inaccurate labels, or erroneous measurements can negatively affect model performance and downstream decisions. Data quality assessments should consider multiple dimensions because accuracy alone does not guarantee that a dataset is suitable. Data can be accurate but incomplete, outdated, or poorly representative. Organizations should establish appropriate data quality requirements based on the AI system’s purpose and risk. Regular validation can help identify and address data quality problems before and after deployment.<\/span><\/p>\nQuestion 302<\/b><\/p>\n
What does data completeness primarily measure?<\/b><\/p>\n\n- Whether required information is present and sufficiently populated<\/span><\/li>\n
- Whether data is encrypted<\/span><\/li>\n
- Whether an AI model is explainable<\/span><\/li>\n
- Whether users have administrative privileges<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Data completeness concerns whether the required information is present and sufficiently populated for the intended purpose. Missing values can affect AI model training, evaluation, and operational performance. For example, if important attributes are missing disproportionately for a particular population, the resulting system may perform differently across groups. Completeness should therefore be evaluated in the context of the specific use case rather than assuming that a dataset with fewer missing values is always better. Organizations may establish completeness thresholds and investigate missing data patterns. Addressing completeness problems may involve collecting additional information, improving data capture processes, or determining whether missing values can be handled appropriately without introducing unacceptable risk.<\/span><\/p>\nQuestion 303<\/b><\/p>\n
Why is representativeness important when evaluating AI training data?<\/b><\/p>\n\n- It helps determine whether the data adequately reflects the relevant population and operating conditions<\/span><\/li>\n
- It guarantees that an AI system will never be biased<\/span><\/li>\n
- It eliminates the need for performance testing<\/span><\/li>\n
- It ensures that every individual appears exactly once<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Representativeness concerns whether the data used to develop or evaluate an AI system appropriately reflects the populations, situations, and conditions relevant to its intended use. If important groups or operating conditions are poorly represented, an AI system may perform well overall while producing significantly worse results for underrepresented populations. Representativeness should be assessed based on the system’s purpose, deployment environment, and affected stakeholders. It does not guarantee the absence of bias because other factors, such as labeling decisions, historical patterns, and model design, can also influence outcomes. Organizations should combine representative data assessment with subgroup testing and other evaluation methods to understand system performance more thoroughly.<\/span><\/p>\nQuestion 304<\/b><\/p>\n
What is the primary purpose of separating training, validation, and test datasets?<\/b><\/p>\n\n- To evaluate model performance using data that was not used to fit the model<\/span><\/li>\n
- To ensure all data is used for training<\/span><\/li>\n
- To eliminate the need for model evaluation<\/span><\/li>\n
- To prevent any future model changes<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Separating training, validation, and test datasets helps organizations evaluate whether an AI model generalizes beyond the data used to develop it. Training data is generally used to fit the model, while validation data can support model selection or tuning. A separate test dataset provides an additional basis for evaluating performance on previously unseen examples. Without appropriate separation, performance measurements may be overly optimistic because the model may have effectively learned characteristics of the evaluation data. Dataset separation should be designed carefully to avoid leakage between sets. This practice supports more credible performance assessment and helps stakeholders understand whether reported model results are likely to reflect real-world behavior.<\/span><\/p>\nQuestion 305<\/b><\/p>\n
What is data leakage in machine learning?<\/b><\/p>\n\n- When information that should not be available during model training or evaluation improperly influences the model<\/span><\/li>\n
- When data is stored in an encrypted database<\/span><\/li>\n
- When users delete unnecessary records<\/span><\/li>\n
- When a model is deployed to a production environment<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Data leakage occurs when information that should not be available to a model during training or evaluation improperly influences the learning process or performance measurement. Leakage can produce artificially strong results that do not reflect how the system will perform in real-world conditions. For example, a feature may contain information that would only become available after the prediction event, or information from a test set may inadvertently influence training. Leakage can therefore undermine the validity of model evaluation and lead organizations to deploy systems with unrealistic expectations. Preventing leakage requires careful dataset design, feature review, preprocessing procedures, and separation of development and evaluation information.<\/span><\/p>\nQuestion 306<\/b><\/p>\n
What is the main benefit of establishing a model performance baseline?<\/b><\/p>\n\n- It provides a reference point for identifying meaningful changes in system performance<\/span><\/li>\n
- It guarantees future performance will remain constant<\/span><\/li>\n
- It eliminates the need for monitoring<\/span><\/li>\n
- It prevents the model from being updated<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
A performance baseline establishes a reference point against which future AI system performance can be compared. Relevant measures may include accuracy, precision, recall, error rates, response time, or other metrics appropriate to the system’s purpose. Once a baseline has been established, organizations can identify significant changes that may indicate model drift, data changes, system degradation, or other problems. A baseline does not guarantee that performance will remain constant because real-world conditions can change. Instead, it provides a useful comparison point for ongoing monitoring. Baselines should be documented with the conditions under which they were measured so that later comparisons remain meaningful and properly interpreted.<\/span><\/p>\nQuestion 307<\/b><\/p>\n
What is model calibration concerned with?<\/b><\/p>\n\n- Whether predicted probabilities or confidence levels appropriately correspond to observed outcomes<\/span><\/li>\n
- Whether the model has enough storage capacity<\/span><\/li>\n
- Whether users have administrative access<\/span><\/li>\n
- Whether the model’s documentation contains enough pages<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Model calibration concerns whether the confidence or probability values produced by a model appropriately correspond to actual observed outcomes. For example, if a system frequently assigns a probability of 80 percent to events, a well-calibrated system should experience those events at approximately that frequency under comparable conditions. Calibration can be important when users rely on model confidence to prioritize actions or make decisions. A model can have good overall predictive accuracy while still being poorly calibrated. Organizations should therefore select evaluation methods appropriate to the system’s purpose and communicate limitations clearly. Calibration monitoring may also be useful after deployment because changes in data or operating conditions can affect model behavior.<\/span><\/p>\nQuestion 308<\/b><\/p>\n
Why should AI systems communicate uncertainty when it is relevant to decisions?<\/b><\/p>\n\n- It helps users understand the limitations and confidence associated with AI outputs<\/span><\/li>\n
- It guarantees every prediction is correct<\/span><\/li>\n
- It removes the need for human judgment<\/span><\/li>\n
- It prevents users from seeing model outputs<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Communicating uncertainty can help users interpret AI outputs appropriately instead of treating predictions as certain facts. Some AI systems produce estimates or classifications that inherently involve uncertainty. Providing meaningful confidence information, limitations, or other contextual indicators can help users determine when additional review is appropriate. The presentation should be understandable to the intended users and should not create a false impression of precision. Uncertainty communication is particularly important when AI outputs contribute to consequential decisions. It does not guarantee correctness, and confidence scores themselves may not always be perfectly calibrated. Organizations should therefore evaluate how uncertainty information is generated and ensure users receive appropriate guidance on interpreting it.<\/span><\/p>\nQuestion 309<\/b><\/p>\n
What is a major concern when an AI system produces highly confident but incorrect outputs?<\/b><\/p>\n\n- Users may accept the incorrect output because the confidence presentation creates excessive trust<\/span><\/li>\n
- The system will automatically correct itself<\/span><\/li>\n
- The output becomes legally valid because it is confident<\/span><\/li>\n
- Human oversight becomes unnecessary<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Highly confident but incorrect AI outputs can create significant risks because users may interpret confidence as evidence that the result is reliable. This can contribute to automation bias and reduce the likelihood that users will independently verify important information. The risk is particularly significant when AI outputs influence consequential decisions or actions. Organizations should evaluate whether confidence measures are meaningful and appropriately calibrated, provide users with guidance about system limitations, and establish review procedures for higher-risk situations. User interface design can also influence how people interpret AI outputs. Confidence should therefore not be presented in a way that encourages unwarranted certainty or discourages appropriate human judgment.<\/span><\/p>\nQuestion 310<\/b><\/p>\n
What is the purpose of a rollback procedure for an AI model deployment?<\/b><\/p>\n\n- To restore a previous known-good version when a new deployment causes unacceptable problems<\/span><\/li>\n
- To permanently delete all model versions<\/span><\/li>\n
- To prevent incident investigation<\/span><\/li>\n
- To allow unauthorized users to modify production systems<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
A rollback procedure allows an organization to return an AI system to a previous known-good version when a new release causes unacceptable performance, security, reliability, or other operational problems. Effective rollback procedures should identify which versions are available, how they can be restored, who has authority to initiate the rollback, and how the process will be monitored. Rollback is particularly useful when rapid remediation is required and a new deployment is causing harm or significant disruption. Organizations should test rollback procedures rather than assuming they will work during an emergency. A successful rollback can reduce the duration and impact of an incident while allowing teams to investigate the underlying cause.<\/span><\/p>\nQuestion 311<\/b><\/p>\n
What is the primary purpose of a canary deployment for an AI system?<\/b><\/p>\n\n- To expose a new version to a limited portion of users or traffic before broader deployment<\/span><\/li>\n
- To permanently prevent model updates<\/span><\/li>\n
- To eliminate all production monitoring<\/span><\/li>\n
- To give every user unrestricted access to an untested model<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
A canary deployment introduces a new AI system version to a limited portion of traffic, users, or operational environments before expanding deployment more broadly. This approach allows organizations to observe real-world behavior and identify unexpected performance, reliability, security, or other problems while limiting exposure. Monitoring during the canary phase should use predefined criteria so that decision-makers can determine whether the release should proceed, pause, or be rolled back. Canary deployments do not replace comprehensive pre-production testing, but they provide an additional layer of controlled validation. They can be especially useful for significant changes where real-world conditions may reveal issues that were not visible during development testing.<\/span><\/p>\nQuestion 312<\/b><\/p>\n
Why is a secure model registry useful in AI governance?<\/b><\/p>\n\n- It can provide controlled storage and traceability for approved model versions and related metadata<\/span><\/li>\n
- It allows anyone to replace production models<\/span><\/li>\n
- It eliminates the need for access controls<\/span><\/li>\n
- It prevents organizations from documenting model changes<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
A secure model registry can help organizations manage AI model versions, metadata, approval status, deployment information, and related artifacts in a controlled environment. Proper access controls can restrict who is allowed to upload, approve, modify, or deploy models. Version tracking also supports traceability by helping teams determine which model version was approved and which version is operating in a particular environment. Depending on the organization’s processes, a registry may also support testing status, ownership, risk classification, and release history. A registry is not a substitute for broader AI governance, but it can provide an important technical control for maintaining model integrity, accountability, and lifecycle management.<\/span><\/p>\nQuestion 313<\/b><\/p>\n
What is a key security concern when AI systems use external APIs?<\/b><\/p>\n\n- Unauthorized access, misuse of credentials, or exposure of sensitive information<\/span><\/li>\n
- The API will automatically improve model accuracy<\/span><\/li>\n
- External APIs eliminate all security risks<\/span><\/li>\n
- API use makes authentication unnecessary<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
External APIs can introduce security risks because AI systems may exchange information with third-party services and depend on externally managed infrastructure. Organizations should consider authentication, authorization, credential protection, data transmission, logging, rate limits, and the handling of sensitive information. API credentials should not be exposed in source code or shared broadly. Organizations should also evaluate what data is sent to the external provider and whether the provider’s handling practices are appropriate for the intended use. Security controls should be proportionate to the sensitivity of the information and potential impact of misuse. Proper API governance helps reduce unauthorized access and prevents avoidable exposure of organizational or personal information.<\/span><\/p>\nQuestion 314<\/b><\/p>\n
Why is secrets management important for AI applications?<\/b><\/p>\n\n- It protects credentials, API keys, tokens, and other sensitive authentication information<\/span><\/li>\n
- It improves the model’s mathematical accuracy<\/span><\/li>\n
- It eliminates the need for user authentication<\/span><\/li>\n
- It makes all AI outputs automatically trustworthy<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
AI applications may rely on API keys, access tokens, database credentials, service accounts, and other secrets to communicate with systems or services. Poor secrets management can allow unauthorized individuals or applications to access sensitive resources. Organizations should therefore use appropriate mechanisms for storing, rotating, limiting, and monitoring secrets rather than embedding them directly in application code or sharing them unnecessarily. Access should follow least-privilege principles, and compromised credentials should be capable of being revoked or replaced. Secrets management is primarily a security control rather than a model-performance control. Protecting credentials helps reduce the likelihood that AI applications become an entry point for unauthorized access or data exposure.<\/span><\/p>\nQuestion 315<\/b><\/p>\n
What is a major governance concern when an organization uses an open-source AI model?<\/b><\/p>\n\n- Understanding its provenance, license, security, limitations, and suitability for the intended use<\/span><\/li>\n
- Assuming that open-source models require no evaluation<\/span><\/li>\n
- Assuming that publicly available models are automatically risk-free<\/span><\/li>\n
- Eliminating all documentation requirements<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Open-source availability does not automatically establish that an AI model is appropriate, secure, accurate, or compliant for a particular use. Organizations should consider the model’s provenance, licensing terms, development history, documentation, known limitations, security characteristics, dependencies, and suitability for the intended application. They should also evaluate whether the model can be maintained and monitored effectively. Depending on the use case, additional testing may be needed to assess performance, fairness, robustness, privacy, and security. Governance should therefore apply appropriate due diligence to open-source models just as it would to other AI components. Reusing publicly available software or model weights without assessment can introduce significant technical and governance risks.<\/span><\/p>\nQuestion 316<\/b><\/p>\n
Why is AI model license compliance important?<\/b><\/p>\n\n- It helps ensure that the organization uses model components according to applicable license terms<\/span><\/li>\n
- It guarantees the model is unbiased<\/span><\/li>\n
- It eliminates cybersecurity risks<\/span><\/li>\n
- It removes the need to track model provenance<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
AI models and related components may be distributed under licenses that impose specific conditions on use, modification, distribution, attribution, or other activities. Organizations should understand and document applicable license requirements before incorporating such components into products or internal systems. License compliance is separate from model accuracy, security, and fairness, so satisfying a license does not automatically mean the model is appropriate for the intended use. Organizations may need legal, procurement, security, or governance review depending on the circumstances. Maintaining provenance information can also help teams understand where model components came from and what obligations may apply. Proper license management reduces the risk of unintended legal or operational problems.<\/span><\/p>\nQuestion 317<\/b><\/p>\n
What is a major risk associated with exposing AI model weights?<\/b><\/p>\n\n- Unauthorized parties may copy, modify, analyze, or misuse the model<\/span><\/li>\n
- The model automatically becomes more accurate<\/span><\/li>\n
- Security controls become unnecessary<\/span><\/li>\n
- The model can no longer be evaluated<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Model weights can represent important intellectual property and may also provide information that could help unauthorized parties reproduce or analyze an AI system. Depending on the model and deployment environment, exposed weights may facilitate unauthorized copying, modification, or misuse. Organizations should therefore consider appropriate access controls, storage protections, deployment architecture, and monitoring. The required level of protection depends on factors such as the sensitivity of the model, business value, threat environment, and potential consequences of misuse. Protecting model weights does not replace application security or other governance controls. Instead, it forms part of a broader approach to protecting AI assets and maintaining control over authorized model use.<\/span><\/p>\nQuestion 318<\/b><\/p>\n
Why do AI agents require careful permission management?<\/b><\/p>\n\n- Agents may be capable of taking actions through connected tools or systems<\/span><\/li>\n
- Agents can never affect external systems<\/span><\/li>\n
- Permissions are irrelevant when AI is used<\/span><\/li>\n
- Agents automatically follow organizational policies perfectly<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
AI agents may interact with external tools, databases, applications, or services and can sometimes perform actions rather than simply generate information. This increases the importance of permission management because an agent with excessive privileges could make unauthorized changes, access sensitive information, or perform unintended transactions. Organizations should apply least-privilege principles and limit agent access to only the resources necessary for the approved task. Additional safeguards may include action confirmation, transaction limits, monitoring, and the ability to disable the agent quickly. Agentic capabilities should therefore be governed according to the potential consequences of the actions the system can perform, not simply according to the sophistication of its language or reasoning capabilities.<\/span><\/p>\nQuestion 319<\/b><\/p>\n
What is the purpose of action confirmation for a high-impact AI agent?<\/b><\/p>\n\n- To require appropriate human approval before certain consequential actions are executed<\/span><\/li>\n
- To guarantee that the agent will never make an error<\/span><\/li>\n
- To give the agent unlimited authority<\/span><\/li>\n
- To prevent all automated processing regardless of risk<\/span><\/li>\n<\/ol>\n
Correct Answer: 1<\/b><\/p>\n
Explanation<\/b><\/p>\n
Action confirmation can provide an important safeguard when an AI agent is capable of performing consequential actions. Instead of allowing the agent to execute every action automatically, the system can require a qualified human to confirm actions that exceed defined risk or transaction thresholds. This can reduce the likelihood that an erroneous or manipulated AI output results in an irreversible or harmful action. Confirmation requirements should be designed according to the potential impact of the action and should provide the reviewer with enough information to make an informed decision. Human confirmation does not guarantee that every action will be correct, but it creates an additional control point for higher-risk activities.<\/span><\/p>\nQuestion 320<\/b><\/p>\n
Why should organizations establish limits on autonomous AI actions?<\/b><\/p>\n\n