AAIR Practice Questions

Model poisoning attacks target the training data or model parameters to degrade performance or

introduce malicious behavior. Supply chain tampering introduces compromised components at

vendor or integration stages. Security by design principles require embedding defenses against these

threats from the earliest design stages.

Why C is Correct: According to ISACA AAIR security by design guidance, adversarial resilience and

data integrity controls address both model poisoning and supply chain tampering at their root.

Adversarial resilience training prepares the model to resist maliciously crafted inputs. Data integrity

controls—cryptographic signing, provenance tracking, integrity verification—detect tampering in

training data and model artifacts across the supply chain. Together, these form the most

comprehensive defense against both attack categories.

Why A is Wrong: Data refreshes with checksums detect post-hoc data corruption but do not build

adversarial resilience into the model itself. Checksums verify file integrity but cannot prevent

poisoning attacks that maintain file integrity while altering data content.

Why B is Wrong: Frequent retraining and bias monitoring address performance drift and fairness but

do not specifically protect against deliberate tampering. A retrained model may still be trained on

poisoned data if integrity controls are absent.

Why D is Wrong: Data tokenization protects sensitive field values from unauthorized access (a privacy

control) but does not address model poisoning or supply chain tampering, which can occur without

accessing or exposing the sensitive field values themselves.

AI governance frameworks that are disconnected from organizational culture and risk tolerance face

adoption resistance and produce policies that are either too restrictive or too permissive. Tailored

governance is more likely to be embraced by stakeholders and produce risk policies calibrated to the

organization's actual risk appetite.

Why B is Correct: The ISACA AAIR Study Guide emphasizes that governance tailored to culture and

risk tolerance produces two primary benefits: stakeholders are more likely to accept and follow

governance policies that reflect their own values and operational realities, and the resulting policies

are appropriately calibrated to actual risk appetite rather than generic standards. Together, these

produce more effective, sustainable governance.

Why A is Wrong: Model explainability is a technical property of individual AI systems, not a

governance tailoring outcome. Regulatory compliance may improve with tailored governance but is a

compliance benefit, not the primary benefit of cultural alignment.

Why C is Wrong: Automation of risk assessment and accountability clarity are process improvements

that may result from better governance design but are not the primary benefit of cultural and risk

tolerance alignment.

Why D is Wrong: Training programs and reskilling are workforce development activities. While

governance reform may highlight training needs, skills development is an enabling activity rather

than the primary benefit of culturally tailored governance.

API integration with external AI services creates data transmission pathways between the

organization and external systems. Customer support contexts involve sensitive personal data—

account information, contact details, inquiry content—that may be transmitted through these API

connections.

Why B is Correct: The ISACA AAIR security and privacy guidance identifies unauthorized disclosure of

sensitive data through insecure API connections as the greatest risk in multi-service AI integration.

APIs can be vulnerable to interception, inadequate authentication, or misconfiguration. In a

customer support context, exposure of personal data via API vulnerabilities creates privacy

violations, regulatory liability, and reputational harm—all more severe than the other listed

concerns.

Why A is Wrong: Bias and inaccuracy in chatbot responses are real quality risks but represent service

quality issues rather than security or privacy breaches. Inaccurate responses are visible and

correctable; data breaches may go undetected.

Why C is Wrong: Customer dissatisfaction from operational delays is a service quality and business

risk. It is a manageable consequence of performance issues rather than the greatest risk from API-

based AI integration.

Why D is Wrong: Insufficient training datasets affect model quality but are a development concern

addressed during the model selection phase. They do not represent the primary operational risk of

deploying multi-service API integrations in production.

Predictive maintenance for equipment components requires continuous analysis of operational

data—vibration, temperature, pressure, electrical signatures—that indicate component health over

time. AI systems performing this function must process high-frequency sensor data to detect

patterns that precede failure.

Why D is Correct: According to ISACA AAIR AI application guidance, real-time sensor monitoring data

analysis is the core capability enabling accurate failure point prediction. By continuously analyzing

sensor readings against learned patterns of pre-failure behavior, AI systems can detect early-stage

degradation signals and forecast time-to-failure with precision unavailable through periodic

inspection or rule-based thresholds.

Why A is Wrong: Root cause identification occurs after a defect has already manifested. For

predictive maintenance—predicting failure before it occurs—post-defect analysis provides no

forward-looking capability.

Why B is Wrong: Replacement product recommendation is a procurement and inventory support

function. It assists in planning responses to predicted failures but is not the capability that enables

the prediction itself.

Why C is Wrong: Dynamic inventory management of spare parts supports maintenance operations

but is a supply chain function dependent on failure predictions, not a capability that generates those

predictions.

AI systems make decisions that can affect individuals, organizations, and society. When no individual

or function is clearly accountable for those decisions, the organization cannot demonstrate due

diligence, remedy harms, or mount a coherent legal defense when challenged.

Why D is Correct: The ISACA AAIR framework identifies legal liability as the greatest organizational

risk from absent accountability mechanisms. When AI outputs cause harm—discriminatory lending

decisions, unsafe autonomous vehicle actions, inaccurate medical diagnoses—the absence of

documented accountability makes it impossible to demonstrate responsible governance to courts,

regulators, and affected parties. This creates maximum legal exposure across contract, tort, and

regulatory law.

Why A is Wrong: Intellectual property exposure is a significant risk in AI contexts (particularly around

training data and model weights) but is not primarily caused by absent accountability mechanisms.

IP risk arises from access controls and contractual protections.

Why B is Wrong: Ineffective model training is a technical quality issue. While accountability for

model development may influence training quality, ineffective training is not the primary risk from

absent accountability for outputs and decisions.

Why C is Wrong: Reduced availability is an operational resilience concern. Accountability gaps do not

directly cause availability failures, which are driven by architectural and operational factors.

Third-party AI service integration introduces distributed accountability challenges. When external

services are incorporated into organizational supply chains, clearly designated ownership of AI

governance and risk is essential to ensure responsibilities do not fall into gaps between internal and

external parties.

Why A is Correct: According to ISACA AAIR third-party risk principles, designating AI governance and

risk ownership is the foundational requirement for supply chain integration. Without clear internal

ownership, third-party AI risks cannot be effectively monitored, escalated, or managed. The

designated owner ensures vendor performance is assessed against risk criteria, contractual

obligations are enforced, and emerging risks are addressed.

Why B is Wrong: Standardizing data transfer protocols is a technical integration concern. While

important for security and operations, it does not address who is responsible for managing the

ongoing governance and risk of the integrated service.

Why C is Wrong: Training third-party staff on internal AI policies is a vendor management activity

that may be contractually required but does not substitute for internal governance ownership. Third-

party training does not remove the organization's responsibility for oversight.

Why D is Wrong: Security policy alignment is important for consistent security posture but represents

one dimension of third-party risk management. Governance and risk ownership encompasses

security policy alignment as well as broader risk management responsibilities.

Effective AI user training must go beyond policy acknowledgment and compliance instruction to

equip employees with the practical skills needed to identify ethical risks and report them

appropriately. This builds an active risk-aware workforce.

Why B is Correct: The ISACA AAIR framework identifies that training on ethical risk indicators and

reporting mechanisms directly reinforces ethical norms by enabling employees to recognize real-

world signs of AI misuse, bias, or harmful outputs. When staff can identify specific risk signals and

know how to escalate them, the organization builds a proactive risk culture grounded in practical

ethical literacy.

Why A is Wrong: Acceptable use policy acknowledgment is a compliance activity, not a culture-

building measure. Acknowledging a document does not ensure employees understand how to apply

ethical principles in practice.

Why C is Wrong: Cyber threat identification addresses security risk, which is narrower than the full

scope of ethical AI risk. Security training does not develop ethical judgment regarding fairness, bias,

or societal impact.

Why D is Wrong: Regulatory compliance checklists address legal obligations but do not develop the

ethical reasoning and risk recognition skills needed to reinforce organizational norms.

Generative AI capabilities and the associated risk landscape evolve rapidly. Governance policies and

controls must be refreshed through a structured, regular process rather than reactively or only when

compliance requirements change.

Why A is Correct: According to ISACA AAIR, establishing a regular review cadence with codified

reassessment procedures is the most robust approach because it creates a systematic, predictable

process for keeping governance current. By documenting when and how policies will be reviewed—

including triggers for ad hoc review (new deployments, incidents, regulatory changes)—the

organization ensures governance never stagnates regardless of external pressures.

Why B is Wrong: Regulatory alignment is an important input to governance refresh but represents a

reactive, external-trigger approach. Relying primarily on regulatory signals means governance lags

behind organizational AI changes not covered by new regulations.

Why C is Wrong: Centralizing authority in executive and technical leadership creates decision

bottlenecks and reduces the operational agility needed to keep pace with rapidly evolving AI

deployments. Distributed governance with clear escalation paths is more effective.

Why D is Wrong: Annual reviews are too infrequent for generative AI tools, which may see significant

capability changes and risk profile shifts multiple times per year. Annual compliance audits cannot

keep governance current in a rapidly evolving AI environment.

Privacy and data protection regulations worldwide—including GDPR, CCPA, and sector-specific

laws—impose strict requirements on the collection, use, and processing of personal information.

Customer data used for AI systems must be obtained through lawful means with appropriate consent

for the specific processing purpose.

Why A is Correct: According to ISACA AAIR guidance on regulatory compliance, the legal basis for

processing personal data is the foundational requirement. An AI system built on data collected

without proper consent or legal authorization exposes the organization to regulatory penalties,

reputational damage, and forced shutdown of the system. Consent must be specific to the AI use

case, not merely generic data collection consent.

Why B is Wrong: Backup and storage protocols address data security and resilience, which are

compliance requirements but secondary to the lawfulness of data collection. Securely storing

improperly obtained data does not cure the regulatory violation.

Why C is Wrong: Human review of recommendations is a governance safeguard for accuracy and

fairness, not a regulatory compliance requirement for data collection. Many regulations do not

require human review of recommendation systems.

Why D is Wrong: Supervised learning is a modeling technique that does not address regulatory

compliance regarding data sourcing. The training methodology is irrelevant to whether the

underlying data was legally obtained.

Accountability in AI governance requires that specific individuals or roles be clearly designated as

responsible for AI system outputs, decisions, and associated risks. Without formal documentation of

ownership, accountability gaps emerge.

Why D is Correct: The ISACA AAIR framework emphasizes that accountability must be explicit and

documented, with named individuals assigned to own AI outcomes. Formal role assignments create

a traceable chain of responsibility that supports auditability, regulatory compliance, and effective

escalation when issues arise. Named ownership prevents diffusion of responsibility.

Why A is Wrong: A centralized task force creates collective responsibility, which can dilute individual

accountability. Governance bodies support oversight but do not replace individual role ownership for

specific outputs.

Why B is Wrong: Continuous monitoring and KPIs are valuable operational controls but represent

monitoring mechanisms, not accountability structures. Monitoring detects issues but does not assign

responsibility for them.

Why C is Wrong: Resource allocation reviews address investment efficiency rather than

accountability for AI decisions and outputs. This is a management activity, not an accountability

framework.