IT Cert Ready

ISACA

AAIR

Advanced in AI Risk

Start practicing

About this certification

AAIR (Advanced in AI Risk) is a ISACA certification. IT Cert Ready offers 1,517 practice questions with domain quizzes, full mock exams, and detailed explanations. Practice by official domain with 3 blueprint-aligned domains.

Official exam domain weights

Verified blueprint weights from the official exam outline.

  • Domain 1 · AI Risk Governance and Framework Integration · 37%
  • Domain 2 · AI Life Cycle Risk Management · 21%
  • Domain 3 · AI Risk Program Management · 42%

Sample AAIR practice questions

Preview realistic multiple-choice items from the IT Cert Ready question bank. Expand each answer for the correct choice and explanation—the same format used in domain quizzes and mock exams.

  1. Sample question 1 · AI Risk Program Management (42%)

    Meridian Analytics, a mid-sized U.S.-based technology company with 800 employees, relies on a third-party LLM vendor called NovaMind AI to power its customer-facing contract analysis platform. During a routine monitoring review, the AI risk manager discovers that NovaMind AI experienced an undisclosed model degradation event three weeks ago that caused the LLM to produce systematically incorrect clause interpretations. Meridian's contract with NovaMind AI includes a 72-hour breach notification clause but is silent on AI model performance incidents. Meridian's current incident response plan (IRP) covers data breaches and system outages but does not address third-party AI model failures. Customers have already flagged anomalous outputs, and the legal team is concerned about liability exposure. The AI risk manager must determine the immediate next step to contain impact and strengthen the incident response posture. As the AI risk manager at Meridian Analytics, which action should you recommend FIRST to address the current incident and close the structural gap exposed by the NovaMind AI model degradation event?

    • A. Invoke the existing data breach notification clause in the NovaMind contract and demand a root-cause analysis report within 72 hours to satisfy regulatory disclosure obligations.
    • B. Activate an emergency incident response bridge that classifies the LLM model degradation as an AI-specific incident, initiates customer impact scoping, and simultaneously flags the contractual gap to legal for immediate contract remediation.
    • C. Suspend all use of the NovaMind AI platform indefinitely and revert all contract analysis workflows to manual review until the vendor provides a written assurance of model stability.
    • D. Escalate the matter directly to the board's audit committee and request an emergency session to authorize a vendor replacement procurement process.
    View answer & explanation

    Correct answer: B. Activate an emergency incident response bridge that classifies the LLM model degradation as an AI-specific incident, initiates customer impact scoping, and simultaneously flags the contractual gap to legal for immediate contract remediation.

    [Domain 3 - AI Risk Program Management, Subtopic 3F - AI Incident Response, BIA, Business Continuity, and Disaster Recovery - Area 1 - AI Incident Response, BIA, Business Continuity, and Disaster Recovery] Core Concept: Effective AI incident response requires that organizations classify AI-specific failure modes—such as model degradation, drift, or systematic output errors—as distinct incident categories within the IRP, separate from traditional data breach or outage categories. When a third-party LLM vendor is involved, the response must simultaneously address immediate customer impact containment and the contractual/structural gaps that allowed the incident to go undetected and unreported. Parallel workstreams—operational containment and governance remediation—are the hallmark of mature AI incident response practice. Correct Answer (B): Activating an emergency incident response bridge that formally classifies the LLM degradation as an AI-specific incident is the correct first step because it triggers structured containment, scoping of affected customers, and evidence preservation. Correct Answer (B): Correct Answer (B): Simultaneously flagging the contractual gap to legal ensures that the organization begins remediation of the missing AI performance incident notification clause before the next vendor contract review cycle, addressing both the immediate crisis and the systemic vulnerability in one coordinated action. Why the other options are incorrect: Why the other options are incorrect: Why the other options are incorrect: Option A: : : Invoking the data breach notification clause is procedurally incorrect because model degradation is not a data breach; misapplying that clause could create legal ambiguity and does not address the customer impact scoping or the structural gap in the IRP. Option C: : : Indefinite suspension without a structured impact assessment and business continuity analysis is disproportionate and could cause greater operational harm than a controlled, scoped response; it also bypasses the BIA step required to justify such a decision. Option D: : : Escalating directly to the board's audit committee as the first action skips the operational incident response and containment steps that must precede executive escalation; board notification is appropriate after initial triage and impact scoping are complete.

  2. Sample question 2 · AI Risk Program Management (42%)

    Meridian Pulse, a fast-growing social media startup headquartered in Singapore with approximately 180 employees, recently deployed an AI content moderation tool called ClearStream to handle user-reported posts across its platform. ClearStream uses a third-party large language model API to classify content as safe, borderline, or violating. Three weeks after deployment, a coordinated adversarial attack causes ClearStream to systematically misclassify a wave of coordinated harassment posts as 'safe,' resulting in a 90-minute window during which thousands of targeted users receive unmoderated abusive content. The incident causes a spike in user account deletions and attracts attention from Singapore's Infocomm Media Development Authority (IMDA). Meridian Pulse's Disaster Recovery Plan (DRP) covers database corruption and infrastructure outages but contains no provisions for AI model integrity failures or third-party AI API compromise. The AI risk manager is asked to present a post-incident recommendation to the executive team. As the AI risk manager at Meridian Pulse, which recommendation MOST effectively addresses the DRP gap exposed by this incident?

    • A. Recommend terminating the contract with the third-party LLM API provider and replacing ClearStream with an internally developed model to eliminate third-party dependency risk.
    • B. Recommend updating the DRP to include AI model integrity failure as a named disaster scenario, defining detection triggers, a manual moderation fallback with defined capacity thresholds, third-party AI vendor notification requirements, and a regulatory notification timeline aligned with IMDA expectations.
    • C. Recommend that the DRP team increase the infrastructure recovery SLA with the third-party API provider from 99.9% to 99.99% uptime to reduce the probability of future AI failures.
    • D. Recommend conducting a full red-team adversarial testing exercise on ClearStream before making any DRP changes, so that the scope of DRP updates is informed by a comprehensive threat model.
    View answer & explanation

    Correct answer: B. Recommend updating the DRP to include AI model integrity failure as a named disaster scenario, defining detection triggers, a manual moderation fallback with defined capacity thresholds, third-party AI vendor notification requirements, and a regulatory notification timeline aligned with IMDA expectations.

    [Domain 3 - AI Risk Program Management, Subtopic 3F - AI Incident Response, BIA, Business Continuity, and Disaster Recovery - Area 1 - AI Incident Response, BIA, Business Continuity, and Disaster Recovery] Core Concept: Disaster Recovery Plans must be updated to explicitly incorporate AI-specific failure modes — including model integrity failures, adversarial manipulation of AI outputs, and third-party AI API compromise — as named disaster scenarios with defined response procedures. A DRP that covers only infrastructure outages will leave the organization without a structured response when the failure originates in AI behavior rather than system availability. Post-incident, the AI risk manager's role is to translate lessons learned into concrete DRP enhancements that address detection, containment, recovery, and regulatory notification obligations specific to AI incidents. Correct Answer (B): Updating the DRP to name AI model integrity failure as a disaster scenario and defining the full response lifecycle — detection triggers, fallback controls, vendor notification, and regulatory reporting timelines — directly closes the structural gap the incident exposed. Correct Answer (B): Correct Answer (B): Aligning the regulatory notification timeline with IMDA expectations is particularly important given the regulator's attention to the incident and Singapore's evolving AI governance framework. Why the other options are incorrect: Why the other options are incorrect: Why the other options are incorrect: Option A: : : Terminating the third-party provider and building an internal model is a long-term strategic option that does not address the immediate DRP gap. Internal models carry their own integrity and adversarial risks, and this recommendation does not produce a recovery procedure for future incidents of any kind. Option C: : : Improving API uptime SLAs addresses availability risk, not model integrity or adversarial manipulation risk. The incident was caused by misclassification behavior, not a service outage, so an uptime SLA improvement would not have prevented or shortened the 90-minute harm window. Option D: : : Conducting adversarial testing before updating the DRP inverts the correct priority sequence. The DRP gap is already confirmed by the incident; waiting for a comprehensive threat model before making any DRP changes prolongs the organization's unprotected exposure. Testing should inform future DRP refinements, not delay foundational updates.

  3. Sample question 3 · AI Risk Program Management (42%)

    Halcyon Aerospace Defense (HAD), a large defense prime contractor headquartered in the UK with U.S. operations, operates an AI-driven threat classification system used to triage sensor data for a classified surveillance program. The system is hosted on a hybrid cloud environment, with the AI inference engine managed by a subcontractor, Nexon AI Ltd. HAD's BIA was last updated 18 months ago and does not account for AI model unavailability as a distinct failure mode. During a scheduled tabletop exercise, the incident response team discovers that the recovery time objective (RTO) for the threat classification system is set at 48 hours—identical to the RTO for the underlying IT infrastructure—without any separate consideration for model revalidation time after a failover event. The program's operational tempo requires continuous threat classification with no more than a 4-hour gap. As the AI risk manager at Halcyon Aerospace Defense, which recommendation most directly addresses the gap identified during the tabletop exercise?

    • A. Recommend that the BIA be revised to establish a separate AI-specific RTO for the threat classification system that accounts for model revalidation and integrity verification steps post-failover, aligned to the 4-hour operational requirement.
    • B. Recommend that Nexon AI Ltd. be required to contractually guarantee a 4-hour RTO for the inference engine, transferring the recovery obligation to the subcontractor.
    • C. Recommend that the incident response plan be updated to include a manual override procedure allowing human analysts to perform threat classification without AI assistance during outages.
    • D. Recommend that the hybrid cloud architecture be redesigned to eliminate single points of failure in the AI inference pipeline before the next BIA update cycle.
    View answer & explanation

    Correct answer: A. Recommend that the BIA be revised to establish a separate AI-specific RTO for the threat classification system that accounts for model revalidation and integrity verification steps post-failover, aligned to the 4-hour operational requirement.

    [Domain 3 - AI Risk Program Management, Subtopic 3F - AI Incident Response, BIA, Business Continuity, and Disaster Recovery - Area 1 - AI Incident Response, BIA, Business Continuity, and Disaster Recovery] Core Concept: A Business Impact Analysis for AI-enabled systems must treat AI model availability as a distinct failure mode from underlying IT infrastructure. AI systems require additional recovery steps—such as model integrity verification, revalidation against known-good baselines, and output sanity checks—that extend recovery timelines beyond standard IT failover. Failing to differentiate AI-specific RTOs from infrastructure RTOs creates a false sense of recovery readiness and can result in operational gaps that violate mission requirements. Correct Answer (A): Revising the BIA to establish a separate AI-specific RTO that incorporates model revalidation and integrity verification steps directly addresses the root cause of the gap: the BIA conflates IT infrastructure recovery with AI system recovery. Aligning the revised RTO to the 4-hour operational requirement ensures that recovery planning is grounded in actual mission impact, which is the core purpose of a BIA. Correct Answer (A): Correct Answer (A): This recommendation is within the AI risk manager's direct accountability and produces a durable governance artifact. Why the other options are incorrect: Why the other options are incorrect: Why the other options are incorrect: Option B: : : Contractually transferring the RTO obligation to Nexon AI Ltd. does not resolve the BIA gap; HAD remains operationally accountable for mission continuity, and a contractual SLA does not substitute for an internally validated recovery plan that accounts for model revalidation steps. Option C: : : Recommending a manual override procedure addresses a contingency workaround but does not correct the BIA's failure to capture AI-specific recovery requirements; it also does not ensure the manual process can meet the 4-hour operational tempo at scale. Option D: : : Redesigning the hybrid cloud architecture is a longer-term infrastructure control that does not address the immediate BIA documentation gap or the missing AI-specific RTO; it also exceeds the AI risk manager's direct scope without first establishing the recovery requirements the architecture must support.

  4. Sample question 4 · AI Risk Program Management (42%)

    Crestline Lending Solutions, a fast-growing Canadian fintech with approximately 400 employees, recently launched an AI credit scoring model called ScoreEdge to serve thin-file borrowers. The company has a basic disaster recovery plan (DRP) that covers core banking infrastructure but was written before ScoreEdge was developed. During a tabletop exercise, the team discovers that if ScoreEdge becomes unavailable, loan officers have no documented manual override process and no defined Recovery Time Objective (RTO) or Recovery Point Objective (RPO) for the model. The Office of the Superintendent of Financial Institutions (OSFI) has recently issued guidance requiring federally regulated entities to address AI system continuity in their DRPs. The AI risk manager must determine the most critical gap to remediate first. As the AI risk manager at Crestline Lending Solutions, which remediation action should you prioritize to address the most critical continuity gap revealed by the tabletop exercise?

    • A. Commission a full rewrite of the enterprise DRP to incorporate all technology systems, including ScoreEdge, before the next OSFI supervisory review cycle.
    • B. Define and document an RTO and RPO for ScoreEdge, and establish a tested manual override decision process for loan officers to use during model unavailability, then integrate these into the existing DRP as an AI-specific annex.
    • C. Procure a redundant cloud-based credit scoring vendor as a hot standby so that ScoreEdge downtime does not affect loan origination volumes, eliminating the need for a manual fallback process.
    • D. Conduct additional tabletop exercises quarterly until the team can demonstrate that they can manage ScoreEdge downtime without formal documentation, building institutional knowledge as the primary continuity control.
    View answer & explanation

    Correct answer: B. Define and document an RTO and RPO for ScoreEdge, and establish a tested manual override decision process for loan officers to use during model unavailability, then integrate these into the existing DRP as an AI-specific annex.

    [Domain 3 - AI Risk Program Management, Subtopic 3F - AI Incident Response, BIA, Business Continuity, and Disaster Recovery - Area 1 - AI Incident Response, BIA, Business Continuity, and Disaster Recovery] Core Concept: Incorporating AI systems into disaster recovery planning requires establishing AI-specific recovery objectives (RTO and RPO) that reflect the model's role in critical business processes, as well as documented fallback procedures that allow the organization to continue operating when the AI system is unavailable. NIST AI RMF MANAGE function and OSFI's technology and cyber risk guidance both emphasize that AI continuity controls must be documented, tested, and integrated into formal DRP artifacts — not left to informal institutional knowledge or assumed to be covered by generic IT recovery plans. Correct Answer (B): Defining the RTO and RPO for ScoreEdge and establishing a tested manual override process directly addresses the two most critical gaps identified in the tabletop exercise: the absence of recovery objectives and the absence of a fallback decision process for loan officers. Correct Answer (B): Correct Answer (B): Integrating these into the existing DRP as an AI-specific annex is the most proportionate and timely remediation, aligns with OSFI's AI continuity guidance, and avoids the delay and resource burden of a full DRP rewrite. Why the other options are incorrect: Why the other options are incorrect: Why the other options are incorrect: Option A: : : A full enterprise DRP rewrite is disproportionate to the immediate gap and would take significantly longer to complete, leaving Crestline exposed to the identified continuity risk in the interim. Risk management best practice calls for targeted, risk-proportionate remediation of the highest-priority gap first. Option C: : : Procuring a redundant vendor as a hot standby may be a valid long-term resilience measure, but it does not address the immediate absence of documented recovery objectives or a manual fallback process. Vendor procurement introduces its own due diligence timeline and does not eliminate the need for a human override capability during any transition period. Option D: : : Relying on repeated tabletop exercises to build institutional knowledge as the primary continuity control is not a substitute for documented, auditable DRP procedures. Undocumented knowledge is highly vulnerable to staff turnover, is not verifiable by regulators, and does not satisfy OSFI's requirement for formal AI continuity documentation.

  5. Sample question 5 · AI Risk Program Management (42%)

    The Meridian County Public Health Authority (MCPHA), a mid-sized government agency, relies on a third-party vendor, ClarityAI Solutions, to operate an AI-driven triage and patient-routing system used across six county clinics. The agency's existing BCP was written before the AI system was deployed and does not reference AI-specific failure modes. During a tabletop exercise, the AI risk manager discovers that ClarityAI's contract contains no provisions for incident notification timelines, no requirement to participate in MCPHA's BCP testing, and no defined recovery time objectives (RTOs) for the AI system. A major model degradation event occurs two weeks later, causing misrouted patient referrals for 18 hours before manual override is activated. The agency's leadership requests a post-incident review and asks the AI risk manager to recommend the most critical contractual and BCP remediation action. As the AI risk manager at MCPHA, which remediation action should you prioritize to most directly address the systemic gap exposed by this incident?

    • A. Require ClarityAI Solutions to obtain an independent SOC 2 Type II audit and submit the report to MCPHA's procurement office annually.
    • B. Amend the vendor contract to include AI-specific incident notification timelines, defined RTOs for the AI system, and mandatory participation in MCPHA's BCP testing exercises.
    • C. Develop an internal AI model monitoring dashboard so MCPHA staff can detect model degradation events without relying on vendor reporting.
    • D. Escalate the incident findings to the county board of supervisors and request emergency funding to replace ClarityAI Solutions with an in-house AI solution.
    View answer & explanation

    Correct answer: B. Amend the vendor contract to include AI-specific incident notification timelines, defined RTOs for the AI system, and mandatory participation in MCPHA's BCP testing exercises.

    [Domain 3 - AI Risk Program Management, Subtopic 3F - AI Incident Response, BIA, Business Continuity, and Disaster Recovery - Area 1 - AI Incident Response, BIA, Business Continuity, and Disaster Recovery] Core Concept: When a government agency's BCP and vendor contracts lack AI-specific provisions, the most direct remediation is to embed AI incident response obligations—notification timelines, RTOs, and joint testing requirements—directly into the vendor agreement. This ensures that third-party AI operators are contractually bound to support the agency's continuity posture and that recovery expectations are formally defined and enforceable. Correct Answer (B): The incident revealed that the absence of contractual AI incident notification timelines, RTOs, and BCP testing obligations was the root systemic gap. Correct Answer (B): Correct Answer (B): Amending the contract to include these provisions directly closes the identified control failure, aligns vendor obligations with MCPHA's BCP, and creates enforceable accountability for future AI-related disruptions—addressing both the immediate finding and the structural deficiency. Why the other options are incorrect: Why the other options are incorrect: Why the other options are incorrect: Option A: : : A SOC 2 Type II audit addresses general security and availability controls but does not specifically define AI incident notification timelines, RTOs, or require the vendor to participate in MCPHA's BCP exercises, leaving the core contractual gap unresolved. Option C: : : An internal monitoring dashboard improves detection capability but does not address the contractual absence of vendor incident response obligations or recovery time commitments, which were the primary systemic failures identified. Option D: : : Escalating to the board and seeking to replace the vendor is a disproportionate response that does not address the immediate contractual and BCP gaps; vendor replacement is a long-term strategic option, not the priority remediation for the identified systemic deficiency.

AAIR exam prep FAQ

What is the passing score for the AAIR exam?
Passing scores for AAIR are set by ISACA and can change between exam versions. Check the latest official exam outline on the ISACA website for the current passing standard.
How many questions are on the AAIR certification exam?
Official AAIR exam length and item types are defined by ISACA. Review the current exam blueprint for the latest question counts and timing.
How do domain quizzes help with AAIR exam prep?
Domain quizzes on IT Cert Ready let you practice AAIR one blueprint domain at a time across 3 official exam domains. You get immediate feedback, detailed explanations, and can revisit weak areas before taking a full timed mock exam.
How many AAIR practice questions are on IT Cert Ready?
IT Cert Ready includes 1,517 published AAIR practice questions with domain quizzes, full-length mock exams, and answer explanations mapped to the official outline.