Question #1
A risk practitioner is developing risk scenarios related to successful data poisoning attacks on an AI model used across the organization. Which of the following is the BEST approach to help ensure the scenarios are relevant?
Answer: D
The question focuses on developing relevant risk scenarios for data poisoning attacks against organizational AI models, a core use case within the All AAIR Questions certification's AI risk assessment domain. Relevance in AI risk scenario development requires alignment with the unique organizational context, including specific AI use cases, data pipeline configurations, threat vectors targeting the organization, and associated business impacts. Engaging key stakeholders ensures all context-specific variables are incorporated into scenario development, eliminating generic, non-applicable scenarios that do not reflect the organization's actual risk exposure. This approach directly addresses the core requirement of scenario relevance by grounding each scenario in the organization's operational, business, and threat reality, per AAIR risk assessment best practices.
Option Analysis:
A. Perform adversarial testing in a sandbox environment. Incorrect. Adversarial testing is a control validation activity used to assess an AI model's resilience to known attack scenarios, not a method for developing relevant risk scenarios. It occurs after scenario development and does not contribute to ensuring scenarios align with the organization's unique risk context, so it does not meet the question's requirement.
B. Gather information on similar attacks impacting industry peers. Incorrect. Industry peer attack data is a useful threat intelligence input for scenario development, but it is generic and does not account for the organization's unique AI deployment, data access controls, business priorities, or specific threat actor motivations. Scenarios based solely on peer attacks are often not relevant to the organization's actual risk profile, so this is not the best approach.
C. Create comprehensive data flow diagrams. Incorrect. Data flow diagrams are a supporting tool that maps data movement through the AI pipeline, which can inform scenario development, but they do not capture critical context such as business impact, threat actor targeting, or use case-specific vulnerabilities. On their own, they cannot ensure scenarios are relevant to organizational risk priorities, so this is not the correct answer.
D. Engage key stakeholders in risk scenario development. Correct. Key stakeholders including business unit owners, AI model developers, data engineering teams, and cybersecurity staff hold unique context about the AI model's use cases, data pipeline access controls, business criticality, and known threats targeting the organization. Their participation ensures scenarios are tailored to the organization's specific operating environment, eliminating irrelevant generic scenarios and ensuring alignment with actual risk exposure, which directly addresses the question's requirement for relevant risk scenarios.
Key Concepts:
1. AI Risk Scenario Tailoring: A core AAIR knowledge area that requires risk scenarios to be customized to the organization's unique context rather than relying on generic industry examples, to ensure they accurately reflect actual risk exposure and support actionable risk response.
2. Cross-Functional Stakeholder Engagement in AI Risk Management: AAIR emphasizes that cross-functional stakeholder input is mandatory for complete, relevant AI risk assessments, as no single team holds full visibility into all aspects of AI deployment, business impact, and threat exposure.
3. Data Poisoning Risk Contextualization: AAIR domain knowledge specifies that data poisoning risk scenarios must account for organization-specific variables including training data sourcing workflows, model retraining frequency, access controls to training and inference data, and business impact of corrupted model outputs, all of which require stakeholder input to accurately define.
References:
NIST AI Risk Management Framework, https://www.nist.gov/itl/ai-risk-management-framework
NIST SP 1800-34: Assuring AI Security: Mitigating Data Poisoning Attacks, https://csrc.nist.gov/publications/detail/sp/1800-34/final