QUIZ: Risk Management for AI Systems

Assessment settings

• Passing grade required: Yes
• Passing grade: 80%

Pass message

Excellent Work!
Your answers demonstrate a clear command of risk identification, categorization, and mitigation strategies for AI systems. You’re well-prepared to implement robust controls and governance protocols that foster trustworthy innovation in AI.

Fail message

Thank You for Completing the Quiz.
Your results suggest revisiting the core concepts around AI risk domains, proactive assessments, and resilience-building practices. Review the lesson’s key frameworks—like D.A.R.E. and FMEA—and apply them to concrete examples before retaking the quiz to deepen your understanding of responsible AI risk management.

Questions and answer key

1. Which set of risk categories is most relevant when assessing AI initiatives?

Type: single

• Operational, public relations, hardware design, philanthropic
• Customer satisfaction, philanthropic investment, financial ROI, shipping logistics
• Staff productivity, public image, quarterly profit, internal communication
• Operational, ethical, security, and compliance risks requiring careful review

Explanation

Correct Answer: D
Explanation: AI involves vulnerabilities beyond simple business metrics—teams must address operational failures, ethical harms, cyber threats, and regulatory obligations throughout the AI lifecycle.

2. During the design phase of an AI system, which approach best promotes proactive risk identification?

Type: single

• Applying structured assessment tools like FMEA or HACCP to detect potential failure modes early
• Focusing on speed of deployment, postponing any risk assessment until post-launch analysis
• Relying on ad hoc judgments from a single engineer, with no cross-functional collaboration
• Assuming that pre-trained models from open-source repositories are inherently safe for production

Explanation

Correct Answer: A
Explanation: FMEA and HACCP help teams methodically uncover weaknesses, enabling early fixes before large-scale deployment. Collaborative input strengthens the overall risk review process.

3. What does “resilience” mean when discussing AI risk management?

Type: single

• Restricting updates to the system after its first version achieves minimal performance benchmarks
• Designing AI to disregard user feedback, ensuring model decisions remain consistent at all costs
• Building robust fail-safes and monitoring protocols so the system recovers gracefully from unexpected setbacks
• Relying on a rigid configuration that never changes, thus preventing new complications from emerging

Explanation

Correct Answer: C
Explanation: Resilience means anticipating and handling disruptions or failures. Redundant systems, active monitoring, and quick response strategies keep AI functional under stress.

4. In the D.A.R.E. model (Data, Algorithm, Regulation, Ethics), which dimension focuses on abiding by existing laws and policies?

Type: single

• Algorithmic oversight, ensuring interpretability of complex neural networks
• Regulatory compliance, addressing mandates like GDPR and sector-specific guidelines
• Ethical consideration of social consequences and moral obligations
• Data protection, assuring proper anonymization and secure storage techniques

Explanation

Correct Answer: B
Explanation: The “Regulation” category of D.A.R.E. covers adherence to local and international legal standards, demanding alignment with evolving policies to avoid fines and reputational damage.

5. How might an organization best address adversarial threats targeting its AI models?

Type: single

• Only employing standard antivirus software and disallowing any updates to training data post-launch
• Publicly sharing detailed code logic, hoping collective transparency negates adversarial exploitation
• Forbidding encryption practices to keep the AI’s internal structure freely available to all stakeholders
• Implementing anomaly detection and strengthening defensive measures that guard against manipulated inputs

Explanation

Correct Answer: D
Explanation: Adversarial attacks often involve subtle data manipulations. Proactive monitoring coupled with secure design resists such exploits and preserves correct AI functioning.

6. Which of the following best illustrates proactive compliance risk management in AI?

Type: single

• Tracking relevant legal developments, conducting regular audits, and adjusting system parameters to uphold standards
• Refusing to adapt the algorithm post-deployment, trusting that initial regulatory analysis will remain sufficient
• Over-collecting user data in all regions, relying on after-the-fact negotiations to resolve conflicts
• Delegating all accountability to external contractors, avoiding internal oversight or regulatory checks

Explanation

Correct Answer: A
Explanation: Compliance demands ongoing monitoring of laws, routine internal reviews, and timely system modifications to ensure alignment with current legal frameworks.

7. Which ethical risk can arise if an AI model is trained on skewed historical data?

Type: single

• The AI becomes fully explainable and fosters broader trust among diverse communities
• The system’s real-time outputs automatically incorporate all perspectives in a balanced manner
• Unfair predictions or decisions that amplify societal biases, negatively affecting marginalized groups
• Comprehensive elimination of any accountability concerns for the organization deploying the AI

Explanation

Correct Answer: C
Explanation: Biased datasets can perpetuate unfair treatment—reinforcing inequalities or stereotypes. Proper vetting and diverse training data are crucial to mitigate harm.

8. To enhance resilience in a critical AI application, which practice is most beneficial?

Type: single

• Repeating the same data-collection methods without iteration, trusting historical records suffice for all new contexts
• Using layered fail-safes, continuous user feedback loops, and real-time anomaly detection to avert major breakdowns
• Relying on a single vendor’s off-the-shelf AI solution and prohibiting modifications to the underlying model
• Prohibiting any direct stakeholder engagement once the initial technical design has been approved

Explanation

Correct Answer: B
Explanation: Fail-safes, robust monitoring, and inclusive feedback maintain consistent operations despite system disruptions, supporting the principle of ongoing resilience.

9. Which step should an AI governance committee include in its standard protocols for effective risk management?

Type: single

• Ignoring cross-departmental views, letting only data scientists establish guidelines for the entire organization
• Producing minimal documentation of AI decisions, retaining no audit logs for model changes over time
• Mapping out risk assessment roles, reviewing compliance post-launch, and engaging ethicists in major decisions
• Restricting public knowledge about governance standards to maintain control and minimize external scrutiny

Explanation

Correct Answer: C
Explanation: Effective governance involves clearly defining responsibilities, ongoing compliance checks, and multidisciplinary engagement (including ethics experts) to address AI’s complex implications.

10. In the context of risk categorization for AI, why is a structured framework crucial?

Type: single

• It trivializes diverse problems by merging them into one universal solution, thus simplifying development
• It ensures technology remains static, preventing new features that could introduce added risk
• It compels all stakeholders to ignore minor issues, concentrating only on the most catastrophic scenarios
• It aids in systematically identifying, prioritizing, and assigning targeted mitigation strategies for various threat domains

Explanation

Correct Answer: D
Explanation: A formal framework organizes risk domains to guide appropriate response efforts. By clarifying and ranking threats, teams can handle them more effectively and stay aligned with best practices.