Responsible AI Scenario Drill
Level Up Team
This post summarizes Responsible AI Scenario Drill, module 4.3 of Level Up’s Azure AI Fundamentals series. The lesson is on YouTube as Azure AI Fundamentals AI-901 Exam: 4.3: Responsible AI Scenario Drill. Module 4.2 is Responsible AI in a Generative Application.
Choose the primary principle, then the control that matches it
For each scenario, name the principle that applies most directly and the mitigation that addresses the stated failure. Several principles can apply. The exam wants the primary one. The six principles are fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.
Ten scenarios
| Scenario | Primary principle | Mitigation |
|---|---|---|
| A loan system approves qualified applicants from one demographic group less often than similarly qualified applicants from another | Fairness | Evaluate approval and error rates across groups, then address unjustified differences |
| A medical assistant recommends urgent treatment with no evidence, confidence, or explanation | Transparency | Provide supporting evidence, communicate uncertainty, and explain the limits |
| A support assistant includes another customer’s account information in a response | Privacy and security | Enforce user-level authorization and isolate retrieved data between sessions |
| A public-service assistant accepts and returns only spoken language | Inclusiveness | Add text input and output, captions, and assistive-technology support |
| An agent treats “reduce cloud costs” as permission to delete production resources | Reliability and safety | Limit permissions and require human approval before destructive actions |
| A customer cannot appeal an AI refund denial, and no one owns the decision | Accountability | Assign a human owner and provide review, appeal, and escalation |
| Customers believe they are chatting with a person because AI use is never disclosed | Transparency | Disclose AI involvement before users rely on the responses |
| A voice assistant frequently misinterprets particular regional accents | Fairness | Test recognition by accent, improve representative data, and offer another input method |
| The order database is down, and the assistant still says an order has shipped | Reliability and safety | Detect the failure, avoid unsupported claims, and degrade to a status message or a person |
| Complete conversations, including payment and health information, are stored after the service no longer needs them | Privacy and security | Minimize collected data and enforce retention and deletion |
How to read the question
| The question focuses on | Primary principle |
|---|---|
| Unequal treatment or performance | Fairness |
| Unsafe or unreliable behavior | Reliability and safety |
| Data exposure or unauthorized access | Privacy and security |
| Barriers to participation | Inclusiveness |
| Missing disclosure or explanation | Transparency |
| Missing ownership, review, or recourse | Accountability |
Then choose the mitigation that directly removes or controls that failure.
Bring this to your team
Level Up can run a personalized Azure AI Fundamentals workshop for your team. Tell us a little about your group and we’ll reply with pricing.