Question 1 A payroll agent will fetch employee salary records. Each employee must see only their own data. Which integration identity design should lead the solution? Use one owner account for all requests Grant the agent broad application permissions Use delegated user identity and enforce payroll service authorization on each request Put employee IDs in hidden prompts Explanation Correct Answer C:Delegation preserves the caller context so the service can reject another employee's record. Incorrect Answers A:An owner account could expose payroll records without distinguishing callers. B:Application-wide permissions exceed the per-employee access requirement. D:A user-supplied or hidden ID is not server-side authorization.
Question 2 A public customer service agent answers shipping FAQs without accessing private orders. What audience design is appropriate? Give guests shared customer credentials Offer anonymous access only to public content with restricted tools Enable all backend connectors for anonymous visitors Require administrator credentials from every visitor Explanation Correct Answer B:Public answers can be available without providing access to order systems. Incorrect Answers A:Shared identities prevent isolation and accountability. C:Unrestricted actions would make private business data accessible to strangers. D:Administrator credentials would expose elevated privileges unnecessarily.
Question 3 An insurance agent uses both Dataverse claims and an external underwriting API. What should the architect capture first? Map data sources, permissions, action contracts and data flows before selecting tools Hardcode an example underwriting response Start by publishing to every channel Disable audit logging for easier prototyping Explanation Correct Answer A:Integration planning identifies trust boundaries and dependencies for the agent. Incorrect Answers B:A fixed response cannot execute the required external assessment. C:Channel publication cannot resolve missing integration permissions. D:Removing traceability undermines a regulated integration.
Question 4 A support agent will launch in Teams for employees and on an external website for customers. What is essential in the channel plan? Assume Teams SSO works identically on all websites Publish one unrestricted agent link everywhere Specify per-channel authentication, availability and audience data boundaries Use the same anonymous permissions for both Explanation Correct Answer C:Internal and public channels can have different sign-in and exposure requirements. Incorrect Answers A:Web channels can require separate OAuth and sign-in configuration. B:A single open link could expose employee-only tools externally. D:Internal data should not be exposed through unauthenticated access.
Question 5 A procurement agent is designed to approve purchases over $10,000. What governance design fits this financial risk? Let the agent infer consent from enthusiastic wording Store approval only in a chat transcript Require explicit approval with identity, policy checks and auditable action records Approve any purchase that passes a prompt Explanation Correct Answer C:High-impact purchases need accountable authorization beyond a model recommendation. Incorrect Answers A:Ambiguous language is not proof of delegated purchase authority. B:An unstructured transcript does not enforce approval workflow controls. D:Prompt validation is not a financial authorization control.
Question 6 A hospital FAQ agent answers generic visiting hours, while a clinical agent accesses patient records. How should the solution be partitioned? Expose patient record lookups in the public FAQ Rely on a disclaimer to prevent misuse Separate public information from authenticated clinical operations and enforce distinct permissions Use identical credentials across both agents Explanation Correct Answer C:Different sensitivity levels require different trust and tool boundaries. Incorrect Answers A:The public agent must not retrieve identifiable patient information. B:Disclaimers cannot substitute for access controls. D:Shared elevated credentials erase separation of clinical duties.
Question 7 A manufacturer needs consistent currency-formatting logic across three agents. Which design supports reuse? Encapsulate the formatter as a governed reusable tool with an explicit contract Store the formatter in an unrelated customer profile Implement three unrelated scripts with no versioning Copy an informal formatting sentence into each prompt Explanation Correct Answer A:Centralized reusable logic reduces divergent behavior and eases validation. Incorrect Answers B:Profile data is not a reusable computation interface. C:Duplicated code multiplies maintenance and regression risks. D:Repeated prose can diverge and lacks a controlled implementation.
Question 8 An agent is meant to escalate only after confirming the user's region and support tier. What should be planned? An input collection and routing design using validated variables before tool execution Wait until after executing the escalation tool to ask Grant every user premium support by default Choose the escalation destination randomly Explanation Correct Answer A:Explicit routing conditions keep unverified user claims away from sensitive actions. Incorrect Answers B:The missing inputs must be validated before the external action. C:It bypasses the stated support-tier policy. D:Random routing ignores entitlement and geographic constraints.
Question 9 A multinational agent uses European employee records and US support contractors. What early architecture review is required? Assess cross-border data flows, retention and applicable residency constraints Delay data-flow documentation until production Copy all employee files to a public index Assume encrypted transit removes all legal obligations Explanation Correct Answer A:Regulated transfers may need controls before indexing or tools send data abroad. Incorrect Answers B:Late discovery can require rework and create compliance risk. C:Public indexing creates exposure beyond the intended audience. D:Encryption does not alone establish a lawful transfer basis.
Question 10 A travel agent helps book expensive nonrefundable flights. Which responsible AI pattern reduces harmful actions? Make reservations silently whenever prices look good Treat the model's highest confidence as user consent Prevent users from seeing the final itinerary Separate suggestions from confirmed transactions and require explicit user confirmation Explanation Correct Answer D:Approval before irreversible purchases prevents speculative action from becoming a booking. Incorrect Answers A:Automated purchases conflict with the user's control of commitments. B:Confidence is not an authorization to spend money. C:Hiding details blocks informed confirmation.